Abstract: The accurate segmentation of wheat leaf disease and pest areas plays an important role in realizing the automation and intelligence of disease detection, and has important theoretical value and practical significance for improving the economic benefits of crops. In order to solve the problems of large workload and poor classification effect in the existing research, this study combines the active learning strategy of manual intervention labeling, and proposes a semi-supervised active learning method for wheat leaf disease and pest region segmentation. By extracting boundary features, color space features and texture features, the feature vector is composed, and the transductive support vector machine is used as the classifier. In each round of training, the k-means algorithm is used to randomly generate the samples to be labeled, and the human-computer interaction system is used to intervene the sample labels to improve the classification efficiency. This paper constructs a LM neural network model for classifying pests and diseases based on a new disease region segmentation method. In the image segmentation experiment, the proposed model is compared with the classical small sample SVM model and the semi-supervised TSVM model. The segmentation effect ranks first, which proves the superiority of the proposed segmentation strategy. In the experiment of pest identification, the model verification results showed that the recognition accuracy of LM neural network classification model using semi-supervised active learning method was 93.75%, which has good classification effect and can provide effective basis for subsequent pest control.

Key words: semi-supervised learning, active learning, TSVM, disease leaf segmentation