IMAGE RECOGNITION OF TYPICAL POTATO DISEASES AND INSECT PESTS USING DEEP LEARNING

As an important food crop, potato is often attacked by pests and diseases. Traditional pest identification relies on the visual observation of agricultural workers for empirical distinction, which has a small detection range, high labor intensity, and low operating efficiency. This paper takes potato pests and diseases images under natural conditions as the research object, and uses image processing and pattern recognition technology to automatically classify pests and diseases. Firstly, in view of the problems of traditional potato pest detection methods, a potato pest detection model based on Faster R-CNN is proposed; Secondly, the residual convolutional network is used to extract image features, Max-pooling is a down-sampling method, the feature pyramid network is introduced into the RPN network to generate object proposals, and the convolutional neural network structure is optimized; Finally, construct a potato pest data set, and implement model training and testing to detect potato pests. The test results based on the TensorFlow framework show that the optimized neural network algorithm has an average recognition accuracy of 97.8% for typical potato pest images. The optimized convolutional neural network recognition model has stronger robustness and applicability, and can provide a reference for the identification and intelligent diagnosis of potato and other crop pests.