面向无人机前端轻量级应用的输电线路鸟巢智能检测

Bird’s nest intelligent detection on transmission lines for unmanned aerial vehicle front-end lightweight application

利用鸟巢颜色单一的特点,提出一种将YOLOX与颜色空间相结合的鸟巢检测方法,以提高巡检无人机在复杂输电线路背景下对鸟巢检测的准确率.在前端设备Jetson Xavier NX里,采用经过模型调优的YOLOX目标检测网络对鸟巢图像进行检测并截取区域子图;根据颜色空间分布过滤非鸟巢区域,实现鸟巢的精筛.实验结果表明,采用上述方法对测试集中的鸟巢图像进行检测,准确率可达97.20%.

In order to improve the detection accuracy of bird’s nest for unmanned aerial vehicle in complex transmission line backgrounds,a bird nest detection method is proposed which combines YOLOX with color space based on the characteristic of single color of bird nests.In the front-end device Jetson Xavier NX,a YOLOX object detection network that has been fine-tuned and trained is used to detect and crop sub-images of bird’s nest images.The non-nest regions are filtered according to the color space distribution to achieve the fine screening of bird nests.The experimental results show that the detection of bird nest images in the test set using the above method can reach an accuracy of 97.20%.