纺织学报 ›› 2024, Vol. 45 ›› Issue (01): 112-119.doi: 10.13475/j.fzxb.20230103301

• 纺织工程 • 上一篇    下一篇

基于边缘填充的单兵迷彩伪装小目标检测

池盼盼1, 梅琛楠1, 王焰2, 肖红2, 钟跃崎1,3()   

  1. 1.东华大学 纺织学院, 上海 201620
    2.军事科学院 系统工程研究院, 北京 100010
    3.东华大学 纺织面料技术教育部重点实验室, 上海 201620
  • 收稿日期:2023-01-28 修回日期:2023-09-25 出版日期:2024-01-15 发布日期:2024-03-14
  • 通讯作者: 钟跃崎(1972—),男,教授,博士。研究方向为数字化纺织服装及人工智能技术应用。E-mail:zhyq@dhu.edu.cn。
  • 作者简介:池盼盼(1998—),女,硕士生。主要研究方向为基于深度学习的迷彩单兵检测。
  • 基金资助:
    上海市自然科学基金项目(21ZR1403000);国家自然科学基金项目(61572124)

Single soldier camouflage small target detection based on boundary-filling

CHI Panpan1, MEI Chennan1, WANG Yan2, XIAO Hong2, ZHONG Yueqi1,3()   

  1. 1. College of Textiles, Donghua University, Shanghai 201620, China
    2. Systems Engineering Research Institute, Academy of Military Sciences, Beijing 100010, China
    3. Key Laboratory of Textiles Science & Technology, Ministry of Education, Donghua University, Shanghai 201620, China
  • Received:2023-01-28 Revised:2023-09-25 Published:2024-01-15 Online:2024-03-14

RichHTML

11

PDF (PC)

28

摘要：

针对迷彩单兵识别存在伪装对象与背景高度相似融合、目标尺寸小等问题,提出了基于边缘填充的单兵迷彩伪装小目标检测模型BFNet(boundary-filled network)。该网络以SCNet(sparse complex-valued neural network)作为骨干网络,在网络的边缘引导阶段,利用边缘先验信息以及边缘的周围环境来挖掘目标信息。在上下文聚合阶段,利用上一级的预测值,使网络学习预测背景与前景的相互关系。实验结果表明:与最先进的BGNet相比,BFNet平均精度提升了0.74%,交并比识别率提升了1.35%,同时自适应E度量、加权F度量以及结构相似度与加权自适应F度量均得到了提高,其中,自适应E度量提升了0.85%,加权F度量提升了0.71%,证明所提出的BFNet能在更大程度上识别出单兵迷彩伪装小目标,且识别精度也得到提升。

关键词: 单兵迷彩伪装自动检测, 伪装物体识别, 深度学习, 小目标检测, 伪装物体分割

Abstract:

Objective In the automatic detection of siagle soldier camouflage, it is necessary to detect the targets at a long distance. In this scenario, the small size of the camouflaged target and the intensification of background fusion substantially increase the difficulty of detection. Therefore, a deep learning approach to tackle this challenge is proposed based on the deep learning network architecture and module structure.

Method The original dataset was extended using data augmentation and the network architecture was designed based on the BGNet model. SCNet was used for feature extraction of images, and EAM (edge-aware module) was used for detecting target edges. EFM (edge-guidance feature module) made use of the output of EAM to guide the network to locate and identify targets, NCD (neighbor con-connection decoder) was used for fusing the features from EFM output, and the CAM (context aggregation module) was employed to aggregate multi-level features to obtain the final output.

Results The quantitative results of the proposed model and the other models showed that PFNet performed poorly in this small target detection, and SINet-V2 and C2FNet had higher recognition rates but with lower recognition accuracy, indicating poor detection accuracy although they intersect with the true values. On the other hand, the BGNet model had lower recognition rates but with higher accuracy and structural similarity. The BFNet proposed in this paper was improved based on the BGNet, and after the improvement, the recognition rate was increased. At the same time, other indices measuring detection accuracy and object similarity were also improved. The proposed BFNet was found to be able to take both recognition rate and accuracy rate into account, and identify targets more accurately and comprehensively. The quantitative evaluation of the ablation experiments was carried out, and it showed that the modified EFM improved the recognition rateIby 1.35%, indicating that more targets are able to be recognized after the improvement. The modified CAM improved the recognition rate I by 0.51%, indicating that the improved CAM further improved the recognition rate I, while S, a measure of structural similarity, and the adaptive F value Fad were also hoisted, indicating that the recall rate was also improved considering the accuracy. With the modified EFM and CAM, the detection accuracy pA was slightly decreased, but the I value is improved by 1.87%. After modifying EFM and CAM, the accuracy pA was improved by 1.74% using SCNet (self-calibrated networks) as the backbone model, proving the SCNet model compensation for the decrease in accuracy caused by the improved module structure. The results of the final improvement scheme showed that the improvement rate of pA was 0.74% and the improvement rate of I was 1.35%, while the adaptive E metric Eϕadand weighted F-measure Fwβ were improved by 0.85% and 0.71%, respectively. The qualitative comparison of the proposed model with other models is shown. The baseline model could barely recognize small targets, while the improved model performs well in small camouflage target recognition task.

Conclusion The experimental results show that the proposed model performs well in the automatic detection tasks of single soldier camouflage, which indicates that the detection model in COS (camouflage object segmentation) task is suitable for single soldier camouflage detection, and the improved model offers higher the recognition rate, especially for detecting small target. The detection algorithm can be used as an aid for combatants and also provides an effective means to evaluate camouflage designs.

Key words: camouflage detection, camouflage object recognition, deep learning, small target detection, camouflage object segmentation

中图分类号: 

  • TS131.9

图1

部分数据集展示"

图2

BFNet模型整体架构 f1~f5为经过SCNet不同卷积层输出的多通道特征图。"

图3

SCNet模型的自校准卷积 注:Conv2D表示二维卷积。"

图4

改进后的边缘引导模块 注:D表示下采样;Max pooling表示最大池化; ?表示逐元相乘;Sum表示特征图内元素加和;Conv1D表示一维卷积; ⊕表示逐元素相加;Conv2D表示二维卷积;GAP表示全局平均池化; f i为当前的第i 级特征; f e为 边缘预测值;Out 为模块输出值。"

图5

改进后的上下文聚合模块 注:U表示上采样;-表示取反; ?表示逐元相乘;Conv1×1表示1×1卷积; ⊕表示逐元素相加;Conv3×3表示3×3卷积,其中D表示扩张率;h为上一级高级特征;l 为当前低级特征。"

图6

与其它方法检则效果的定性比较 注:图中每行依次表示原图和标注真值以及使用不同网络方法的检测结果图。"

图7

消融实验的定性分析 注:图中每行依次表示原图和标注真值以及使用不同网络方法的检测结果图。"

表2

与几种文献出现方法的定量比较"

方法 评估指标
p A I $E_{\phi}^{\mathrm{ad}}$ F a d M S F β w
PFNet[13] 0.784 0.968 0.942 0.807 0.004 0.867 0.762
SINet-V2[12] 0.779 0.986 0.928 0.830 0.004 0.877 0.777
C2FNet[6] 0.785 0.981 0.946 0.816 0.004 0.872 0.772
BGNet[14] 0.814 0.964 0.943 0.840 0.003 0.891 0.801
BFNet(本文方法) 0.820 0.977 0.951 0.846 0.003 0.892 0.803

表3

消融实验的定量评估"

序号 方法 评估指标
EFM+ EFM CAM+ CAM SCNet 随机裁剪 pA I $E_{\phi}^{\mathrm{ad}}$ Fad M S F β w
1# √ √ √ 0.814 0.964 0.943 0.840 0.003 0.891 0.801
2# √ √ √ 0.810 0.977 0.949 0.845 0.003 0.890 0.799
3# √ √ √ √ 0.818 0.973 0.952 0.847 0.003 0.892 0.804
4# √ √ √ 0.806 0.982 0.945 0.847 0.003 0.892 0.798
5# √ √ √ √ 0.820 0.977 0.951 0.846 0.003 0.892 0.803
6# √ √ √ 0.815 0.977 0.953 0.847 0.003 0.892 0.806
[1] TANKUS A, YESHURUN Y. Convexity-based visual camouflage breaking[J]. Computer Vision and Image Understanding, 2009, 82(3):208-237.
doi: 10.1006/cviu.2001.0912
[2] NAGABHUSAN U N. Camouflage defect identification: a novel approach[C]// 9th International Conference on Information Technology (ICIT'06). Orlando FL: IEEE Computer Society, 2006: 145-148.
[3] SENGOTTUVELAN P, WAHI A, SHANMUGAM A. Performance of decamouflaging through exploratory image analysis[C]// 2008 First International Conference on Emerging Trends in Engineering and Technology. Nagpur: IEEE Computer Society, 2008: 6-10.
[4] ZHENG Yunfei, ZHANG Xiongwei, CAO Tieyong, et al. Detection of people with camouflage pattern via dense deconvolution network[J]. IEEE Signal Processing Letters, 2019, 26(1):29-33.
doi: 10.1109/LSP.2018.2825959
[5] FANG Zheng, ZHANG Xiongwei, DENG Xiaotong, et al. Camouflage people detection via strong semantic dilation network[C]// Proceedings of the ACM Turing Celebration Conference-China. New York: Association for Computing Machinery, 2019:1-7.
[6] SUN Yujia, CHEN Geng, ZHOU Tao, et al. Context-aware cross-level fusion network for camouflaged object detection[C]// Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence Main Track: IJCAI, 2021:1025-1031.
[7] YANG F, ZHAI Q, LI X, et al. Uncertainty-guided transformer reasoning for camouflaged object detec-tion[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal: IEEE/CVF, 2021: 4146-4155.
[8] LE T N, NGUYEN T V, NIE Z, et al. Anabranch network for camouflaged object segmentation[J]. Computer Vision and Image Understanding, 2019, 184: 45-56.
doi: 10.1016/j.cviu.2019.04.006
[9] ZHAI Qiang, LI Xin, YANG Fan, et al. Mutual graph learning for camouflaged object detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE/CVF, 2021: 12997-13007.
[10] CHEN Tianyou, XIAO Jin, HU Xiaoguang, et al. Boundary-guided network for camouflaged object detection[J]. Knowledge-Based Systems, 2022.DOI:10.1016/j.knosys.2022.108901.
[11] LV Y, Zhang J, DAI Y, et al. Simultaneously localize, segment and rank the camouflaged objects[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE/CVF, 2021: 11591-11601.
[12] FAN Dengping, JI Gepeng, CHENG Mingming, et al. Concealed Object Detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(10): 6024-6042.
doi: 10.1109/TPAMI.2021.3085766
[13] MEI Haiyang, JI Gepeng, WEI Ziqi, et al. Camouflaged object segmentation with distraction mining[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE/CVF, 2021: 8772-8781.
[14] SUN Yujia, WANG Shuo, CHEN Chenglizhao, et al. Boundary-guided camouflaged object detection[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal: IEEE/CVF, 2022:1335-1341.
[15] LIU Jiangjiang, HOU Qibin, CHENG Mingming, et al. Improving convolutional networks with self-calibrated convolutions[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE/CVF, 2020:10093-10102.
[16] WEI Jun, WANG Shuhui, HUANG Qingming. F3NET: fusion, feedback and focus for salient object detection[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7):12321-12328.
doi: 10.1609/aaai.v34i07.6916
[17] XIE Enze, WANG Wenjia, WANG Wenhai, et al. Segmenting transparent objects in the wild[C]// Computer Vision-ECCV 2020: 16th European Conference. Berlin:Springer-Verlag, 2020:696-711.
[18] PERAZZI F, KRAHENBULHL P, PRITCH Y, et al. Saliency filters: Contrast based filtering for salient region detection[C]// 2012 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE/CVF, 2012: 733-740.
[19] FAN Dengping, JI Gepeng, CHENG Mingming, et al. Cognitive vision inspired object segmentation metric and loss function[J]. Scientia Sinica Informationis, 2021, 51(9):1475-1489.
doi: 10.1360/SSI-2020-0370
[20] MARGOLIN Ran, ZELNIK-MANOR L, TAL A. How to evaluate foreground maps?[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2014: 248-255.
[21] FAN Dengping, CHENG Mingming, LIU Yun, et al. Structre-measure: a new way to evaluate foreground maps[C]// 2017 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2017:4558-4567.
[1] 陆伟健, 屠佳佳, 王俊茹, 韩思捷, 史伟民. 基于改进残差网络的空纱筒识别模型[J]. 纺织学报, 2024, 45(01): 194-202.
[2] 杨宏脉, 张效栋, 闫宁, 朱琳琳, 李娜娜. 一种高鲁棒性经编机上断纱在线检测算法[J]. 纺织学报, 2023, 44(05): 139-146.
[3] 顾冰菲, 张健, 徐凯忆, 赵崧灵, 叶凡, 侯珏. 复杂背景下人体轮廓及其参数提取[J]. 纺织学报, 2023, 44(03): 168-175.
[4] 李杨, 彭来湖, 李建强, 刘建廷, 郑秋扬, 胡旭东. 基于深度信念网络的织物疵点检测[J]. 纺织学报, 2023, 44(02): 143-150.
[5] 陈佳, 杨聪聪, 刘军平, 何儒汉, 梁金星. 手绘草图到服装图像的跨域生成[J]. 纺织学报, 2023, 44(01): 171-178.
[6] 王斌, 李敏, 雷承霖, 何儒汉. 基于深度学习的织物疵点检测研究进展[J]. 纺织学报, 2023, 44(01): 219-227.
[7] 安亦锦, 薛文良, 丁亦, 张顺连. 基于图像处理的纺织品耐摩擦色牢度评级[J]. 纺织学报, 2022, 43(12): 131-137.
[8] 陈金广, 李雪, 邵景峰, 马丽丽. 改进YOLOv5网络的轻量级服装目标检测方法[J]. 纺织学报, 2022, 43(10): 155-160.
[9] 江慧, 马彪. 基于服装风格的款式相似度算法[J]. 纺织学报, 2021, 42(11): 129-136.
[10] 杨争妍, 薛文良, 张传雄, 丁亦, 马颜雪. 基于生成式对抗网络的用户下装搭配推荐[J]. 纺织学报, 2021, 42(07): 164-168.
[11] 许倩, 陈敏之. 基于深度学习的服装丝缕平衡性评价系统[J]. 纺织学报, 2019, 40(10): 191-195.
[12] 刘正东, 刘以涵, 王首人. 西装识别的深度学习方法[J]. 纺织学报, 2019, 40(04): 158-164.
[13] 汪珊娜 张华熊 康锋. 基于卷积神经网络的领带花型情感分类[J]. 纺织学报, 2018, 39(08): 117-123.
[14] 何晓昀 韦平 张林 邓斌攸 潘云峰 苏真伟. 基于深度学习的籽棉中异性纤维检测方法[J]. 纺织学报, 2018, 39(06): 131-135.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
京ICP备14006648号　 京公网安备11010502044800号
版权所有 © 2021 《纺织学报》编辑部
地址： 北京市朝阳区延静里中街3号主楼6层《纺织学报》杂志社　邮政编码：100025　
电话：010-65017711,65917740　传真：010-65016539　Email:fangzhixuebao@vip.126.com
本系统由北京玛格泰克科技发展有限公司设计开发