Image Classification Of Crop Diseases And Pests Based On Convolutional Neural Network

The plant of diseases and insect pests is one of the main challenges in agricultural production.Because of its variety and complexity,it is very easy to outbreak in many specific environments,resulting in the decline of agricultural production or even crop failure.With the continuous progress and development of society,the total population of the world continues to grow,and agricultural land continues to shrink;how to ensure that the production capacity of crops and market demand to meet the benign has become an increasingly serious issue;and the problem of pests and diseases is the key link in this topic.In this paper,a method of image classification of crop diseases and insect pests based on convolution neural network is proposed.The data set(61 classes,47 637 pictures)provided by the Global AI Challenge,which is the largest in the field of agriculture in China so far,is used for in-depth study.The main research work of this paper is as follows:(1)According to the task of crop pest classification,the classification model of crop pests and diseases based on VGG-16 was used to classify them into 61 categories.Experiments show that the model can effectively identify crop pests and diseases,and the overall recognition accuracy is 76.8%.It has good application prospects.(2)The similarities and differences between traditional image classification task and crop pest classification are analyzed in detail.The data set of pests and diseases used in this paper includes 10 species,27 diseases and 10 health classifications,while the traditional image classification task can only classify pests and diseases into 61 categories,so a multi-branches structure is designed specifically for the data set of pests and diseases.Multi-branches CNN model uses multi-branches structure to classify pest and disease data.Each kind of crop is a separate branch.The label of crop species is used as command signal to select branch.Finally,61 kinds of classification problems are transformed into multi-branches training problems.Experiments show that the model can effectively identify crop pests and diseases,the overall recognition accuracy is 82.1%,compared with the CNN model improved by about 5%,indicating that the Multi-branches CNN model has a certain improvement in the performance of crop pest classification tasks compared with the CNN model,and has good research prospects.(3)There is little difference in the degree of pests and diseases of the same crop,so combining the fine-grained image classification task with the image classification task of crop pests and diseases,this paper proposes a method of combining the mainstream fine-grained image weak supervised classification model Bilinear CNN and Multi-Branches CNN to classify the crop pests and diseases images,and extracts thefeatures of CNN by Bilinear processing.Multi-branches B-CNN model.Experiments show that the model can effectively identify crop diseases and insect pests,and the overall recognition accuracy is 89.2%.Compared with CNN model and Multi-Branches CNN model,the classification performance of the model has been greatly improved.This shows that the multi-Branches B-CNN model can improve the classification performance of crop diseases and insect pests,and has a certain reference significance for the Research of other crop diseases and insect pests classification.Finally,the theoretical methods of crop pest identification discussed in this paper have fundamentally changed and improved the performance of key data compared with the traditional methods,which has a better effect on solving the problems of pests and diseases that may arise in the actual production process.