A Monitoring Scheme for Pine Wood Nematode Disease Tree Based on Deep Learning and Ground Monitoring

Abstract

AbstractThe detection of pine nematode trees in forest areas is a very important task to prevent and control pine nematode epidemic, where the timely and accurate detection in large and complex scenarios is a challenging work. To address this issue, we propose a ground-based monitoring detection scheme to acquire the large-scale images, and a single-stage deep neural network is applied to detect the infected tree. By installing monitoring cameras on the ground foundation, cameras are controlled to collect images automatically according to predetermined parameters. We designed an improved YOLOv3 based detection model, named as YOLO-S. Specifically, an attention mechanism is added to the backbone feature extraction network and the bottom-up feature pyramid structure is added to the original feature pyramid to enhance the detection accuracy. Experiments results show that the proposed YOLO-S algorithm can provide higher detection accuracy than that of other models.KeywordsGround-based monitoringPine nematode diseaseAttention mechanismDeep learning