Classification Of Crop Diseases Based On Convolutional Neural Network

Around the world,crop diseases will seriously affect crop yields,resulting in slower agricultural development in various countries.How to accurately identify different types of diseases and intelligently diagnose diseases is an important research direction of agricultural intelligence.Crop diseases not only affect agricultural production,but also threaten food safety.Due to the complex and variable disease symptoms,even experienced agronomists and plant pathologists often fail to successfully diagnose specific diseases,which leads to wrong conclusions and treatments;and manual identification of crop diseases is time-consuming and labor-intensive.The subjective nature of the pesticide will lead to the wrong use of pesticides,so it is very important to study intelligent identification systems that can quickly and accurately identify crop diseases.In this paper,in order to improve the accuracy of crop disease image recognition,the classification and recognition method of convolutional neural network is studied.The main research contents are as follows:(1)To solve the problem that the convolutional neural network requires many training parameters in tomato disease recognition,and the training is very time-consuming,the classification method of tomato leaf images in 10 common categories(including disease and health)is studied and adopted.A method based on transfer learning to classify tomato diseases.The mature Alexnet model that has been trained on the Imagenet image data set and its parameter migration are used.The tomato disease image is used as the input of the network to retrain the network.The trained model is used to classify 10 types of tomato leaves.The experimental results show that this research method Reduced network training time and improved classification accuracy.(2)Aiming at the problem that the convolutional neural network needs to improve the accuracy and real-time performance of crop disease identification,a multi-scale convolutional neural network is proposed to classify 28 types of diseased and healthy leaves of various crops.The multi-scale network model is a multi-scale model built on the basis of the ZFnet network structure.The low-level network extracts the local features of the image,the high-level network extracts the global features,and the multi-scale features are stitched and input to the fully connected layer.The final layer of the fully connected layer classifier outputs the classification results,and then realizes a variety of crop disease image classification.(3)Due to the large gap in the number of samples corresponding to each category in the crop disease image set,the classifier will be biased towards the majority category.The classification accuracy of categories with a large number of samples is high,and the accuracy of category recognition with a small number of samples is not high.In view of the imbalance of the crop disease data set,this study will use a deep convolutional generation adversarial network to expand the samples of a few categories to balance the number of samples in each category,and then use the convolutional neural network to classify the balanced image set.The realization result proves that the data balancing method used in this study can effectively expand the data samples and improve the classification accuracy of a few categories.