基于Faster R-CNN的松材线虫病受害木识别与定位

松材线虫病是一种毁灭性松树传染病,其传播速度快、发病时间短、致病力强,及时发现、确定受害木的位置,并采取安全处理措施是目前控制松材线虫病蔓延的有效手段。本文通过小型无人机搭载可见光RGB数码相机获取超高空间分辨率影像,采用Faster R-CNN目标检测算法实现对染病变色松树的自动识别,与传统受害木识别方法不同,本文考虑了其他枯死树和红色阔叶树对受害木识别的影响。实验结果表明,根据受害木的冠幅大小修改区域生成网络中的锚框(anchor)尺寸,并考虑其他枯死树和红色阔叶树的影响,有利于提高受害木识别效果和检测精度。改进后受害木识别总体精度从75.64%提高到82.42%,提高了6.78个百分点,能够满足森林防护人员对受害木定位处理的需求。通过坐标转换的方式得到受害木的精确位置信息与空间分布情况,结合点位合并过程,最终正确定位出494棵受害木。本文通过无人机遥感结合目标检测算法能监测松材线虫病的发生和获取受害木的分布情况,可为松材线虫病的防控提供技术支持。

Detection and Location of Pine Wilt Disease Induced Dead Pine Trees Based on Faster R-CNN

Pine wilt disease (PWD) is a devastating infectious disease for the rapid spread, short disease period, and strong pathogenic ability. At present, detecting the PWD induced dead pine trees (DPT) timely and then taking corresponding measures are vital to control the spread of PWD. An unmanned aerial vehicle (UAV) platform equipped with the Vis-RGB digital camera was used to obtain the ultra-high spatial resolution images. Deep learning object detection of Faster R-CNN was adopted to detect the DPT automatically. Different from the previous research on the DPT identification, the influences of other dead trees and red broad leaved trees on DPT identification were considered. The results showed that Faster R-CNN can effectively identify the DPT. The 6.78 percentage points detection accuracy of the DPT would be improved when taking the anchor size, other dead trees and red broad-leaved trees into consideration. The overall accuracy of DPT detection can reach 82.42%, which can meet the protector for felling of the DPT. Finally, the position of predicted DPT was calculated accurately using coordinate transformation. Combined with the point combination process, 494 DPT were correctly located. This research had the advantages of low cost, high efficiency and automatic identification, and can provide technical support for the prevention and control of PWD. The combination of UAV remote sensing and object detection algorithms was a promising method to monitor the occurrence of PWD and the distribution of the DPT, which provided important basis for the consequence harmless treatment of PWD induced DPT.