基于深度学习的变电站机器人仪表检测研究

为提升变电站巡检机器人对仪表的识别检测能力,将深度学习技术应用于变电站仪表检测中,提出一种适用于变电站巡检机器人的轻量级卷积神经网络.该网络以深度可分离卷积替代传统卷积作为特征提取基本单元,有效降低网络参数.同时,针对不同尺寸的目标,采用反卷积多通道融合和多支路空洞卷积相结合的方式,提升网络对图像多尺度特征信息的获取,保证网络检测精度.通过实验表明,所提网络在目标检测精度以及效率上都有较大提升,并能较好地应用于变电站巡检机器人仪表检测中,实现高效识别检测.

Research on Instrument Detection Method of Substation Inspection Robot Based on Deep Learning

In order to improve the detection ability to the instrument of substation inspection robot, a lightweight convolutional neural network was used to detect substation instrument. In this network, the traditional convolution is replaced by the depth separable convolution as the basic unit of feature extraction, which effectively reduces the network parameters. At the same time, in view of different size objects, the method of deconvolution multi-channel fusion and multi-branch dilated convolution is used to improve the network''s multi-scale feature information acquisition to ensure network detection accuracy. Experimental results show that the proposed network improves the accuracy and efficiency of object detection effectively, and it can be better applied to the instrument detection of substation inspection robot and achieves high-efficiency detection.