计算机科学 ›› 2026, Vol. 53 ›› Issue (4): 299-307.doi: 10.11896/jsjkx.250100105
宋建华1,3,4, 刘淳2, 张龑2,3
SONG Jianhua1,3,4, LIU Chun2, ZHANG Yan2,3
摘要： 伪装目标检测在自然场景分析与安全监控中发挥着重要作用,但伪装目标的复杂性和多样性对检测模型的性能提出了严峻挑战。现有知识蒸馏方法多用于模型压缩,通过对教师网络与学生网络的输出层特征对齐,实现轻量化。然而,现有知识蒸馏方法通常忽略了教师网络中间特征的丰富语义信息。此外,固定学习率策略难以适应教师和学生模型规模差距过大的情况,导致蒸馏过程不稳定。为此,设计了一种基于结构化知识蒸馏的轻量级伪装目标检测模型,利用结构化知识改进传统的软硬标签损失计算,从而显著提升蒸馏效果。同时,将学习率优化问题建模为一个最优化任务,以稳定蒸馏过程中的性能波动。实验结果表明,该方法在COD10K-V3和CAMO伪装目标检测数据集上,Sm分别达到82.9%和81.0%,且训练时间减少至6.5 h。
中图分类号:
| [1]FAN D P,JI G P,SUN G L,et al.Camouflaged object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020:2777-2787. [2]HINTON G,VINYALS O,DEAN J.Distilling the knowledge in a neural network[J].arXiv:1503.02531,2015. [3]CHEN H,WANG Y,XU C,et al.Learning student networksvia feature embedding[J].IEEE Transactions on Neural Networks and Learning Systems,2020,32(1):25-35. [4]FAN D P,JI G P,CHENG M M,et al.Concealed object detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2021,44(10):6024-6042. [5]WANG K,BI H,ZHANG Y,et al.D2C-Net:A Dual-Branch,Dual-Guidance and Cross-Refine Network for Camouflaged Object Detection[J].IEEE Transactions on Industrial Electronics,2021,69(5):5364-5374. [6]LYU 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.2021:11591-11601. [7]REN J,HU X,ZHU L,et al.Deep texture-aware features forcamouflaged object detection[J].IEEE Transactions on Circuits and Systems for Video Technology,2021,33(3):1157-1167. [8]JI G P,FAN D P,CHOU Y C,et al.Deep gradient learning for efficient camouflaged object detection[J].Machine Intelligence Research,2023,20(1):92-108. [9]SUN Y,WANG S,CHEN C,et al.Boundary-guided camou-flaged object detection[J].arXiv:2207.00794,2022. [10]ZHANG J,LYU Y,XIANG M,et al.Depth confidence-aware camouflaged object detection[J].arXiv:2106.05608,2021. [11]ZHONG Y,LI B,TANG L,et al.Detecting camouflaged object in frequency domain[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2022:4504-4513. [12]YANG F,ZHAI Q,LI X,et al.Uncertainty-guided transformer reasoning for camouflaged object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2021:4146-4155. [13]LIU Z,ZHANG Z,TAN Y,et al.Boosting camouflaged object detection with dual-task interactive transformer[C]//2022 26th International Conference on Pattern Recognition(ICPR).IEEE,2022:140-146. [14]YIN B,ZHANG X,FAN D P,et al.Camoformer:Masked separable attention for camouflaged object detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2024,46(12):10362-10374. [15]MENG Z,LI J,ZHAO Y,et al.Conditional teacher-studentlearning[C]//2019 IEEE International Conference on Acoustics,Speech and Signal Processing(ICASSP 2019).IEEE,2019:6445-6449. [16]LI T,LI J,LIU Z,et al.Few sample knowledge distillation for efficient network compression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020:14639-14647. [17]ROMERO A,BALLAS N,KAHOU S E,et al.Fitnets:Hintsfor thin deep nets[J].arXiv:1412.6550,2014. [18]GAO M,SHEN Y,LI Q,et al.An embarrassingly simple approach for knowledge distillation[J].arXiv:1812.01819,2018. [19]ZHOU G,FAN Y,CUI R,et al.Rocket launching:A universal and efficient framework for training well-performing light net[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2018. [20]ZHU X,GONG S.Knowledge distillation by on-the-fly nativeensemble[C]//Advances in Neural Information Processing Systems.2018. [21]SI Y H,CHENG Q,HUANG J C.A dynamic multi-teacher setting method for knowledge distillation[J].Computer Science,2025,52(5):241-247. [22]ZHENG P,GAO D,FAN D P,et al.Bilateral reference for high-resolution dichotomous image segmentation[J].arXiv:2401.03407,2024. [23]ZHAO H W,WU H,MA K,et al.Image classification frame-work based on knowledge distillation[J].Journal of Jilin University(Engineering and Technology Edition),2024,54(8):2307-2312. [24]MIRZADEH S I,FARAJTABAR M,LI A,et al.Improvedknowledge distillation via teacher assistant[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2020:5191-5198. [25]LOSHCHILOV I,HUTTER F.Sgdr:Stochastic gradient de-scent with warm restarts[J].arXiv:1608.03983,2016. [26]FAN D P,CHENG M M,LIU Y,et al.Structure-measure:A new way to evaluate foreground maps[C]//Proceedings of the IEEE International Conference on Computer Vision.2017:4548-4557. [27]ACHANTA R,HEMAMI S,ESTRADA F,et al.Frequency-tuned salient region detection[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2009:1597-1604. [28]FAN D P,GONG C,CAO Y,et al.Enhanced-alignment measure for binary foreground map evaluation[J].arXiv:1805.10421,2018. [29]BORJI A,CHENG M M,JIANG H,et al.Salient object detection:A benchmark[J].IEEE Transactions on Image Proces-sing,2015,24(12):5706-5722. [30]PANG Y,ZHAO X,XIANG T Z,et al.Zoom in and out:A mixed-scale triplet network for camouflaged object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2022:2160-2170. [31]HUANG Z,DAI H,XIANG T Z,et al.Feature shrinkage pyramid for camouflaged object detection with transformers[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2023:5557-5566. [32]JIA Q,YAO S,LIU Y,et al.Segment,magnify and reiterate:Detecting camouflaged objects the hard way[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2022:4713-4722. [33]HE C,LI K,ZHANG Y,et al.Camouflaged object detectionwith feature decomposition and edge reconstruction[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2023:22046-22055. [34]SONG J Y,LUO X Z.A lightweight camouflaged object detection model based on attention mechanism and knowledge distillation[J].Journal of Hubei University(Natural Science Edition),2024,46(5):611-620. |
| [1] | 潘嘉豪, 冯翔, 虞慧群. 基于多任务强化学习的优先级加权软模块化方法:SM-PHT SM-PHT:Robust,Scalable,and Efficient Method for Multi-task Reinforcement Learning 计算机科学, 2026, 53(4): 366-376. https://doi.org/10.11896/jsjkx.250700198 |
| [2] | 吴荞睿, 罗丽, 赵才荣. 基于循环一致性约束的LLM增强型语言模型训练框架 LLM-augmented Training Framework with Cycle-Consistency Constraints 计算机科学, 2026, 53(4): 377-383. https://doi.org/10.11896/jsjkx.250600032 |
| [3] | 赵斌贝, 朱力, 赵红礼, 李雨彤. 计算机视觉在轨道交通中的应用 Computer Vision Applications in Rail Transit Systems 计算机科学, 2026, 53(3): 214-224. https://doi.org/10.11896/jsjkx.250400009 |
| [4] | 唐心亮, 潘晓润, 王建超, 苏鹤. 融合ByteTrack的EAP-YOLOv8无人机Marker点检测与追踪 Integrate ByteTrack’s EAP-YOLOv8 UAV Marker Point Detection and Tracking 计算机科学, 2026, 53(3): 266-276. https://doi.org/10.11896/jsjkx.241100115 |
| [5] | 黄靖, 王腾, 刘健, 胡凯, 彭鑫, 黄亚敏, 文元桥. 多模态水声图像目标视觉检测 Multimodal Visual Detection for Underwater Sonar Target Images 计算机科学, 2026, 53(2): 227-235. https://doi.org/10.11896/jsjkx.241200082 |
| [6] | 范家斌, 王宝会, 陈继轩. 基于文本-图像多模态融合的变电所布局图纸图符检测方法 Method for Symbol Detection in Substation Layout Diagrams Based on Text-Image MultimodalFusion 计算机科学, 2026, 53(1): 206-215. https://doi.org/10.11896/jsjkx.250200090 |
| [7] | 周秉泉, 蒋杰, 陈江民, 詹礼新. EvR-DETR:融合事件与RGB图像的轻量级端到端目标检测 EvR-DETR:Event-RGB Fusion for Lightweight End-to-End Object Detection 计算机科学, 2026, 53(1): 153-162. https://doi.org/10.11896/jsjkx.250300021 |
| [8] | 李芳芳, 孔雨秋, 刘洋, 李朋玥. 基于类别标签引导的协同显著性目标检测方法 Co-salient Object Detection Guided by Category Labels 计算机科学, 2026, 53(1): 163-172. https://doi.org/10.11896/jsjkx.250100071 |
| [9] | 蒋云良, 金森洋, 张雄涛, 刘凯宁, 申情. 多尺度多粒度解耦蒸馏模糊分类器及其在癫痫脑电信号检测中的应用 Multi-scale Multi-granularity Decoupled Distillation Fuzzy Classifier and Its Application inEpileptic EEG Signal Detection 计算机科学, 2025, 52(9): 37-46. https://doi.org/10.11896/jsjkx.250300096 |
| [10] | 邓佳燕, 田时瑞, 刘香丽, 欧阳红巍, 焦韵嘉, 段明星. 基于多阶段行人特征挖掘的轨迹预测方法 Trajectory Prediction Method Based on Multi-stage Pedestrian Feature Mining 计算机科学, 2025, 52(9): 241-248. https://doi.org/10.11896/jsjkx.250700138 |
| [11] | 沈涛, 张秀再, 许岱. 改进RT-DETR的遥感图像小目标检测算法 Improved RT-DETR Algorithm for Small Object Detection in Remote Sensing Images 计算机科学, 2025, 52(8): 214-221. https://doi.org/10.11896/jsjkx.241000019 |
| [12] | 刘乐, 肖蓉, 杨肖. 解耦知识蒸馏在文档级关系抽取中的应用 Application of Decoupled Knowledge Distillation Method in Document-level RelationExtraction 计算机科学, 2025, 52(8): 277-287. https://doi.org/10.11896/jsjkx.240600050 |
| [13] | 刘成壮, 翟素兰, 刘海庆, 王鲲鹏. 基于多模态特征对齐的弱对齐RGBT显著目标检测 Weakly-aligned RGBT Salient Object Detection Based on Multi-modal Feature Alignment 计算机科学, 2025, 52(7): 142-150. https://doi.org/10.11896/jsjkx.240600033 |
| [14] | 徐永伟, 任好盼, 王棚飞. 基于YOLOv8增强的目标检测算法及其应用规范 Object Detection Algorithm Based on YOLOv8 Enhancement and Its Application Norms 计算机科学, 2025, 52(7): 189-200. https://doi.org/10.11896/jsjkx.250100108 |
| [15] | 陈世嘉, 叶剑元, 龚轩, 曾康, 倪鹏程. 基于改进YOLOv5s的飞机起落架安全销检测算法 Aircraft Landing Gear Safety Pin Detection Algorithm Based on Improved YOlOv5s 计算机科学, 2025, 52(6A): 240400189-7. https://doi.org/10.11896/jsjkx.240400189 |
|
||
首页