基于轻量级深度卷积神经网络的绝缘子检测

在无人机巡检图像中,检测出绝缘子是实现输电线路状态分析的关键.本研究采用轻量级卷积神经网络代替传统的人工特征提取器,获取输入图像的深层特征;利用深度学习目标检测网络对所提取特征进行处理和训练学习,实现多尺度、多种类的绝缘子目标检测.实验结果表明:该方法可以准确快速地识别出以山林背景为主的瓷质和复合两类绝缘子,其检测精度分别达到96.29%和90.85%,且整体检测速度高达43 F·s~(-1),有效满足电力巡线中的绝缘子实时检测要求.

Insulator detection based on lightweight deep convolutional neural network

Efficient detection of insulators in aerial images is one of the key techniques for the intelligent inspection of trans-mission lines. This paper proposes a deep learning based insulator detection method to address the problem of the complexity of the feature extractor and detection efficiency. The proposed method utilizes lightweight convolutional neuron network instead of traditional artificial feature extractor to extract the deep image features. Then an object detection network of deep learning is employed to process the extracted features for multi-scale and multi-type insu-lator detection. The experimental results show that the proposed method can accurately and quickly detect the porce-lain insulator and composite insulator from the aerial images with the mountain forest background. The detection ac-curacies of two types of insulators are 96.29% and 90.85%, respectively. The obtained detection speed achieves 43 frames per second, which implies the proposed method can meet the real-time requirement of insulator detection for the transmission lines inspection.