基于改进DeepLabV3+的轻量化SAR图像冰间水道分割

doi:10.3778/j.issn.1002-8331.2401-0300

摘要/Abstract

摘要： 冰间水道对极地航行具有重要意义。为获取准确实时的冰间水道分布情况，提出了一种基于DeepLabV3+的轻量化合成孔径雷达（synthetic aperture radar，SAR）图像冰间水道分割模型。采用MobileNetV3作为主干网络，提出改进的空洞空间金字塔池化模块（SE-GASPP）实现编码器部分轻量化及通道特征增强；采用局部注意力模块提取主干网络输出特征的局部信息，并提出形状保持模块生成形状注意力以关注水道形状信息和改进特征表示；联合形状损失与分割损失训练模型。为解决冰间水道分割任务训练数据有限的问题，利用SAR双极化数据构建数据集。实验结果表明，所提轻量化分割模型在参数量仅为DeepLabV3+的7.8%、浮点运算次数下降70.56%、推理速度提高10%的情况下，其平均交并比提高了4.92个百分点且优于其他分割模型，实现模型轻量化的同时具有较高的分割性能。

关键词: 合成孔径雷达（SAR）, 冰间水道分割, 轻量化, 局部注意力, 形状注意力

Abstract: Sea ice leads are of great significance for polar navigation. In order to obtain accurate and real-time distribution of leads, a lightweight synthetic aperture radar (SAR) image segmentation model based on DeepLabV3+ is proposed. MobileNetV3 is employed as the backbone network, and an improved atrous spatial pyramid pooling module using dilated ghost convolution and squeeze-and-excitation channel attention (SE-GASPP) is proposed to achieve encoder lightweight and channel feature enhancement. A local attention module is employed to extract local information from the output features of the backbone network, and a shape preservation module is proposed to generate shape attention to focus on lead shape information and improve feature representation. The model is trained by combining shape loss and segmentation loss. To solve the problem of limited training data for lead segmentation tasks, a dataset is constructed by using SAR dual-polarization data. The experimental results show that the proposed lightweight segmentation model improves the mean intersection over union by 4.92?percentage points, and is superior to other segmentation models, when the parameters are only 7.8% of DeepLabV3+, the floating-point operations are reduced by 70.56%, and the reasoning speed is increased by 10% compared to DeepLabV3+. The proposed lightweight segmentation model has a high segmentation performance while achieving lightweight.

Key words: synthetic aperture radar (SAR), lead segmentation, lightweight, local attention, shape attention

宋巍, 祝敏, 石少华, 柳彬, 贺琪. 基于改进DeepLabV3+的轻量化SAR图像冰间水道分割[J]. 计算机工程与应用, 2025, 61(10): 258-266.

SONG Wei, ZHU Min, SHI Shaohua, LIU Bin, HE Qi. Lightweight SAR Image Lead Segmentation Based on Improved DeepLabV3+[J]. Computer Engineering and Applications, 2025, 61(10): 258-266.

参考文献

[1] 屈猛，庞小平, 赵羲, 等. 利用多源遥感数据识别波弗特海冰间水道[J]. 武汉大学学报 (信息科学版), 2019, 44(6): 917-924.
QU M, PANG X P, ZHAO X, et al. Detection of sea ice lead in Beaufort Sea based on multisensory remote sensing images[J]. Geomatics and Information Science of Wuhan University, 2019, 44(6): 917-924.
[2] 屈猛, 赵羲, 庞小平, 等. 北极冰间水道区域的物理过程和遥感观测研究进展[J]. 地球科学进展, 2022, 37(4): 382-391.
QU M, ZHAO X, PANG X P, et al. Review of Arctic Sea ice leads: physics and remote sensing[J]. Advances in Earth Science, 2022, 37(4): 382-391.
[3] HOFFMAN J P, ACKERMAN S A, LIU Y H, et al. Application of a convolutional neural network for the detection of sea ice leads[J]. Remote Sensing, 2021, 13(22): 4571.
[4] 罗杨, 卞春江, 陈红珍. 基于特征解耦的SAR图像舰船检测蒸馏[J]. 计算机工程与应用, 2024, 60(2): 171-179.
LUO Y, BIAN C J, CHEN H Z. Distillation for SAR ship detectors based on decoupled features[J]. Computer Engineering and Applications, 2024, 60(2): 171-179.
[5] KOMAROV A S, BUEHNER M. Adaptive probability thresholding in automated ice and open water detection from RADARSAT?2 images[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(4): 552-556.
[6] MURASHKIN D, SPREEN G, HUNTEMANN M, et al. Method for detection of leads from Sentinel-1 SAR images[J]. Annals of Glaciology, 2018, 59(76pt2): 124-136.
[7] LIANG Z Y, PANG X P, JI Q, et al. An entropy-weighted network for polar sea ice open lead detection from Sentinel-1 SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 4304714.
[8] LIU S W, LI M C, XU M M, et al. An improved lightweight U-Net for sea ice lead extraction from multi-polarization SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 2000705.
[9] CHEN L C, ZHU Y K, PAPANDREOU G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]//Proceedings of the 15th European Conference on Computer Vision. Cham: Springer, 2018: 833-851.
[10] HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[J]. arXiv:1704.04861, 2017.
[11] 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, 2018, 40(4): 834-848.
[12] HAN K, WANG Y H, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 1577-1586.
[13] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 7132-7141.
[14] XU Y Y, XIE Z, FENG Y X, et al. Road extraction from high-resolution remote sensing imagery using deep learning[J]. Remote Sensing, 2018, 10(9): 1461.
[15] 沈怀艳, 吴云. 基于MSFA-Net的肝脏CT图像分割方法[J]. 计算机科学与探索, 2023, 17(3): 646-656.
SHEN H Y, WU Y. Liver CT image segmentation method based on MSFA-net[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(3): 646-656.
[16] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 770-778.
[17] MILLETARI F, NAVAB N, AHMADI S A. V-net: fully convolutional neural networks for volumetric medical image segmentation[C]//Proceedings of the 2016 4th International Conference on 3D Vision, 2016: 565-571.
[18] 龙丽红, 朱宇霆, 闫敬文, 等. 新型语义分割D-UNet的建筑物提取[J]. 遥感学报, 2023, 27(11): 2593-2602.
LONG L H, ZHU Y T, YAN J W, et al. New building extraction method based on semantic segmentation[J]. National Remote Sensing Bulletin, 2023, 27(11): 2593-2602.
[19] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2017: 2999-3007.
[20] FILIPPONI F. Sentinel-1 GRD preprocessing workflow[C]//Proceedings of the 3rd International Electronic Conference on Remote Sensing, 2019: 11.
[21] ZAKHVATKINA N, SMIRNOV V, BYCHKOVA I, et al. Detection of the leads in the Arctic drifting sea ice on SAR images[C]//Proceedings of the 2021 IEEE International Geoscience and Remote Sensing Symposium. Piscataway: IEEE, 2021: 4276-4279.
[22] HONG D B, YANG C S. Automatic discrimination approach of sea ice in the Arctic Ocean using Sentinel-1 Extra Wide Swath dual-polarized SAR data[J]. International Journal of Remote Sensing, 2018, 39(13): 4469-4483.
[23] PASSARO M, MüLLER F L, DETTMERING D. Lead detection using Cryosat-2 delay-Doppler processing and Sentinel-1 SAR images[J]. Advances in Space Research, 2018, 62(6): 1610-1625.
[24] RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//Proceedings of the 18th International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2015: 234-241.
[25] CHEN J N, LU Y Y, YU Q H, et al. TransUNet: transformers make strong encoders for medical image segmentation[J]. arXiv:2102.04306, 2021.
[26] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 6000-6010.