Neural network based crop disease identification method

Abstract

The invention provides a neural network based crop disease identification method comprising the following steps: the category of diseased leaf images is estimated via use of a constructed self-adaptive global pooled convolutional neural network; the self-adaptive global pooled convolutional neural network is formed by orderly connecting one input layer, one batch normalization layer, six hidden layers and one classification output layer; each of the first four of the hidden layers includes convolution operation, activation operation, maximum pooling operation, and local response normalizationoperation; a fifth hidden layer includes convolution and activation operation, and a sixth hidden layer is global pooling operation; self-adaptive learning rates adopted in the method disclosed in theinvention can help greatly shorten time required for training and overcome problems of insufficient learning and trapping in local optimum caused by fixed learning rates; convergence speed, generalization ability and stability of a network can be improved.

Description

technical field [0001] The invention relates to the technical field of crop disease identification, in particular to a neural network-based crop disease identification method. Background technique [0002] Crop diseases seriously affect the yield and quality of crops. Disease control is an important link in crop production. To prevent and control diseases, we must first identify the types of diseases. Disease leaf symptoms are the main basis for disease identification. The research on disease recognition methods based on diseased leaf images has always been an important research topic in the fields of computer perspective, image processing and machine learning. However, due to the ever-changing color, shape, and texture of crop leaf disease images, and the background of the actual collected leaf images is relatively complex, traditional crop disease identification methods and technologies cannot meet the needs of the actual crop leaf disease monitoring system based on the...

AI Technical Summary

Problems solved by technology

In order to overcome the low recognition rate of the traditional crop disease recognition method and the slow convergence speed of the existing CNN, which is easy to fall into local optimum, the present invention provides an adaptive global pooling convolutional neural network (Adaptive global pooling CNN, AGPCNN)

Images