摘要
为提升变电站巡检机器人对道路场景的识别理解能力,将深度学习技术应用于变电站巡检机器人中,提出了一种适用于变电站道路场景的全卷积语义分割网络。该网络借鉴ENet编码结构提取图像特征,同时融入多种解码结构来获取更多有效特征,恢复图像目标信息。同时,针对巡检机器人以及变电站道路特点,将语义分割结果转化为机器人前方目标信息以及机器人偏离情况信息,辅助机器人导航避障。实验结果表明:所提出的网络有效地提升了图像分割精度,并能较好地适应于实际变电站环境中。研究结果为机器人提供了有效的道路场景信息,辅助机器人导航避障。
In order to improve the ability of substation inspection robots to recognize and understand road scenes,a full convolution semantic segmentation network based on deep learning technology was proposed to segment substation road scene.The network draws on the ENet coding structure to extract image features,and multiple decoding structures were fused to obtain more effective features to recover image target information.At the same time,according to the characteristics of inspection robots and substation roads,the semantic segmentation results were converted into robot target information and robot deviation information to assist robot navigation and obstacle avoidance.Experimental results show that the proposed network improves the image segmentation accuracy effectively.Meanwhile,it shows excellent scene recognition performance in actual substation scenarios and provides effective road scene information for the robots.
作者
鲜开义
杨利萍
周仁彬
梁洪军
蒋鑫
查盛
XIAN Kai-yi;YANG Li-ping;ZHOU Ren-bin;LIANG Hong-jun;JIANG Xin;ZHA Sheng(Shenzhen Launch Digital Technology Co.,Ltd.,Chengdu 610000,China)
出处
《科学技术与工程》
北大核心
2020年第15期6151-6157,共7页
Science Technology and Engineering
基金
四川省重大科技专项(18ZDZX0162)。
关键词
全卷积神经网络
语义分割
变电站巡检机器人
避障
fully convolutional network
semantic segmentation
substation inspection robot
obstacles avoidance