Abstract: In order to improve the accuracy of identifying small target pests and diseases of trees, a small target pest and disease recognition method based on the improved NanoDet-Plus network was proposed. To enhance the ability to extract small target features, the backbone network of the NanoDet Plus network was improved through attention mechanism, and the improved NanoDet-Plus network was used to identify pine wood nematode disease in pine epidemic areas. The results showed that at an IOU of 0.55, the average accuracy of the improved NanoDet-Plus network recognition reached 94.51%;when the IOU was 0.80, the average accuracy of the improved NanoDet-Plus network recognition was 66.57%;when the IOU was between 0.55 and 0.95, the average accuracy of the improved NanoDet-Plus network recognition was 60.05%. The average accuracy of the improved NanoDet-Plus network recognition was significantly higher than that of traditional Faster R-CNN network, YOLO v6s network, NanoDet network, and NanoDet Plus network, and it performed the best in positioning accuracy and stability.From this, it could be concluded that the improved NanoDet-Plus network had good recognition performance and could be used for forestry pest and disease detection, which could improve prevention and control efficiency.

Key words: improved NanoDet-Plus network, tree, pine wood nematode disease, small goals, diseases and pests, recognition