摘要
为提升道路场景语义分割的性能以及实际应用性,将传统的图像处理算法与深度学习技术相结合,提出了一种多特征融合的轻量级道路场景语义分割网络模型。首先,该模型利用颜色空间转化、图像均衡化、边缘检测等算法对图像多种特征信息进行增强;其次,以深度可分离卷积为基本单元搭建高效率特征提取结构,对特征增强后的图像进行信息融合和提取,并结合跳层上采样操作完成初步分割;最后,引入边缘检测支路来对分割图像的目标边界信息进行细化,保障网络高精度分割。通过实验结果表明,所提网络在分割精度、计算效率上得到了较好的平衡,同时,在实际变电站道路场景应用中,该网络也能实现高效语义分割,为巡检机器人提供有效的道路信息。
In order to improve the performance and practical application of road scene semantic segmentation,a lightweight multi feature fusion network combining traditional image processing and deep learning technology was proposed for road scene semantic segmentation.Firstly,algorithms such as color space transformation,image equalization and edge detection were used to enhance image feature information.Secondly,the depth separable convolution was used as the basic unit to build a high-efficiency feature extraction structure,which was used for information fusion and extraction of the image after feature enhancement,and the preliminary segmentation was realized combined with the skip layer up sampling operation.Finally,the edge detection branch was introduced to refine the boundary information of the segmented image to ensure the high-precision segmentation of the network.Experimental results show that the proposed network effectively balances the network segmentation accuracy and computational efficiency.At the same time,in the actual application of substation road scene,the network can also achieve efficient semantic segmentation and provide effective road information for inspection robots.
作者
谷湘煜
刘晓熠
周仁彬
GU Xiang-yu;LIU Xiao-yi;ZHOU Ren-bin(Chengdu Branch, Shenzhen Launch Digital Technology Co., Ltd., Chengdu 610000, China)
出处
《科学技术与工程》
北大核心
2021年第33期14251-14257,共7页
Science Technology and Engineering
基金
广州市科技计划(202002020019)。
关键词
深度学习
图像处理
特征融合
道路场景
语义分割
deep learning
image processing
feature fusion
road scene
semantic segmentation
