Research On Image Recognition Of Crop Diseases And Insect Pests Based On Convolutional Neural Networke

The rapid and accurate classification of crop diseases and insect pests images by computer is helpful for the accurate diagnosis of crop diseases and insect pests and the targeted pesticide spraying in the later period,which has great significance for the development of the agricultural field.In response to this,this paper mainly designs and implements a network model for image recognition to quickly and accurately diagnose crop diseases and insect pests,and at the same time has a strong anti-interference ability to improve,in order to accurately judge in the actual shooting environment,thus Promote the intelligent identification of crop diseases and insect pests and the popularization and promotion of accurate agricultural knowledge,improve the diagnosis and prevention of crop diseases and insect pests,and provide a basis for identification in later intelligent agriculture.The main innovation of this article is to introduce the self-attention network [1] and recognize the third layer in the convolutional neural network,and change the Soft Max in the self-attention framework to Sigmoid to achieve network memory and forgetting(similar to LSTM)[2],Add a self-attention network between the later convolutional layers,because at this time after multiple pooling layers and convolution kernel extraction,the image matrix is reduced,using Soft Max can better extract the matrix globally,these two types of self-attention The combined use of the network can better allow the model to forget the interference area,and pay attention to the nuances of agricultural disease and insect pest images,so as to achieve more accurate processing of images with blurry and small pest areas.It has been experimentally verified that the combination of these two self-attention networks is better than the benefits of using only one of them.On the artificial intelligence challenge data set,the test accuracy rate reaches 90.59%,breaking the 90% limit It is better than the existing method by 0.45%,and the model performs better than other methods after adding interference noise.In summary,based on the research proposed in this paper,it mainly implements a high-accuracy,strong anti-interference ability rapid diagnosis model for crop disease and insect image recognition,which provides technical support for targeted pesticide spraying and prevention in the later stage of the agricultural field Good theoretical research significance and application value.