Identification Of Leaf Crop Diseases Using Convolutional Neural Networks

The economic impact of crop diseases caused by environmental pollution can be complex and go beyond their immediate effect on directly affected agricultural producers.Some of the possible effects are food insecurity,health risk,environmental degradation,increased financial costs,reduced yield of crop food production,and increased production costs.Automatic detection and classification of crop diseases at an early age could help the producers to take corresponding prevention measures in time and avoids the abuse of pesticides and protects the environment we live on,and could also improve the yield and quality of crop food production.This paper discussed the question of how to identify and classify leaf crop diseases using Convolutional Neural Networks.In order to make the experience a real-life experience,we use 13,243 soybean leaf crop diseases,and 17,664 cucumber leaf crop diseases.The acquired leaf disease images are usually messy,come from different sources and present some visible symptoms and invisible symptoms.In order to feed a dataset of images to a Convolutional Neural Network,they must all be preprocessed.In this paper,we implemented Unsupervised fuzzy clustering,and color space transformation combined with k-means clustering to differently preprocessed the input images.The main works and achievements of this paper are as follows:Firstly,we designed a Convolutional Neural Network based on LeNet to perform leaf crop disease recognition and classification using affected areas of disease spots.The affected areas of disease spots were segmented from the leaves images using the Unsupervised fuzzy clustering algorithm.It achieved a test accuracy of 95.82%with segmented soybean spot diseases,and 96.04%with segmented cucumber spot diseases.The proposed algorithm was evaluated through comparison,and the comparative experiments were conducted using SVM and VGG16.VGG16 model achieved a performance reaching a 97.05%success rate with segmented soybean spot diseases,and 97.34%success rate with segmented cucumber spot diseases and the SVM model achieved a testing accuracy of 88.17%with segmented soybean spot diseases,and 88.32%with segmented cucumber spot diseases.It is found that segmented spot disease does not improve the recognition performance,and the characteristics of the input images such as the shape,the texture,and the margin are influencing factors to the CNN algorithm.In order to improve previous performance recognition results.Secondly,we applied the color space transformation on the inputs images,to covert the color from RGB color space to l*a*b color space,and using k-means clustering to Classify the colors in a*b*space,then label every pixel in the image to make the symptoms more visible without segmented the spot diseases or modify any characteristics of the input images such as the shape,the texture,and the margin.Compared with the previous results,the performance of the CNN model was successfully improved by 2.73%when using preprocessed soybean leaf images,and 2.61%when using preprocessed cucumber leaf images.Thirdly.we propose a new Convolutional Neural Network based on Segnet and VGG16,that we gave the nickname "Encoder-classification network "E-C N".The performance of the proposed algorithm was tested using the same datasets includes soybean healthy leaves and cucumber healthy leaves collected in the real cultivation condition in the field.The proposed CNN model "E-CN" achieved an excellent recognition performance of 99.47%success rate with soybean leaf images and a 99.42%success rate with cucumber leaf images.