Pest Detection and Classification Using YOLO AND CNN
Shashidhar Cheeti
1,a), Dr G.A.E. Satish Kumar
2,b), J. Swetha Priyanka
3,c), Ghazala Firdous
4, d), Pogaku Rani Ranjeeva
5, e)1,2,3,4,5Department of Electronics and Communication Engineering, Vardhaman College of Engineering
(Autonomous), Hyderabad, India-501218
a)[email protected],b) [email protected],c) [email protected],
d)[email protected],e)[email protected] Abstract
The ideology of the paper presents the detection and Classification of pests using YOLO AND CNN. The brisk increase of human population leads to the increase in the demand of the food. Due to the illiteracy rate and deprivation in our country we mislay a large amount of crops due to climatic conditions and pests. Enormous quantities of crops are ruined annually due to the presence of pests. So the pest must be detected and classified in order to guarantee superior production in agricultural fields. Prior detection of pests in the image is important for control of pests in the fields. Due to this, classifying the pest in the images has been an onerous task. The main intention of this paper is to classify the pests and to apply certain measures to protect the crops from the pests. For detection of pest we use YOLO (You look only once) algorithm and for classification of pest we use CNN (Convolution Neural Network).
Keywords
Alexnet CNN; Pest Classification; Pest detection; YOLO Segmentation.
Introduction
Pest is an organism which is harmful to human and human needs. Pests mainly damage the agricultural crops due to which there is a loss to the farmers. In the Modern era due to rapid increase in population there is shortage of food.
Global warming is another reason due to which there is shortage of food. Due to the global warming there is a change in the climatic conditions and rise in sea levels. These factors affect the growth of the agricultural crops.
Pest detection should be done so as to detect the pest present on a crop. We could identify the pests through our eyes but it is a long process to identify multiple pests. There are various methods invented to detect the pests. For the practical implementation of pest detection and classification the embedded systems are used widely for easy implementation of software with hardware. The early pest detection techniques involved the single pest detection in a single image but in the recent developments involved multiple pest detection in a single image due to which all the pests present on the crops are detected.
Pest detection is not sufficient for efficient crop growth. Pests must be classified so as to use proper pesticides on the pests. Permethrin is a pesticide used to kill grasshopper in the crops. The same pesticide cannot kill another pest like ladybug. So pests must be classified to use a proper pesticide on the pests
In this paper we deal with pest detection and classification used YOLO and CNN, where YOLO (You look only once) algorithm is used for detection of pest in an image and Alex net CNN is used for pest classification. 2 networks are being used for pest detection and classification for better accuracy of the pest in an image.
Previous Works
In this section, several existing types of research towards various pest detection and classification are given below
In the work of Apurva Sriwastwa [1] describes about the spotting of pests using image segmentation technique.
The segmentation technique is based on color. This segmentation technique along with the clustering algorithm is used on different types of infected crops.
This work of B. Rajesh, M. Vishnu Sai Vardhan and L. Sujihelen [6] proposed a system which uses a decision tree to recognize and analyze the leaf infection and extension of the spotting accuracy with less amount of time.
In this work of Jia Shijie, Jia Peiyi, Hu Siping, Liu Haibo [7] explains the spotting algorithms on leaf images and establish the CNN model to spot the tomato pests and infections based on VGG16[8] and transfer learning.
In the work of N.Vinushree, B.Hemalatha, Vishnu Kumar Kaliappan [8] presents a clustering technique; the popular KFCM algorithm is used to recognize density of pest in plant. A supervised learning network is applied to separation of feature extraction of leaf in the plants.
In the work of Vivek Agnihotri [9] explains about the detection of pests in crops and identification of pests in fields by using a microprocessor device along with infrared and normal camera would be connected to the quadcopter that will fly over the field region and recognizes the pest.
In the work of Tu-Liang Lin, Hong-Yi Chang, Kai-Hong Chen [10], explains about a Faster R-CNN algorithm for pest and infection recognition in the growing of peppers.
Proposed Design
Figure 1. Block Diagram A. Database
The pest images are being taken manually from the internet. The size of the database can be varied. In this method we use four different types of pests namely Colorado Bettle, Grasshopper, Japanese Bettle, Lady Bug. These pests are present on leaves and flowers. The database consists of pest images which are being captured in various angles. For non real time purposes the database is prepared by manually collecting the images where the pests are present. In the real time conditions the image is captured by image acquisition toolbox in MATLAB. The input image which consists of pest is being pre-processed before the segmentation process.
Figure 2. Input images in the database B. Pre-processing
The pre-processing of the image from the database involves the resizing of the image as per size required by the networks used in the methodology. There are 2 networks used and the input image is being resized without changing the aspect ratio of original image. The aspect ratio states as the ratio of height and width of an image. The resizing operation is the only operation performed in pre-processing.
The noise and uneven illuminations of sunlight are eliminated in the YOLO Segmentation.
C. YOLO Based Segmentation
YOLO (You Look Only Once) is a real time object detection system. There are different object detection techniques present which include Regional Convolution Neural Network and You look only once (YOLO) v2, v3. For this pest detection we use yolov2ObjectDetector. YOLO v3 object detector upgrades upon YOLO version2 object detector by adding detection at different scales to assist detection of min objects. The YOLO v2 is sufficient for pest detection on the crops. The first step in the YOLO segmentation involves the preparing the ground truth. The image Labeler is used to create the bounding box for the pest in an image. We need to create the bounding boxes for all the images present in the database. After the creation of the bounding boxes we will load the layers required by the network. Next step involves setting the training options which involves the use of classifier, min batch size, initial learn rate, Max epoch’s size and checkpoint path. The network is trained for certain iterations.
Figure 3. Training of YOLO Network
need to train the network same as the YOLO network. The first step involves the loading a pre-trained neural network. The layer architecture is reviewed and the next step involves the modifying the pre- trained network. In this the fully connected layer and classification layers are modified along with the change in the maxEpochs and minibatch size. The training options are set as per the required network.
The finishing step of the Alexnet CNN is training. We will train the network and graphical representation of training is being observed.
Figure 5. Training of Alexnet CNN F. Pest Classification
The final step after training involves the identification of pest present on the image. The pest classification gives the name of the pest present in the input image. The single pest or multiple pests are being detected and classified.
Results and Discussions
The output of the pest detection and classification involves the detection of pest by the bounding box around the pest and identification of pest by give label to the pest. The Graphic User interface is used for the representation of output of the pest detection and classification. Initially we will browse the image from the database and the detection button on the GUI is pressed after that both pest detection and classification output is displayed. The pest detection and classification can be observed on single pest in an image or multiple pests in an image. The following figures show the output for single and multiple pests.
Figure 6. Output for Lady Bug detection and identification in the image
Figure 7. Output for multiple Colarodo Bettle detection and identification in the image
Advantages of the proposed method over existing are:
The YOLO Based Segmentation technique provides the accurate pest detection by eliminating the maximum unwanted area from the input image.
The usage of Convolution Neural Network gives more accurate result with less number of errors.
Conclusion
The pest detection and classification is performed for detection of pests at early stages such that the farmers use proper amount of pesticides to kill the pests. The pests are classified so to use the required pesticide on the particular pest. The pest detection and classification can be implemented in hardware and the farmers could use at a wide range so they could save the fields from the pests. The future scope in the pest detection involves the usage of the pest detection and classification method by the farmers in a wide range as of
2012, pp. 1-5, doi: 10.1109/ICCCNT.2012.6395903.
[3]
A. Martin, D. Satish, C. Balachander, T. Hariprasath and G. Krishnamoorthi, "Identification and counting of pests using extended region grow algorithm," 2015 2nd International Conference on Electronics and Communication Systems (ICECS), Coimbatore, India, 2015, pp. 1229-1234, doi:10.1109/ECS.2015.7124779.
[4]
Y. Kumar, A. K. Dubey and A. Jothi, "Pest detection using adaptive thresholding," 2017 International Conference on Computing, Communication and Automation (ICCCA), Greater Noida, India, 2017, pp. 42-46, doi: 10.1109/CCAA.2017.8229828.[5]
P. Rajan, B. Radhakrishnan and L. P. Suresh, "Detection and classification of pests from crop images using Support Vector Machine," 2016 International Conference on Emerging Technological Trends (ICETT), Kollam, India, 2016, pp. 1-6, doi: 10.1109/ICETT.2016.7873750.[6]
B. Rajesh, M. V. Sai Vardhan and L. Sujihelen, "Leaf Disease Detection and Classification by Decision Tree," 2020 4th International Conference on Trends in Electronics and Informatics (ICOEI)(48184), Tirunelveli, India, 2020, pp. 705-708, doi: 10.1109/ICOEI48184.2020.9142988.[7]
J. Shijie, J. Peiyi, H. Siping and s. Haibo, "Automatic detection of tomato diseases and pests based on leaf images," 2017 Chinese Automation Congress (CAC), Jinan, China, 2017, pp. 2537-2510, doi: 10.1109/CAC.2017.8243388.[8]
N. Vinushree, B. Hemalatha and V. K. Kaliappan, "Efficient Kernel-Based Fuzzy C-Means Clustering for Pest Detection and Classification," 2014 World Congress on Computing and Communication Technologies, Trichirappalli, India, 2014, pp. 179-181, doi:10.1109/WCCCT.2014.61.
