融合边缘检测模块的自然地貌语义分割模型研究

doi:10.16182/j.issn1004731x.joss.20-0756

摘要/Abstract

摘要：

针对遥感图像自然地貌边缘的像素点归类问题，提出融合边缘检测模块的多通道融合模型与解码器端模块模型。边缘检测模块以Canny算子为基础进行闭运算及均值滤波处理得到精确化的图像边缘。语义分割网络以DeepLabV3+为基础，分别从编码器及解码器端并联边缘计策模块。实验结果表明，改进后的2种网络相比原DeepLabV3+网络在高分辨率自然地貌图像数据集上均取得更好的分割效果，且解码器端融合网络取得了最高72.60%的交互比(IoU ，intersection over union)和86.64%的F1score，可用于面向自然地貌的识别与分割。

关键词: 语义分割, deeplabv3+, 边缘检测, 自然地貌

Abstract:

To classify pixels of natural landform edges in remote sensing images, this paper proposes a multi-channel fusion model and a decoder-side module model both integrating an edge detection module. The edge detection module takes the Canny operator as the base to perform closed operations and mean filtering, as a result of which accurate image edges can be achieved. Based on DeepLabV3+, the semantic segmentation network is connected with an edge planning module in parallel at encoder and decoder sides respectively. The experimental results show that the two improved networks can achieve a better segmentation effect on a high-resolution natural landform image data set compared with the original DeepLabV3+ network. Particularly, the network with fusion at the decoder side achieves the highest intersection over union (IoU) of 72.60% and F1score of 86.64%, which can be used for the recognition and segmentation of natural landforms.

Key words: semantic segmentation, deeplabv3+, edge detection, natural landform

中图分类号:

TP391

沈祺宗, 高春艳. 融合边缘检测模块的自然地貌语义分割模型研究[J]. 系统仿真学报, 2022, 34(2): 293-302.

Qizong Shen, Chunyan Gao. Research on Semantic Segmentation of Natural Landform Based on Edge Detection Module[J]. Journal of System Simulation, 2022, 34(2): 293-302.

图/表 19

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

图 13

表 1

图14

表 2

表 3

表4

图 15

参考文献 15

1	Badrinarayanan V, Kendall A, Cipolla R. SegNet: A Deep Convolutional Encoder-Decoder Architecture for Scene Segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence(S0162-8828), 2017, 39(12) 2481-2495.
2	Lin G, Milan A, Shen C, et al. RefineNet: Multi-path Refinement Networks for High-Resolution Semantic Segmentation[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, 2017:5168-5177.
3	Liu Z, Li X, Ping L, et al. Deep Learning Markov Random Field for Semantic Segmentation[J]. IEEE Trans Pattern Anal Mach Intell(S0162-8828), 2018: 1814-1828.
4	陈聪. 基于深层网络的城市街道场景的语义分割的方法研究[D]. 西安: 西安电子科技大学, 2019.
	Chen Cong.Reserch on Semantic Segmentation of Cityscape Based on Deep Neural Network[D].Xi′an Xidian University, 2019.
5	Long J, Shelhamer E, Darrell T. Fully Convolutional Networks for Semantic Segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence(S0162-8828), 2014, 39(4): 640-651.
6	林志洁, 罗壮, 赵磊, 等.特征金字塔多尺度全卷积目标检测算法[J].浙江大学学报(工学版), 2019, 53(3): 533-540.
	Lin Zhijie, Luo Zhuang, Zhao Lei, et al. Multi-scale Full Convolution Target Detection Algorithm for Feature Pyramid [J]. Journal of Zhejiang University (Engineering Edition), 2019, 53(3): 533-540.
7	Wu Dihua,Yin Xuqiang,Jiang Bo,et al. Detection of the Respiratory Rate of Standing Cows by Combining the Deeplab V3+ Semantic Segmentation Model with the Phase-based Video Magnification algorithm[J]. Biosystems Engineering(S1615-7591), 2020: 72-89.
8	Chen L C, Papandreou G, Kokkinos I, et al. Deeplab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected Crfs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence(S0162-8828), 2018, 40(4): 834-848.
9	Tang Wei,Zou Dongsheng,Yang Su,et al. A Two-stage Approach for Automatic Liver Segmentation with Faster R-CNN and DeepLab[J]. Neural Computing and Applications(S0941-0643), 2020, 32(11): 6769-6778.
10	Wu Dihua,Yin Xuqiang,Jiang Bo,et al. Detection of the Respiratory Rate of Standing Cows by Combining the Deeplab V3+ Semantic Segmentation Model with the Phase-based Video Magnification Algorithm[J]. Biosystems Engineering(S1537-5110), 2020: 72-89.
11	Zhao H, Shi J, Qi X, et al. Pyramid Scene Parsing Network[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2881-2890.
12	Lin G, Milan A, Shen C, et al. Refinenet: Multi-path Refinement Networks for High-resolution Semantic Segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1925-1934.
13	Xu Wang,Chen Renwen,Huang Bin,et al. Single Image Super-Resolution Based on Global Dense Feature Fusion Convolutional Network[J]. Sensors(S1424-8220), 2019, 19(2): 316-335.
14	Liu Z, Li X, Ping L, et al. Deep Learning Markov Random Field for Semantic Segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence(S0162-8828), 2018, 40(8): 1418-1428.
15	Zhang Xiao, Chen Fuen. Lane Line Edge Detection Based on Improved Adaptive Canny Algorithm[J]. Journal of Physics: Conference Series(S1742-6588), 2020, 1549(2): 1362-1375.

方法	植被茂盛		沙漠戈壁
方法	mIoU	F1score	mIoU	F1score
FC-DenseNet56	62.88	81.35	62.95	79.49
GCN	67.75	81.23	67.37	80.67
SegNet	67.25	84.44	68.19	84.28
DeepLabV3	70.47	85.15	70.26	85.06
多通道输入	72.51	86.59	72.36	86.13
解码器输入	72.60	86.64	72.44	86.25

地貌类型	标签	RGB	识别精度
山地岩石	01	（128,128,0）	0.638
沙漠沙砾	02	（128,0,128）	0.657
平原草地	03	（0,128,0）	0.712
河流水面	04	（0,0,128）	0.614

地貌类型	标签	RGB	识别精度
山地岩石	01	（128,128,0）	0.640
沙漠沙砾	02	（128,0,128）	0.659
平原草地	03	（0,128,0）	0.715
河流水面	04	（0,0,128）	0.617

地貌类型	标签	RGB	识别精度
山地岩石	01	（128,128,0）	0.643
沙漠沙砾	02	（128,0,128）	0.661
平原草地	03	（0,128,0）	0.720
河流水面	04	（0,0,128）	0.619

[1]	陈立家, 王凯, 李世刚, 田延飞. 基于航空影像的航海模拟器视景快速建模方法[J]. 系统仿真学报, 2021, 33(7): 1565-1573.
[2]	蔡兴泉, 杨哲, 何鑫, 陈超. 结合伺服电机的古建筑物增强现实交互展示系统[J]. 系统仿真学报, 2020, 32(6): 1145-1154.
[3]	金炜东, 胡燕花, 唐鹏, 李伟. 基于结构化森林的前向车载视频图像的细节强化[J]. 系统仿真学报, 2018, 30(12): 4602-4610.
[4]	孙逊, 王玄玉, 王献, 刘顺华. 实时修正的烟幕随机游走仿真模型[J]. 系统仿真学报, 2016, 28(9): 1979-1984.
[5]	姚莉, 吴琼颖, 许辉. 基于网格映射的虚拟视点合成[J]. 系统仿真学报, 2016, 28(10): 2321-2328.