无人机航拍图像中裸露地表的识别

电力输电线路下方或附近的无覆盖物的裸露地表,是引起输电线路事故的主要隐患之一;从无人机电力巡检航拍图像中识别裸露地表可以预防类似事故的发生;由于Mask RCNN识别无人机电力巡检航拍图像中裸露地表的精度较低,提出一种图像特征融合的方法,即人工提取HOG和LBP两种不同的图像特征,经过不同权重的融合共同表征图像中裸露地表区域的特征,再对SVM进行训练并用于识别;实验结果表明,采用该方法识别率可以达到80％以上,识别时间少于60 ms;HOG和LBP两种特征在进行融合时,当两种特征的数量级相当时,得到的识别率最高;可见,该方法在具有较高识别率的同时,具有比较好的实时性,适合于无人机机载平台对航拍图像的初筛,且训练时间较少,权重参数规模小,为无人机航拍图像中目标物的识别提供一种新思路.

Recognizing Bared Earth Region of Aerial Images from UAV

The bared earth under or near power transmission lines is one of the main hidden trouble. It is one of the important tasks to recognize the bared earth region from aerial images of Unmanned Aerial Vehicle(UAV) power inspection. After failing to employ the Mask RCNN to recognize the bared earth, this paper proposed an approach of image feature fusion. The HOG and LBP features of aerial images were extracted and fused by different weights. The experiments show that, (1)the average precision of recognizing the bared earth reaches 80%; (2) the average recognizing precision is best when the weights of two features make the order of magnitude of the two features equal. The approach proposed in this paper is suitable for the first image filting by the UAV airborne platform because it has not only quite good recognizing precision but also quite good real-time performance, which provides a new idea for target recognition in UAV aerial images.