[1]宋 宇,王小瑀,梁 超,等.基于多级特征图联合上采样的实时语义分割[J].计算机技术与发展,2022,32(02):82-87.[doi:10. 3969 / j. issn. 1673-629X. 2022. 02. 013]
　SONG Yu,WANG Xiao-yu,LIANG Chao,et al.Real-time Semantic Segmentation Based on Multi-scale Feature Map Joint Pyramid Upsamping[J].,2022,32(02):82-87.[doi:10. 3969 / j. issn. 1673-629X. 2022. 02. 013]
点击复制

基于多级特征图联合上采样的实时语义分割()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年02期
页码:
82-87
栏目:
图形与图像
出版日期:
2022-02-10

文章信息/Info

Title:
Real-time Semantic Segmentation Based on Multi-scale Feature Map Joint Pyramid Upsamping
文章编号:
1673-629X(2022)02-0082-06
作者:
宋 宇; 王小瑀; 梁 超; 程 超
长春工业大学 计算机科学与工程学院,吉林 长春 130012
Author(s):
SONG Yu; WANG Xiao-yu; LIANG Chao; CHENG Chao
School of Computer Science and Engineering,Changchun University of Technology,Changchun 130012,China
关键词:
无人驾驶; 语义分割; 卷积神经网络; 深度学习; 空洞卷积
Keywords:
driverless; semantic segmentation; convolution neural network; deep learning; atrous convolution
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 02. 013
摘要:
视觉感知是无人驾驶技术中的重要一环,而语义分割技术又是实现视觉感知的主要技术手段之一。 现在的语义分割技术多采用计算量大、内存占用高的空洞卷积来提取高分辨率特征图,从而导致现在主流的语义分割网络分割速度不足,无法有效应用于无人驾驶的场景中。 针对这一问题,提出了一种实时性更好的语义分割网络。 首先,采用了一种轻量级的卷积神经网络作为编码器,并且使用跨步卷积和常规卷积替换了耗时、耗内存的空洞卷积。 然后,为了得到与DeepLabv v3+相似的特征图,提出了一种新的联合上采样模块:多级特征图联合上采样模块( multi-scale feature map jointpyramid upsamping,MJPU) ,通过融合编码器的多个特征图,生成了语义信息更加丰富的高分辨率特征图。 通过 Cityscapes数据集上的实验表明,相比于主流语义分割网络 Deeplabv3 +,该网络在不损失大量性能的前提下,可以将分割速度提高2郾 25 倍,达到 32. 3 FPS / s。 从而使网络具有更好的实时性,更加适合应用于无人驾驶场景。
Abstract:
Vision-based perception is an import link in driverless technology,and semantic segmentation is one of the main technique torealize visual perception in driverless technology. The current semantic segmentation technology mostly uses atrous convolution with alarge amount of computation and high memory consumption to extract high-resolution feature maps. As a result,the current mainstreamsemantic segmentation network lacks the segmentation speed and cannot be effectively applied in driverless technology. To solve thisproblem,a semantic segmentation network with better real - time performance is proposed. Firstly, a lightweight convolutional neuralnetwork is used as the encoder,and stride convolution and regular convolution are used to replace the time-consuming and memory-consuming atrous convolution. Secondly,in order to obtain the feature map similar to Deeplabv V3 +,a new joint upsampling module,multi-scale feature map joint pyramid upsamping,is proposed. By fusing multiple feature maps in the encoder,a high resolution feature mapwith richer semantic information is generated. Experiments on the Cityscapes dataset show that compared with the popular semantic segmentation network Deeplab V3+,the proposed network can improve the segmentation speed by 2. 25 times to 32. 3 FPS / s without losinga lot of performance. Therefore,the network proposed has better real-time performance and is more suitable for driverless scenes.

相似文献/References:

[1]张泽宇,郭 斌,张太红*.基于 DCNN 的马匹图像分割算法研究[J].计算机技术与发展,2020,30(10):210.[doi:10. 3969 / j. issn. 1673-629X. 2020. 10. 037]
　ZHANG Ze-yu,GUO Bin,ZHANG Tai-hong.Research on Horse Image Segmentation Algorithm Based on DCNN[J].,2020,30(02):210.[doi:10. 3969 / j. issn. 1673-629X. 2020. 10. 037]
[2]龚 安,赵 莉,姚鑫杰.基于改进的 U-Net 网络模型的气胸分割算法[J].计算机技术与发展,2021,31(10):173.[doi:10. 3969 / j. issn. 1673-629X. 2021. 10. 029]
　GONG An,ZHAO Li,YAO Xin-jie.Pneumothorax Segmentation Algorithm Based on Improved U-Net Network Model[J].,2021,31(02):173.[doi:10. 3969 / j. issn. 1673-629X. 2021. 10. 029]
[3]姚芷馨,张太红,赵昀杰.基于卷积神经网络的多模型交通场景识别研究[J].计算机技术与发展,2022,32(07):93.[doi:10. 3969 / j. issn. 1673-629X. 2022. 07. 016]
　YAO Zhi-xin,ZHANG Tai-hong,ZHAO Yun-jie.Research on Multi-model Traffic Scene Recognition Based on Convolution Neural Network[J].,2022,32(02):93.[doi:10. 3969 / j. issn. 1673-629X. 2022. 07. 016]
[4]张 鑫,陈 黎.基于多尺度特征融合的消防车通道占用检测[J].计算机技术与发展,2022,32(10):51.[doi:10. 3969 / j. issn. 1673-629X. 2022. 10. 009]
　ZHANG Xin,CHEN Li.Fire Truck Passages Occupancy Detection Based on Multi-scale Feature Fusion[J].,2022,32(02):51.[doi:10. 3969 / j. issn. 1673-629X. 2022. 10. 009]
[5]李松宇.基于 HED—UNet 遥感图像建筑物语义分割[J].计算机技术与发展,2022,32(S2):58.[doi:10. 3969 / j. issn. 1673-629X. 2022. S2. 010]
　LI Song-yu.Semantic Segmentation of Buildings in Remote Sensing Images Based on HED-UNet[J].,2022,32(02):58.[doi:10. 3969 / j. issn. 1673-629X. 2022. S2. 010]
[6]杜睿山,宋健辉,孟令东.基于注意力机制的岩石铸体薄片轻量化分割[J].计算机技术与发展,2023,33(10):128.[doi:10. 3969 / j. issn. 1673-629X. 2023. 10. 020]
　DU Rui-shan,SONG Jian-hui,MENG Ling-dong.Lightweight Segmentation of Rock Casting Sheet Based on Attention Mechanism[J].,2023,33(02):128.[doi:10. 3969 / j. issn. 1673-629X. 2023. 10. 020]
[7]刘贤梅,刘鹏飞,贾 迪,等.基于多特征融合的城市场景三维点云语义分割[J].计算机技术与发展,2023,33(11):78.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 012]
　LIU Xian-mei,LIU Peng-fei,JIA Di,et al.3D Point Cloud Semantic Segmentation of Urban Scene Based on Multi-feature Fusion[J].,2023,33(02):78.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 012]
[8]陈世婕,王卫星*,彭 莉.基于多尺度网络的苗绣绣片纹样分割算法研究[J].计算机技术与发展,2023,33(11):149.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 022]
　CHEN Shi-jie,WANG Wei-xing*,PENG Li.Research on Miao Embroidery Pattern Segmentation Algorithm Based on Multi-scale Network[J].,2023,33(02):149.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 022]
[9]曾碧卿,杨 睿,李一娴,等.基于卷积神经网络的零件圆检测方法[J].计算机技术与发展,2023,33(11):64.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 010]
　ZENG Bi-qing,YANG Rui,LI Yi-xian,et al.Part Circle Detection Method Based on Convolutional Neural Network[J].,2023,33(02):64.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 010]
[10]张 婧,张 策*,张 茹,等.图像分割述评:基本概貌、典型算法及比较分析[J].计算机技术与发展,2024,34(01):1.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 001]
　ZHANG Jing,ZHANG Ce*,ZHANG Ru,et al.Review of Image Segmentation:Basic Overview,Typical Algorithms and Comparative Analysis[J].,2024,34(02):1.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 001]

更新日期/Last Update: 2022-02-10