基于深度学习的遥感影像松材线虫病树提取

Remote Sensing Image Extraction of Pine Wood Nematode Disease Tree Based on Deep Learning

摘要: 松材线虫病对我国松树类物种具有极大的伤害，需要对病虫区域进行准确高效的确定以提早防治。这种病在传播方式上具有跳跃特性。它具有传播途径多种多样、发病部位较隐蔽难以发现、病情潜伏时间长、发病速度迅速、治理不方便等特点，严重时会导致大量松树病死，导致环境和森林景观的严重损坏，并可能导致严重的经济和环境损失。本论文通过使用深度学习目标检测中的RetinaNet方法，将无人机拍摄的影像作为训练样本，充分利用深度学习目标检测方法的优势，将其对比SSD和YOLO v3方法的识别效果，实现病虫树木的高效判别。对松材线虫病树区域展开定位研究，在节省人工成本的同时能迅速防止病虫害对松树的疾病扩散，为清除和防治病害区域扩散至更大范围提供有效帮助。

Abstract: Pine wood nematode disease is very harmful to pine species in China, so it is necessary to determine the pest area accurately and efficiently in order to prevent it in advance. The disease has a leaping characteristic in its mode of transmission. It has the characteristics of diversity, strong concealment of transmission route, long incubation period of disease, fast transmission speed and inconvenient management. When it is serious, a large number of pine trees will die, causing serious damage to environment and forest landscape, and may lead to serious economic and environmental damage. In this paper, through the use of RetinaNet method in deep learning target detection, the unmanned aerial vehicle images are taken as training samples, making full use of the advantages of deep learning target detection method, the recognition effect of SSD and Yolo V3 method is compared, and the efficient identification of pest trees is realized. The research on the location of pine wood nematode disease tree area can save the labor cost and prevent the spread of diseases and insect pests on pine trees quickly, which can provide effective help for clearing and controlling the spread of disease area to a wider range.