• Nu S-Au Găsit Rezultate

View of An Automated Detection and Classification of Plant Diseases from the Leaves Using Image processing and Machine Learning Techniques: A State-of-the-Art Review

N/A
N/A
Protected

Academic year: 2022

Share "View of An Automated Detection and Classification of Plant Diseases from the Leaves Using Image processing and Machine Learning Techniques: A State-of-the-Art Review"

Copied!
18
0
0

Text complet

(1)

15933 http://annalsofrscb.ro

An Automated Detection and Classification of Plant Diseases from the Leaves Using Image processing and Machine Learning Techniques: A

State-of-the-Art Review

S. KRITHIKHA SANJU

Department of Computer Science and Engineering, CEG, Anna University, Chennai, India.

DR. B. L. VELAMMAL Associate professor

Department of Computer Science and Engineering, CEG, Anna University, Chennai, India.

Abstract

India is being rapidly developing country, and for years agriculture was the cornerstone on which the country‟s economy grew and developed. The agriculture domain has been experiencing challenges owing to massive industrialization and globalization. Further, it is crucial that the knowledge of cultivation and its significance be disseminated to the millennial generation. Today, while the sweeping winds of change have made their presence felt everywhere, agriculture, sadly, continues to lag behind in the implementation of new technology and is still stuck in time with primitive methodology and outdated practices. A failure to recognize plant disease correctly results in enormous losses in terms of production, energy, resources and product quality. A comprehensive study of the different conditions in which food crops grow plays a pivotal role in effective planting. Monitoring plant disease using automated methodology helps to lower huge tracking jobs on large crop farms, and identifies the symptoms of pathogens at the earliest stages, that is, when they show up on leaves. This article presents a survey on the machine learning and image processing strategies used to identify and classify plant leaf disease automatically, using which an automated system for leaf disease detection and classification can be effectively implemented. Once a disease is rightly identified, fertilizers can be appropriately used to maximize agricultural productivity.

Keywords: Back propagation neural network, leaf disease detection, image processing, k-means clustering, gray level co-occurrence matrix, support vector machine, diseases classification, convolutional neural network Introduction

An image constitutes of a defined number of parts, each part has distinct location and value. Such components are known as pixels or picture element. Digital image processing [27] involves processing images using digital computers. Images are digitized and certain interventions executed to get improved images from which relevant information is obtained. Image processing has applications in disciplines suchas nuclear medicine [28], astronomical observation [29], signature identification [30], number plate detection [31], farming [32, 33] etc.

Digital image processing is used extensively in the agriculture domain.

Plant disease results in economic devastation, as well as environmental and social losses. In order to protect the quantity with quality of crops, accurate identification of plant disease at the earliest stages is imperative. Plant pathology can be tracked in several ways. Many plant diseases have no related adverse side effects, however, and are not easily recognized by farmers. In such circumstances, a comprehensive analysis is normally required, typically via powerful microscopes [34]. In other instances, signals can only be observed by an imaging system that spans nearly the entire electromagnetic spectrum [35], from gamma to radio waves that are not apparent to humans. Farmers use a naked-eye detection system to determine infections from experience and knowledge, but this approach makes it difficult to recognize the diseases effectively. The automated systems [36] are required to easily detect infections using various classification techniques to test the leaves of different plants.

In India, huge number of crops are cultivated that mostly serve as hosts for a range of insect pests [37] and pathogens [38]. The agro-climate produces more insect pests than disease generating pathogens, given that much of India lies in sub-tropical and tropical zones. Early detection and early intervention of plant pathogens are vital to restricting damage to plants and maximizing high-quality production. Hence, farmers need to inspect their crops periodically and identify the very first signs of disease so as to control its spread. Using the naked

(2)

15934 http://annalsofrscb.ro

eye to diagnose disease often results in high error rates and inaccurate classification.

The objective of this research is a survey that addresses problems in plant leaf disease detection and classification by incorporating machine learning and image processing algorithms to track and classify leaf diseases. Given that even professional agronomists and plant pathologists often diagnose plant diseases inaccurately, an automated system that locates, detects and classifies leaf disease effectively is critical.

Pesticides and fertilizers can be used to increase yields in line with the classification results produced.In recent years, organic farming has come into its own, producing high-quality yields with almost no pesticide and fertilizer use. Consequently, leaf disease is to be rightly diagnosed in order to enhance agricultural productivity, which is a key driver in the country‟s economy.

The article is divided into seven sections. Section II starts with a background discussion on disease-causing agents and different types of plant disease. Section III describes each step in the leaf disease detection and classification system. Section IV presents a detailed literature survey in regard to the leaf disease detection and classification of 26 articles. Section V states the comparison among different models. Section VI describes the trends and challenges for leaf disease detection and classification. Section VII summarizes the importance of leaf disease detection and classification in the form of conclusion.

Background

Both abiotic and biotic [39] factors cause plant disease. Biotic factors have their origins from living organisms such as bacteria, viruses, fungi, viroids, algae, nematodes and parasitic plants. These biotic factors are commonly known as pathogens. Abiotic factors include nutritional deficiencies, an excess of nutrients, extreme temperatures, low or extreme sunlight, soil acidity or alkalinity, toxic air pollutants, high moisture, lack of oxygen etc. Much of the research done has focused on plant disease caused by three main pathogens: bacteria, fungi and viruses [40, 41, 42]. This survey article, likewise, is also based on the three pathogens in plant leaves.

Figure 1 depicts the various types of plant diseases caused by bacteria, viruses and fungi.

Fig 1. Types of plant diseases caused by bacteria, viruses and fungi.

(3)

15935 http://annalsofrscb.ro

Fig 2. Generalized architecture diagram of an automated leaf disease detection and classification system.

Leaf Disease Detection And Classification System

Figure 2 shows a generalized architecture of an automated leaf disease detection and classification system. The process of automated leaf disease detection and classification system is divided into phases: image processing and machine learning. The image processing phase incorporates image acquisition, preprocessing, segmentation and feature extraction, while the machine learning phase constitutes classification.

Image Acquisition

Image acquisition [43, 44] is the process of collecting or gathering plant leaf images, with and without disease.

The accuracy of the system chiefly depends on the types of images used, since training is done from these.

Images are captured using a digital camera [45] from the farm, or gathered from the internet. Image quality depends on the type of digital camera used and its orientation. The Canon EOS 1500D, Nikon D5600, Sony Cybershot, Nikon Coolpix P900, and Canon Powershot SX430 are a few widely used digital cameras, specifically employed to capture leaf images. Images accessed from the internet are those from Google and publicly available datasets like the Plant Village dataset, IPM Images, APS Images, and IRRI (International Rice Research Institute) databases, etc.

Image Preprocessing

Image preprocessing [46, 47] follows image acquisition. Image preprocessing refines images through de-noising [48, 49], enhancing [50, 52], resizing [53, 54, 55], data augmentation [56], cropping [57, 58, 59], color space conversion [60], and smoothing [61,62] etc. The leaf images captured may show insects, insect excrement, dewdrops and dust, etc, all of which are considered noise that is to be removed. Distorted images are enhanced, with the distortions removed using noise removal filters. Where image contrast is low, contrast enhancement techniques are called for. Only leaf images are required for the work, and the remaining parts are considered the background. Background removal techniques are therefore used to extract leaves from whole images.

Image Segmentation

In leaf disease detection, image segmentation [51, 63] plays a vital role, because the region of interest is extracted from preprocessed images. Segmentation is the process of dividing robustly-correlated images with respect to the region of interest. Image segmentation is carried out on the basis of similarities or discontinuities.

Segmentation based on similarities divides images with specific predefined criteria, with Otsu„s thresholding [64, 65] being a prime example of this method. Discontinuity-based segmentation considers intensity values in dividing images, and edge detection is an example of this method. Infected leaves show color differences, and such leaf images are segmented using the k-means clustering method to extract diseased portions from the leaves. Numerous studies have used Otsu‟s thresholding and k-means clustering-based segmentation [66, 67], rather than the Prewitt, Sobel and Canny-based segmentation approaches, since the former are ideally suited to leaf disease detection.

Feature Extraction

The feature extraction [68, 69] is the process of identifying and extracting inherent characteristics, known as features that describe disease in the images. Generally, color, texture and shape features are extracted. Color features differentiate one disease from another, based on color, with the histogram and moments being the principal color features. Texture features, which show how image textures are scattered, are extracted for disease classification. Entropy, homogeneity, and contrast are examples of texture features. The shape features extracted reveal how disease symptoms differ in shape. Gray level co-occurence matrix (GLCM) [70, 71] is the most common and effective feature extraction method for leaf disease detection process.

(4)

15936 http://annalsofrscb.ro

Leaf Diseases Classification

The features extracted are used for leaf disease classification [72, 73, 74]. Classification is a supervised learning method of mapping leaf images to different disease classes. By learning from images with disease labels, the classifier method generated will describe the predefined set of disease classes. This learning phase is known as the training step. The trained classifier is used to test the images, and the accuracy obtained is based on the classifier trained. The support vector machine [75], K-nearest neighbour [76, 78] and neural network [77, 79]

are the most commonly used and effective classifiers for leaf disease classification.

State of the Art of Leaf Disease Detection and Classification

Wang et al. [1] analyzed four types of diseases, including grape downy mildew, grape powdery mildew, wheat stripe rust and wheat leaf rust, using image processing and pattern recognition techniques. Images were gathered using a digital camera, and preprocessed using cropping and denoising methods, with no change in image resolution.After preprocessing, the k-means clustering method was used for segmentation and the diseased portion extracted from the image. Color (21), shape (4) and texture features (25) were taken for feature extraction using MATLAB 7.6, and utilized for disease classification using the backpropagation neural network.

Two kinds of work were undertaken: (i) firstly, feature data dimensions were reduced to produce a classification accuracy of 97.14% for grape disease and 100% for wheat disease, and ii) secondly, feature data dimensions were not reduced, and produced 100% classification accuracy for both grape and wheat diseases.

Kutty et al. [2] used the MATLAB neural network pattern recognition with the Statistical Package for the Social Sciences (SPSS) to classify the watermelon leaf diseases of downey mildew and anthracnose. A digital camera was used to collect leaf samples, and samples with regions of interest, that is, infected portions, were identified and cropped. The cropped images were enhanced and segmented using the median filtering method. Finally, the images were analysed using the SPSS software, and the neural network pattern recognition toolbox was used to classify the diseases, producing 75.9% accuracy based on the color features extracted from the RGB color model.

Rothe et al. [3] proposed a pattern recognition method to detect and classify the cotton leaf diseases of Alternaria, Myrothecium and bacterial blight. Images were taken from cotton fields in Buldana, Wardha and the Central Institute of Cotton Research, Nagpur. Image segmentation was carried out using an active contour model, after preprocessing with the Gaussian filter. Invariant moments were used as features to train the neural network. A set of seven invariant moments, known as Hu‟s Moments, were extracted. The backpropagation neural network used to classify the three cotton leaf diseases obtained a classification accuracy of 85.52%.

Barbedo et al. [4] proposed a system for disease detection using color transformations and color histograms, and the pairwise-based classification method was employed. The dataset used in the work consisted of 82 different diseases of 12 plants. The images were preprocessed using the guided active contour method, following which basic image processing was carried out, including leaf segmentation, symptom segmentation and color transformation.For feature extraction and pairwise classification, intensity histogram calculation, histogram cross-correlation, pairwise correlation comparison and disease likelihood estimation techniques were utilized.

The overall classification accuracy was 58%, while individual accuracies ranged from between 40% (corn) and 76% (cotton).

Padol et al. [5] used downy mildew and powdery mildew grape leaf disease images, taken using a digital camera. Preprocessing resized the images and thresholded them to extract all the green components, while the Gaussian filter denoised them. K-means clustering was applied to segment the diseased portions in the leaves, following which the color (3) and texture features (9) were extracted to classify leaf diseases using the linear support vector machine (LSVM). The proposed classification method obtained 88.89% accuracy.

Anand et al. [6] proposed a method to detect brinjal leaf disease using an application-oriented image processing technique. The research classified images of bacterial wilt, tobacco mosaic virus, and Cercospora leaf spot diseases of brinjal leaves. Histogram equalization was used to enhance image quality and k-means clustering to segment diseased leaf parts. Features were extracted from segmented images using the gray-level co-occurrence matrix (GLCM) and the artificial neural network (ANN) for brinjal leaf disease classification. A graphical user interface was also created for the leaf disease classification application.

(5)

15937 http://annalsofrscb.ro

Francis et al. [7] used RGB images of leaves and stems of the pepper crop, captured using a digital camera. The research work classified pepper crop images as healthy or unhealthy, using image processing techniques. The images were converted to the HSV color space and the segmentation done in two phases, masking and thresholding-based. Texture features were extracted using the GLCM, and the feed-forward back propagation neural network classified the pepper crop images as healthy or unhealthy, using the features.

Pinki et al. [8] proposed an automated system to diagnose three common diseases of paddy leaf (leaf blast, bacterial blight and brown spot) and, based on the severity of the disease, the system recommends fertilizers or pesticides accordingly. Images were captured using a camera and collected from the internet. Image resizing, median filtering and contrast enhancement were carried out in the preprocessing stage, and the k-means clustering algorithm used for segmentation. Color (5), texture (5), and shape features were extracted from the segmented images. Using the extracted features, the support vector machine (SVM) classified the diseased paddy leaves, and recommended appropriate dosages of pesticides or fertilizers for the diseased portions. The overall accuracy of the proposed automated system was 92.06%.

A vision-based recognition algorithm for grape leaf disease detection and classification was proposed by Agarwal et al. [9]. Taken from the Plant Village database, the dataset contained images of black rot, esca, and leaf blight, as well as healthy grape plant leaves. The preprocessing phase resized, enhanced and smoothed the images, following which they were converted to both HSI and LAB color models. Both the HIS and LAB color models were used in the research, with the k-means clustering algorithm segmenting both. For feature extraction, the GLCM method and standard MATLAB functions were used for the LAB and HIS colored segmented images, respectively. The multiclass SVM used for the classification of grape leaf disease produced, respectively, an average accuracy of 82.5% and 90% for the LAB and HSI color models.

Healthy leaves, and leaves depicting eight different maize leaf diseases, were detected and classified using an improved deep convolutional neural network proposed by Zhang et al. [10]. Images collected from the Plant Village database and Google websites were preprocessed using dataset augmentation, resizing and labelling.

The improved GoogLeNet and CIFAR-10 pretrained models used for classification utilized fewer parameters than the VGG and AlexNet models. The improved GoogLeNet and CIFAR-10 models produced classification accuracies of 98.9% and 98.8%, respectively.

Sardogan et al. [11] proposed a method for disease detection and classification of tomato leaves using a convolutional neural network with the Learning Vector Quantization (LVQ) algorithm. Images of septoria leaf spot, late blight, bacterial spot and yellow curl diseased tomato leaves, alongside healthy tomato leaves, were taken from the PlantVillage database. The average accuracy was 86% for tomato leaf disease classification.

Diamante Max is a specific breed of tomato plants. Target spot, leaf miner and phoma rot diseases of the Diamante Max breed of tomato plants were detected and classified by a classifier proposed by Luna et al. [12].

A dataset comprising 4923 images of healthy and diseased tomato leaves was put together using a motor- controlled image-capturing box. The fast region convolutional neural network (FRCNN) used to classify the tomato plant leaves achieved 91.67% accuracy.

Khitthuk et al. [13] presented an unsupervised neural network model for grape leaf disease identification and classification. Grape leaf images of scab, rust, and downy mildew, as well as those with no disease, were preprocessed using a Gaussian low-pass filter. Feature extraction was carried out on the diseased leaves using a statistic-based GLCM method and texture feature equations. The unsupervised simplified fuzzy ARTMAP neural network was used as the classifier, with an overall classification accuracy of more than 90%.

Kurale et al. [14] proposed an approach using the decision ree, k-nearest neighbor (KNN), SVM and neural network for plant leaf disease identification and classification. Early blight, black rot, and late blight disease- affected leaves, along with healthy leaves, were captured using a digital camera. Median filtering was used for preprocessing and Otsu thresholding for segmentation. The GLCM method was used to extract texture and shape features with pixel values. A comparison of the performance of all 4 classifiers showed the k-nearest neighbour with the highest classification accuracy of 96.9%.

Bhimte et al. [15] proposed an automatic system for diagnosing cotton leaf disease using image processing techniques with the SVM classifier. Four types of cotton leaf diseases such as bacterial blight, Alternaria leaf spot, gray mildew and magnesium deficiency were considered for the work. Images were captured using a digital camera, and image preprocessing methods like resizing, cropping, filtering, contrast enhancement and color transformation were executed to improve image quality and de-noise the images. The diseased portions

(6)

15938 http://annalsofrscb.ro

were segmented using k-means clustering. The GLCM was used to extract color and texture features from the segmented images and used SVM for cotton leaf disease classification.

Five types of apple leaf diseases, including brown spot, mosaic, Alternaria leaf spot, rust and grey spot, were detected using a deep learning approach based on an improved convolutional network proposed by Wang et al.

[16]. In all, 26,377 images from the apple leaf disease dataset (ALDD) were used for the research, and the image dataset extracted was properly constructed using image annotation and augmentation methods. The deep CNN model proposed combined the GoogLeNet Inception structure with Rainbow concatenation to extract features and classify leaf diseases based on the features. The disease detection performance of the proposed model was 78.80% mAP for the ALDD images.

Anthracnose is a fungal disease that affects the leaves and fruit of the mango tree. Singh et al. [17] proposed a multilayer convolutional neural network (MCNN) to classify Anthracnose disease-infected mango leaf images.

The 1070 mango leaf images were captured in real time and 1130 taken from the PlantVillage database. The histogram equalization method and the central square crop method were used in the image preprocessing phase.

Features were extracted, and mango leaf disease classified using the MCNN model, with an overall classification accuracy of 97.13%.

Too et al. [18] provided a comparative study on fine-tuning deep learning models for classifying plant leaf disease images from the PlantVillage dataset. Deep learning models such as Inception V4, VGG 16, DenseNet with 121 layers, and ResNet with 50, 101 and 152 layers were used for plant leaf disease classification. Of all the deep learning models mentioned above, the DenseNet was found to have the best classification accuracy of 99.75%.

Pantazi et al. [19] proposed an automated leaf disease detection method using image features analysed with a one-class classifier, where the training was done using vine leaves, and tested for a variety of crops. Images of healthy leaves, as well as leaves affected by powdery mildew, downy mildew and black rot disease were taken and the region of interest obtained using the hue saturation value (HSV) with the GrabCut algorithm. Features were extracted using local binary patterns (LBPs) and the one-class SVM was employed to classify diseased leaves, differentiating them from healthy leaves. Totally, 46 plant-condition combinations were tested and 95%

of classification accuracy was obtained.

A new computer vision-based neutrosophic method for basil leaf disease detection and classification was proposed by Dhingra [20]. The leaf image dataset comprised 4 types of healthy and diseased leaf images collected from Punjabi University, National Institute of Pharmaceutical Education and Research (NIPER), and the herb garden of Punjabi Agriculture University, Ludhiana. A neutrosophic logic-based segmentation method was used, along with a novel fuzzy set extended version. The segmented images had one of three membership elements (false, true or intermediate region) for each segmented region in the image. The feature subset from the segmented regions was evaluated using the histogram, color, texture and disease sequences to distinguish between healthy and diseased basil leaves. Nine classifiers were used to evaluate the feature set extracted and the random forest classifier obtained a high classification accuracy of 98.4%.

Karadag et al. [21] proposed an approach to discriminate between healthy pepper leaves and those affected by fusarium, using spectral reflectance from the spectrora diometer that absorbs reflections from leaves. The process of detecting pepper leaf disease took place in two phases: in the first, feature vectors were extracted using the wavelet transform and in the second, the vectors were classified. The classification was performed using the Naïve Bayes, ANN, and KNN classifiers, with the KNN classifier producing the highest classification accuracy of 100%.

Coulibaly et al. [22] introduced a method to classify mildew in pearl millet using transfer learning with feature extraction. Images gathered from the internet and captured using digital devices were preprocessed using augmentation and transformation techniques. The research work used the VGG16 model to extract features and classify millet crop disease, producing 95% classification accuracy.

Barbedo [23] proposed a deep learning method to identify plant leaf disease from individual lesions and spots.

Images were captured using different sensors, with the system classifying disease in terms of five major symptoms. Data augmentation was performed and the pretrained GoogLeNet CNN model used in MATLAB to extract features and classifying diseases in line with the symptoms. It was found that a leaf can have multiple diseases, and the method obtained around 88.5% accuracy for the classification of leaf diseases, using both

balanced and imbalanced datasets.

(7)

15939 http://annalsofrscb.ro

Elhassouny et al. [24] used 7176 images of tomato leaves to detect and distinguish between diseased and healthy leaves. The work proposed a CNN model to classify 10 different types of tomato leaf disease using a smart mobile application. The proposed CNN model, inspired by the MobileNet CNN model, was used for feature extraction and diseased leaf classification. The proposed method used different optimizers and learning rates to classify tomato leaf disease, obtaining a high 90.3% accuracy with a learning rate of 0.001.

Kamal et al. [25] used six different model architectures of CNN pretrained models to classify 28 leaf disease classes of 15 different plants using the fine-tuning process that applies the concept of freezing blocks of layers.

Images from the PlantVillage dataset were used for the research. Of the six model CNN architectures used, the DenseNet169 produced a high plant leaf disease classification accuracy of 99.74%.

Bashir et al. [26] implemented the SVM for detecting and classifying three types of rice diseases: false smut, brown spot and bacterial blight. Images were gathered from the Rice Knowledge Bank, American Phytopathological Society, ShutterStock and Rice Research Institute. Using EM GUI functions, the images were converted to grayscale images in the preprocessing step. Feature detection was done using the SIFT algorithm and the features clustered using the k-means clustering algorithm. Finally, the SVM classifier was used to classify rice diseases by training the normalized histogram. The overall disease classification accuracy was 94.16%.

Comparison Among Different Models

Figure 3 represents the accuracy obtained by each articles in the literature survey. Figure 3 shows that research article [21] achieved a high accuracy of 100% using the KNN for pepper leaf disease, while research article [4]

achieved a low accuracy of 58% using the disease likelihood estimation method for 82 different diseases of 12 plants. Table 1 presents a summary of the articles constituting the literature survey, using image processing and machine learning techniques. Table 1 names the methods used for all the processes in every paper.

Fig 3. Literature survey articles with their classification methods Vsaccuracyobtained from those classification methods in percentage.

Table 1 A summary of the literature survey

Reference Dataset used

Image preprocessing

techniques

color space

Segmentation method

Feature extraction method

Classification method

(8)

15940 http://annalsofrscb.ro

[1] Grape leaf images captured using a digital camera

Cropping and denoising

RGB k-means

clustering

MATLAB 7.6 to extract color, shape and texture features

Backpropagation neural network

[2] Watermelon leaf images captured using a digital camera

Cropping method to crop the region of interest in the images.

RGB Median

filtering

SPSS software Neural Network Pattern Recognition toolbox in MATLAB

[3] Cotton leaf images taken from cotton fields

Gaussian filtering

RGB Active contour model

The function cent moment in MATLAB to extract Hu‟s Moments

Backpropagation neural network

[4] 82 diseases of 12 plant leaves collectedusing a digital camera and mobile devices

Guided Active Contour

HSV, LAB, CMYK

Leaf segmentation and symptom segmentation

Intensity histogram calculation, histogram cross correlation

Pairwise based classification ( pairwise correlation comparison, disease likelihood estimation ) [5] Images of grape

leaves captured using a digital camera

Resizing, thresholding, gaussian filtering

RGB K-means

clustering

3 color and 9 texture features extracted using mathematical equations

LSVM

[6] Images of brinjal leaves

Histogram equalization

RGB Means

clustering

GLCM ANN

[7] Images of pepper plant leaves and stems captured using a digital camera

Color transformation

HSV Masking and thresholding

GLCM Feed forward

backpropagation neural network

[8] Images of paddy leaves collected from the internet and by using a digital camera

Resizing, median filtering, contrast enhancement

RGB K-means

clustering

5 color features, 5 texture features and shape features extracted using mathematical equations

SVM

[9] Images of grape leaves takenfrom the PlantVillage dataset

Resizing, image enhancemen, smoothing

HSI, LAB K-means clustering

GLCM and MATLAB functions

Multiclass SVM

[10] Images of maize leaves taken from the PlantVillage dataset

Data augmentation, resizing, labelling

RGB - Improved

GoogLeNet and Cifar10 models

Improved GoogLeNet and Cifar10 models

[11] Images of tomato leaves from thePlantVillage dataset

- RGB - CNN CNN with LVQ

(9)

15941 http://annalsofrscb.ro

[12] Images of tomato leaves captured using a motor controlled image capturing box

- RGB - FRCNN FRCNN

[13] Images of grape leaves

Gaussian low pass filtering

RGB - GLCM and texture

feature equations

Unsupervised simplified fuzzy ARTMAP neural network [14] Leaf images

captured using a digital camera

Median filtering RGB Otsu thresholding method

GLCM Decision tree, KNN,

SVM and neural network [15] Images of cotton

leaves

Resizing, image cropping, filtering, contrast enhancement, color transformation

LAB K-means

clustering

GLCM SVM

[16] Images of apple leaves takenfrom the ALDD

Image annotation and

augmentation

RGB - Rainbow

concatenation

GoogLeNet inception deep CNN model

[17] Images of mango leaves taken from the PlantVillage dataset and captured using a digital camera

Histogram equalization and central square crop method

RGB - MCNN MCNN

[18] Leaf images from the PlantVillage dataset

- RGB - Inception V4,

VGG16, DenseNet 121 and ResNet 50, 100, 152

Inception V4, VGG16, DenseNet 121 and ResNet 50, 100, 152

[19] Images of vine leaves

- RGB GrabCut

algorithm

LBP method One class SVM

[20] Images of basil leaves

- RGB Neutrosophic

logic based segmentation

Histogram, color, texture and disease sequence using the feature subset method

Decision tree, Naive Bayes, KNN, SVM, Random forest, AdaBoost, ANN, Discriminant analysis, GLM [21] Reflections from

pepper leaves

- RGB - Wavelet transform

method

Naive Bayes, ANN, KNN

[22] Images of pearl millet crop captured using digital devices and gathered from the internet

- RGB - VGG16 VGG16

[23] Images of plant leaves captured using different sensors

Data augmentation

RGB - GoogLeNet CNN GoogLeNet CNN

(10)

15942 http://annalsofrscb.ro

[24] Images of tomato leaves

- RGB - MobileNetCNN MobileNet CNN

[25] Leaf images from the PlantVillage dataset

- RGB - VGG19, ResNet

50, MobileNet, NasNetMobile, Inception V3, DenseNet

VGG19, ResNet 50, MobileNet, NasNetMobile, Inception V3, DenseNet [26] Images of rice Color

transformation

Grayscale - SIFT algorithm, k- means clustering

SVM

Figure 4 and Figure 5 describe the number of research articles available each year, from 2000 to 2020, for the queries, “leaf disease detection and classification” and “leaf disease identification and classification”, on Google Scholar on January 19th 2021. Figure 6 and Figure 7 describe the number of research articles available each year from 2010 to 2020 for the queries, “leaf disease detection and classification” and “leaf disease identification and classification”, in IEEE Xplore on January19th 2021.

Fig 4. Number of research articles available each year from 2000 to 2020 for the query, “leaf disease detection and classification”, on Google Scholar on January 19th 2021.

Fig 5. Number of research articles available each year from 2000 to 2020 for the query, “leaf disease identification and classification”, on Google Scholar on January 19th 2021.

(11)

15943 http://annalsofrscb.ro

Fig 6. Number of research articles available each year from 2010 to 2020 for the query, “leaf disease detection and classification”, in IEEE Xplore on January 19th 2021.

Fig 7. Number of research articles available each year from 2010 to 2020 for the query, “leaf disease identification and classification”, in IEEE Xplore on January 19th 2021.

Trends and Challenges

Research in agriculture has been increasing, and the objective of all the research has been to contribute innovative models that increase agricultural productivity while offering the best quality. A number of studies based on automated leaf disease detection and classification systems have been successfully implemented and achieved high accuracies. However, these have only been done for a few images, or for leaf diseases of a particular plant. In the image processing research domain, creating an automated system for leaf disease detection and classification in terms of big data perspective have not been successfully created. Also, all diseases are not considered for the leaf diseases of a particular plant. The biggest challenge in agriculture today

(12)

15944 http://annalsofrscb.ro

is that of leaf disease detection and classification, particularly in terms of creating an automated model with respect to big data. Such a model must detect and classify leaf disease with high accuracy, using image processing with machine learning techniques.

Conclusion

Leaf disease in plants causes huge losses in agriculture. Consequently, an automated system is called for that detects and classifies leaf disease in plants, using image processing with machine learning techniques. The image processing techniques has been used for detecting the presence of leaf diseases and the machine learning techniques has been used to classify the various leaf diseases. Hence, this survey paper has reviewed and summarized the major techniques used in image processing with machine learning techniques for leaf image preprocessing, segmentation, feature extraction and classification. It has described, in detail, the step-wise process involved in leaf disease identification and classification in automated systems. The machine learning methods used to classify diseased leaf images assist farmers greatly by creating an automated system that easily detects and classifies leaf diseases. Farmers can thus maximize their agricultural productivity through the use of appropriate doses of fertilizers and pesticides to tackle leaf disease.

REFERENCES

[1] Wang, H., Li, G., Ma, Z. and Li, X., 2012, October. Image recognition of plant diseases based on backpropagation networks. In 2012 5th International Congress on Image and Signal Processing (pp. 894-900).

IEEE.

[2] Kutty, S.B., Abdullah, N.E., Hashim, H., Kusim, A.S., Yaakub, T.N.T., Yunus, P.N.A.M. and Rahman, M.F.A., 2013, April. Classification of watermelon leaf diseases using neural network analysis. In 2013 IEEE Business Engineering and Industrial Applications Colloquium (BEIAC) (pp. 459-464). IEEE.

[3] Rothe, P.R. and Kshirsagar, R.V., 2015, January. Cotton leaf disease identification using pattern recognition techniques. In 2015 International Conference on Pervasive Computing (ICPC) (pp. 1-6). IEEE.

[4] Barbedo, J.G.A., Koenigkan, L.V. and Santos, T.T., 2016. Identifying multiple plant diseases using digital image processing. Biosystems Engineering, 147, pp.104-116.

[5] Padol, P.B. and Yadav, A.A., 2016, June. SVM classifier-based grape leaf disease detection. In 2016 Conference on Advances in Signal Processing (CASP) (pp. 175-179). IEEE.

[6] Anand, R., Veni, S. and Aravinth, J., 2016, April. An application of image processing techniques for detection of diseases on brinjal leaves using k-means clustering method. In 2016 International Conference on Recent Trends in Information Technology (ICRTIT) (pp. 1-6). IEEE.

[7] Francis, J. and Anoop, B.K., 2016, March. Identification of leaf diseases in pepper plants using soft computing techniques. In 2016 Conference on Emerging Devices and Smart Systems (ICEDSS) (pp. 168-173).

IEEE.

[8] Pinki, F.T., Khatun, N. and Islam, S.M., 2017, December. Content-based paddy leaf disease recognition and remedy prediction using support vector machines. In 2017 20th International Conference of Computer and

(13)

15945 http://annalsofrscb.ro

Information Technology (ICCIT) (pp. 1-5). IEEE.

[9] Agrawal, N., Singhai, J. and Agarwal, D.K., 2017, October. Grape leaf disease detection and classification using a multi-class support vector machine. In 2017 International Conference on Recent Innovations in SignalPprocessing and Embedded Systems (RISE) (pp. 238-244). IEEE.

[10] Zhang, X., Qiao, Y., Meng, F., Fan, C. and Zhang, M., 2018. Identification of maize leaf diseases using improved deep convolutional neural networks. IEEE Access, 6, pp.30370-30377.

[11] Sardogan, M., Tuncer, A. and Ozen, Y., 2018, September. Plant leaf disease detection and classification based on the CNN with the LVQ algorithm. In 2018 3rd International Conference on Computer Science and Engineering (UBMK) (pp. 382-385). IEEE.

[12] de Luna, R.G., Dadios, E.P. and Bandala, A.A., 2018, October. Automated Image Capturing System for Deep Learning-based Tomato Plant Leaf Disease Detection and Recognition. In TENCON 2018-2018 IEEE Region 10 Conference (pp. 1414-1419). IEEE.

[13] Khitthuk, C., Srikaew, A., Attakitmongcol, K. and Kumsawat, P., 2018, March. Plant Leaf Disease Diagnosis from Color Imagery Using the Co-Occurrence Matrix and Artificial Intelligence System. In 2018 International Electrical Engineering Congress (iEECON) (pp. 1-4). IEEE.

[14] Kurale, N.G. and Vaidya, M.V., 2018, July. Classification of Leaf Disease Using the Texture Feature and Neural Network Classifier. In 2018 International Conference on Inventive Research in Computing Applications (ICIRCA) (pp. 1-6). IEEE.

[15] Bhimte, N.R. and Thool, V.R., 2018, June. Disease Detection of Cotton Leaf Spot Using Image Processing and the SVM Classifier. In 2018 Second International Conference on Intelligent Computing and Control Systems (ICICCS) (pp. 340-344). IEEE.

[16] Jiang, P., Chen, Y., Liu, B., He, D. and Liang, C., 2019. Real-time detection of apple leaf diseases using the deep learning approach based on improved convolutional neural networks. IEEE Access, 7, pp.59069-59080.

[17] Singh, U.P., Chouhan, S.S., Jain, S. and Jain, S., 2019. Multilayer convolution neural network for the classification of mango leaves infected by anthracnose disease. IEEE Access, 7, pp.43721-43729.

[18] Too, E.C., Yujian, L., Njuki, S. and Yingchun, L., 2019. A comparative study of fine-tuning deep learning models for plant disease identification. Computers and Electronics in Agriculture, 161, pp.272-279.

[19] Pantazi, X.E., Moshou, D. and Tamouridou, A.A., 2019. Automated leaf disease detection in different crop species through image features analysis and One-Class Classifiers. Computers and Electronics in Agriculture, 156, pp.96-104.

[20] Dhingra, G., Kumar, V. and Joshi, H.D., 2019. A novel computer vision-based neutrosophic approach for leaf disease identification and classification. Measurement, 135, pp.782-794.

(14)

15946 http://annalsofrscb.ro

[21] Karadag, K., Tenekeci, M.E., Taşaltın, R. and Bilgili, A., 2019. Detection of pepper fusarium disease using machine learning algorithms based on spectral reflectance. Sustainable Computing: Informatics and Systems.

[22] Coulibaly, S., Kamsu-Foguem, B., Kamissoko, D. and Traore, D., 2019. Deep neural networks with transfer learning in millet crop images. Computers in Industry, 108, pp.115-120.

[23] Barbedo, J.G.A., 2019. Plant disease identification from individual lesions and spots using deep learning. Biosystems Engineering, 180, pp.96-107.

[24] Elhassouny, A. and Smarandache, F., 2019, July. Smart mobile application to recognize tomato leaf diseases using Convolutional Neural Networks. In 2019 International Conference of Computer Science and Renewable Energies (ICCSRE) (pp. 1-4). IEEE.

[25] Kamal, K.C., Yin, Z., Li, B., Ma, B. and Wu, M., 2019, September. Transfer Learning for Fine-Grained Crop Disease Classification Based on Leaf Images. In 2019 10th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS) (pp. 1-5). IEEE.

[26] Bashir, K., Rehman, M. and Bari, M., 2019. Detection and classification of rice diseases: An automated approach using textural features. Mehran University Research Journal of Engineering and Technology, 38(1), pp.239-250.

[27] Paul, A., Ghosh, S., Das, A.K., Goswami, S., Choudhury, S.D. and Sen, S., 2020. A Review on Agricultural Advancement Based on Computer Vision and Machine Learning. In Emerging Technology in Modelling and Graphics (pp. 567-581). Springer, Singapore.

[28] Ibrahim, A., Vallieres, M., Woodruff, H., Primakov, S., Beheshti, M., Keek, S., Sanduleanu, S., Walsh, S., Morin, O., Lambin, P. and Hustinx, R., 2019, June. Radiomics analysis for clinical decision support in nuclear medicine. In Seminars in Nuclear Medicine. WB Saunders.

[29] Azhar, M.A.D.M., Hamid, N.S.A., Kamil, W.M.A.W.M. and Mohamad, N.S., 2019, July. Urban Night Sky Conditions Determination Method Based on Low-Resolution All-Sky Images. In 2019 6th International Conference on Space Science and Communication (IconSpace) (pp. 158-162). IEEE.

[30] Al-Azzeh, J., Zahran, B., Alqadi, Z., Ayyoub, B. and Mesleh, M., 2019. Creating Color Image Signature Based on the Laplacian Equation. JOIV: International Journal on Informatics Visualization, 3(2), pp.94-99.

[31] Satapathy, S.K., Mishra, S., Sundeep, R.S., Teja, U.S.R., Mallick, P.K., Shruti, M. and Shravya, K., 2019.

Deep learning-based image recognition for vehicle number information. International Journal of Innovative Technology and Exploring Engineering, 8, pp.52-55.

[32] Xin, J., Mingyong, L., Kaixuan, Z., Jiangtao, J., Hao, M. and Zhaomei, Q., 2019. Development of vegetable intelligent farming device based on a mobile APP. Cluster Computing, 22(4), pp.8847-8857.

[33] Zainuddin, Z., Manjang, S. and Wijaya, A.S., 2019, April. Rice Farming Age Detection Uses Drones Based

(15)

15947 http://annalsofrscb.ro

on SVM Histogram Image Classification. In Journal of Physics: Conference Series (Vol. 1198, No. 9, p.

092001). IOP Publishing.

[34] Ortega, S., Guerra, R., Diaz, M., Fabelo, H., López, S., Callico, G.M. and Sarmiento, R., 2019.

Hyperspectral Push-Broom Microscope Development and Characterization. IEEE Access, 7, pp.122473- 122491.

[35] Horstrand, P., Guerra, R., DÍaz, M., Morales, A., Jiménez, A., López, S. and López, J.F., 2019, November.

A spectral imaging system for precision agriculture: From its inception till a pre-commercial prototype. In 2019 XXXIV Conference on Design of Circuits and Integrated Systems (DCIS) (pp. 1-6). IEEE.

[36] Karthik, R., Hariharan, M., Anand, S., Mathikshara, P., Johnson, A. and Menaka, R., 2020. Attention- embedded residual CNN for disease detection in tomato leaves. Applied Soft Computing, 86, p.105933.

[37] Jactel, H., Koricheva, J. and Castagneyrol, B., 2019. Responses of forest insect pests to climate change: Not so simple. Current Opinion in Insect Science.

[38] Eschenbrenner, C.J., Feurtey, A. and Stukenbrock, E.H., 2020. Population Genomics of Fungal Plant Pathogens and the Analyses of Rapidly Evolving Genome Compartments. In Statistical Population Genomics (pp. 337-355). Humana, New York, NY.

[39] Meena, K.K., Shinde, A.L., Sorty, A.M., Bitla, U.M., Meena, H. and Singh, N.P., 2019. Microbe-Mediated Biotic and Abiotic Stress Tolerance in Crop Plants. In Microbial Interventions in Agriculture and Environment (pp. 315-329). Springer, Singapore.

[40] Raghavendra, B.K., 2019, March. Disease Detection of Various Plant Leaves Using Image Processing Techniques: A Review. In 2019 5th International Conference on Advanced Computing & Communication Systems (ICACCS) (pp. 313-316). IEEE.

[41] Shruthi, U., Nagaveni, V. and Raghavendra, B.K., 2019, March. A review on machine learning classification techniques for plant disease detection. In 2019 5th International Conference on Advanced Computing & Communication Systems (ICACCS) (pp. 281-284). IEEE.

[42] Annabel, L.S.P., Annapoorani, T. and Deepalakshmi, P., 2019, April. Machine Learning for Plant Leaf Disease Detection and Classification – A Review. In 2019 International Conference on Communication and Signal Processing (ICCSP) (pp. 0538-0542). IEEE.

[43] Marcelpoil, R.R., Orny, C. and Morel, D., BD KIESTRA BV, 2019. System and method for image acquisition using supervised high-quality imaging. U.S. Patent 10,200,625.

[44] Supriyanti, R., Rifai, A., Ramadhani, Y. and Siswandari, W., 2019, October. Influence of camera types in histogram distribution on morphological identification of myeloblast cell-based image segmentation. In Journal of Physics: Conference Series (Vol. 1321, No. 3, p. 032017). IOP Publishing.

(16)

15948 http://annalsofrscb.ro

[45] Richardson, A.D., 2019. Tracking seasonal rhythms of plants in diverse ecosystems with digital camera imagery. New Phytologist, 222(4), pp.1742-1750.

[46] Agnihotri, V., 2019, June. Machine Learning-based Pest Identification in Paddy Plants. In 2019 3rd International Conference on Electronics, Communication and Aerospace Technology (ICECA) (pp. 246-250).

IEEE.

[47] Ke, H., Yue, H. and Qiao, J., 2019. Study on Image Preprocessing of Soybean Moisture Content Measurement.

[48] Sadhukhan, T., Chatterjee, S., Das, R.K., Ghosh, S., Das, D. and Sanyal, J., 2019, October. Efficient Removal of Noise from an Image using HSV Filtering. In 2019 Global Conference for Advancements in Technology (GCAT) (pp. 1-4). IEEE.

[49] Lin, P.H., Chen, B.H., Cheng, F.C. and Huang, S.C., 2016. A morphological mean filter for impulse noise removal. Journal of Display Technology, 12(4), pp.344-350.

[50] Ashok Kumar, K., Chowdary, D.D. and Sree, C.D., 2019, October. Data Analysis and Prediction on Cloud Computing for Enhancing Productivity In Agriculture. In IOP Conference Series: Materials Science and Engineering (Vol. 590, No. 1, p. 012014). IOP Publishing.

[51] Ghazal, M., Mahmoud, A., Shalaby, A. and El-Baz, A., 2019. Automated framework for accurate segmentation of leaf images for plant health assessment. Environmental Monitoring and Assessment, 191(8), p.491.

[52] Mittal, N., Raj, A. and Goyal, H., 2019, June. Enhancement and Removal of Noise from Medical Images by the Wavelet Transform Method. In 2019 3rd International Conference on Electronics, Communication and Aerospace Technology (ICECA) (pp. 1126-1130). IEEE.

[53] Liu, C.Y., Chang, C.C., Way, D.L. and Tai, W.K., 2019. Image resizing using fuzzy inferences. IET Image Processing, 13(8), pp.1328-1336.

[54] Mukhopadhyay, J., 2017, July. Image resizing in the compressed domain. In 2017 International Symposium on Signals, Circuits and Systems (ISSCS) (pp. 1-4). IEEE.

[55] Arar, M., Danon, D., Cohen-Or, D. and Shamir, A., 2019. Image Resizing by Reconstruction from Deep Features. arXiv preprint arXiv:1904.08475.

[56] Frid-Adar, M., Diamant, I., Klang, E., Amitai, M., Goldberger, J. and Greenspan, H., 2018. GAN-based synthetic medical image augmentation for increased CNN performance in liver lesion classification. Neurocomputing, 321, pp.321-331.

[57] Lu, W., Xing, X., Cai, B. and Xu, X., 2019. Listwise View Ranking for Image Cropping. IEEE Access, 7, pp.91904-91911.

(17)

15949 http://annalsofrscb.ro

[58] Chen, J., Bai, G., Liang, S. and Li, Z., 2016. Automatic image cropping: A computational complexity study. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 507-515).

[59] Zeng, H., Li, L., Cao, Z. and Zhang, L., 2019. Reliable and efficient image cropping: A grid anchor- based approach. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 5949- 5957).

[60] Takeuchi, M., Saika, S., Sakamoto, Y., Matsuo, Y. and Katto, J., 2018, January. A study on the color-space conversion method considering color information restoration. In 2018 IEEE International Conference on Consumer Electronics (ICCE) (pp. 1-2). IEEE.

[61] Zhu, F., Liang, Z., Jia, X., Zhang, L. and Yu, Y., 2019. A benchmark for edge-preserving image smoothing. IEEE Transactions on Image Processing, 28(7), pp.3556-3570.

[62] Fan, Q., Wipf, D.P., Hua, G. and Chen, B., 2017. Revisiting deep image smoothing and intrinsic image decomposition. arXiv preprint arXiv:1701.02965, 4.

[63] Divya, D. and Babu, T.G., 2019, March. A Survey on Image Segmentation Techniques. In International Conference on Emerging Current Trends in Computing and Expert Technology (pp. 1107-1114). Springer, Cham.

[64] Feng, Y., Zhao, H., Li, X., Zhang, X. and Li, H., 2017. A multi-scale 3D Otsu thresholding algorithm for medical image segmentation. Digital Signal Processing, 60, pp.186-199.

[65] Sha, C., Hou, J. and Cui, H., 2016. A robust 2D Otsu‟s thresholding method in image segmentation. Journal of Visual Communication and Image Representation, 41, pp.339-351.

[66] Baghel, J. and Jain, P., 2016. K-means segmentation method for automatic leaf disease detection. Int. J.

Eng. Res. Appl, 6(3), pp.83-86.

[67] Lin, H., Li, S., Zhao, J. and Qiu, G., 2019, October. Images based on histogram and K-Means clustering segmentation algorithm. In 2019 IEEE 3rd Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC) (pp. 892-895). IEEE.

[68] Nixon, M. and Aguado, A., 2019. Feature extraction and image processing for computer vision. Academic Press.

[69] Tzanakou, E.M. ed., 2017. Supervised and unsupervised pattern recognition: Feature extraction and computational intelligence. CRC press.

[70] Olaniyi, E.O., Adekunle, A.A., Odekuoye, T. and Khashman, A., 2017. Automatic system for grading bananas using GLCM texture feature extraction and neural network arbitrations. Journal of Food Process Engineering, 40(6), p.e12575.

[71] Indriani, O.R., Kusuma, E.J., Sari, C.A. and Rachmawanto, E.H., 2017, November. Tomato classification

(18)

15950 http://annalsofrscb.ro

using K-NN based on GLCM and HSV color space. In 2017 international conference on innovative and creative information technology (ICITech) (pp. 1-6). IEEE.

[72] Kaur, S., Pandey, S. and Goel, S., 2018. Semi-automatic leaf disease detection and classification system for soybean culture. IET Image Processing, 12(6), pp.1038-1048.

[73] Bhattacharya, S., Mukherjee, A. and Phadikar, S., 2020. A Deep Learning Approach for the Classification of Rice Leaf Diseases. In Intelligence Enabled Research (pp. 61-69). Springer, Singapore.

[74] Singh, H., Rani, R. and Mahajan, S., 2020. Detection and Classification of Citrus Leaf Disease Using Hybrid Features. In Soft Computing: Theories and Applications (pp. 737-745). Springer, Singapore.

[75] Reddy, V.H., Kumari, S., Muralidharan, V., Gigoo, K. and Thakare, B.S., 2020. A Literature Survey Food Recognition and Calorie Measurement Using Image Processing and Machine Learning Techniques. In ICCCE 2019 (pp. 23-37). Springer, Singapore.

[76] Sivaraj, S., Malmathanraj, R. and Palanisamy, P., 2020. Detecting anomalous growth of skin lesions using the threshold-based segmentation algorithm and fuzzy K-Nearest Neighbor classifier.

[77] Madgi, M., Danti, A. and Anami, B., 2019, March. Recognition of Green Colour Vegetables' Images Using an Artificial Neural Network. In 2019 International Conference on Data Science and Communication (IconDSC) (pp. 1-5). IEEE.

[78] Larijani, M.R., Asli‐Ardeh, E.A., Kozegar, E. and Loni, R., 2019. Evaluation of image processing techniques in identifying rice blast disease in field conditions based on KNN algorithm improvement by K‐means. Food Science & Nutrition, 7(12), pp.3922-3930.

[79] Sudhesh, R., Nagalakshmi, V. and Amirthasaravanan, A., 2019, March. A Systematic Study on Disease Recognition, Categorization, and Quantification in Agricultural Plants using Image Processing. In 2019 IEEE International Conference on System, Computation, Automation and Networking (ICSCAN) (pp. 1-5). IEEE.

Referințe

DOCUMENTE SIMILARE

Vijayalakshmi M M, Melanoma Skin Cancer Detection using Image Processing and Machine Learning, International Journal of Trend in Scientific Research and

Feature extraction mainly refers to the geometric measurements of the leaf image obtained. The three features such as colour, texture and shape-based features are

We then go on to examine a number of prototype techniques proposed for engineering agent systems, including methodologies for agent-oriented analysis and design, formal

I use a ConcurrentHashMap to store for every thread a HashMap of all the methods it calls and the statistics associated with

had suggested machine learning based automatic segmentation and hybrid feature analysis for Diabetic Retinopathy classification using fundus Image (2020).. This study used

The algorithm is split into following parts: Capture image, Pre-processing, Plate region extraction, Segmentation of character within the extracted number

and Kulkarni, S.B., A Hybrid Approach for Plant Leaf Disease Detection and Classification Using Digital Image Processing Methods. and Abinaya, S., Plant Leaf

In the wake of preprocessing, helpful picture highlights are extricated utilizing highlight extraction procedure that will be utilized as preparing tests for the help