融合二阶池化注意力的类边界均衡小样本红外目标检测

doi:10.3778/j.issn.1002-8331.2403-0003

摘要/Abstract

摘要： 针对红外目标检测存在复杂背景干扰下已有的小样本目标检测模型无法充分挖掘支持信息，导致检测性能下降的问题，提出了融合二阶池化注意力的类边界均衡小样本目标检测算法。类边界均衡算法通过类边界对抗最小-最大正则化来实现新类之间边界平衡，解决了小样本目标检测任务中由于基类和新类特征原型分布混乱造成的检测性能下降的问题。但是由于复杂背景干扰，类边界均衡算法直接用于红外目标检测无法充分利用支持集图像的有效信息。提出二阶池化注意力机制来抑制背景干扰，增强对支持图像有效信息的学习，进而加强利用支持信息对查询信息调节的功能。该机制沿通道维度计算输入特征图各个通道之间的协方差，来获取各个通道之间的统计依赖性，进而捕获重要通道的高阶特征信息。同时沿通道维度计算输入特征图各个通道的标准差，并将两个通道的协方差除以两个通道的标准差以减弱噪声对协方差估计的影响，增强协方差计算的准确性。在类边界均衡算法权重模块中融入二阶池化注意力机制，来引导检测算法将特征学习聚焦在目标及其邻域，并抑制复杂背景的干扰。实验结果表明，相对于经典算法，提出的小样本目标检测算法在10-shot任务上获得了最佳性能，在自制的红外目标数据集的新类别上的mAP达到了56.4%。

关键词: 红外目标检测, 小样本, 类边界均衡, 二阶池化注意力, 协方差

Abstract: To address the issue of decreased detection performance in infrared object detection due to the inability of existing few-shot object detection models to fully exploit support information in the presence of complex background interference, a novel few-shot object detection algorithm is proposed that integrates second-order pooling attention with class margin equilibrium. The algorithm of class margin equilibrium achieves boundary balance between new classes by adversarial minimum-maximum regularization on class margin, addressing the performance degradation in few-shot object detection tasks caused by the confusion of feature prototype distributions between base classes and novel classes. However, due to complex background interference, directly applying the algorithm of class margin equilibrium to infrared object detection cannot fully utilize the effective information from the support set images. The second-order pooling attention mechanism is proposed to suppress background interference, enhance the learning of effective information from support set images, and thereby strengthen the function of utilizing support information to adjust query information. This mechanism calculates the covariance between different channels of the input feature map along the channel dimension to capture the statistical dependencies among channels, and thereby capture the high-order feature information of important channels. At the same time, the standard deviation of each channel of the input feature map is computed along the channel dimension. The covariance between two channels is then divided by the product of their respective standard deviations to reduce the impact of noise on the covariance estimation and enhance the accuracy of covariance calculation. Incorporating the second-order pooling attention mechanism into the weighting module of class margin equilibrium guides the detection algorithm to focus feature learning on object and its surroundings, suppressing interference from complex backgrounds. The experimental results indicate that compared to classical algorithms, the proposed few-shot object detection algorithm achieves the best performance in the 10-shot task, with an mAP of 56.4% on novel classes in homemade infrared object dataset.

Key words: infrared object detection, few-shot, class margin equilibrium, second-order pooling attention, covariance

司起峰, 刘刚, 徐红鹏, 陈会祥. 融合二阶池化注意力的类边界均衡小样本红外目标检测[J]. 计算机工程与应用, 2025, 61(10): 279-287.

SI Qifeng, LIU Gang, XU Hongpeng, CHEN Huixiang. Few-Shot Infrared Object Detection with Class Margin Equilibrium Based on Second-Order Pooling Attention[J]. Computer Engineering and Applications, 2025, 61(10): 279-287.

参考文献

[1] 张毅, 张焱, 张宇, 等. 基于多级特征增强融合的红外飞机目标检测方法[J]. 航空学报, 2023, 44(22): 48-62.
ZHANG Y, ZHANG Y, ZHANG Y, et al. Infrared aircraft target detection method based on multi-level feature enhancement fusion[J]. Acta Aeronautica et Astronautica Sinica, 2023, 44(22): 48-62.
[2] WANG X, LU R, BI H, et al. An infrared small target detection method based on attention mechanism[J]. Sensors (Basel, Switzerland), 2023, 23(20): 8608.
[3] 刘春磊, 陈天恩, 王聪, 等. 小样本目标检测研究综述[J]. 计算机科学与探索, 2023, 17(1): 53-73.
LIU C L, CHEN T E, WANG C, et al. Survey of few-shot object detection[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(1): 53-73.
[4] 张振伟, 郝建国, 黄健, 等. 小样本图像目标检测研究综述[J]. 计算机工程与应用, 2022, 58(5): 1-11.
ZHANG Z W, HAO J G, HUANG J, et al. Review of few-shot object detection[J]. Computer Engineering and Applications, 2022, 58(5): 1-11.
[5] WU J P, ZENG J J, ZHOU Y B, et al. Few-shot electrical equipment image recognition method based on an improved two-stage fine-tuning approach[J]. The Journal of Engineering, 2023(9): e12313.
[6] FAN Q, ZHUO W, TANG C K, et al. Few-shot object detection with attention-RPN and multi-relation detector[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 4013- 4022.
[7] 丁政伟, 白鹤翔, 胡深. 多尺度深层特征加强的CME小样本目标检测模型[J]. 计算机工程与应用, 2024, 60(6): 222-229.
DING Z W, BAI H X, HU S. CME-based few-shot detection model with enhanced multiscale deep features[J]. Computer Engineering and Applications, 2024, 60(6): 222-229.
[8] TIAN Y J, ZHAO X X, HUANG W. Meta-learning approaches for learning-to-learn in deep learning: a survey[J]. Neurocomputing, 2022, 494: 203-223.
[9] YAN X P, CHEN Z L, XU A N, et al. Meta R-CNN: towards general solver for instance-level low-shot learning[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 9576-9585.
[10] AMJOUD A B, AMROUCH M. Object detection using deep learning, CNNs and vision transformers: a review[J]. IEEE Access, 2023, 11: 35479-35516.
[11] 华杰, 刘学亮, 赵烨. 基于特征融合的小样本目标检测[J]. 计算机科学, 2023, 50(2): 209-213.
HUA J, LIU X L, ZHAO Y. Few-shot object detection based on feature fusion[J]. Computer Science, 2023, 50(2): 209-213.
[12] 李丽芬, 范新烨. 元学习与多尺度特征融合的小样本目标检测[J]. 小型微型计算机系统, 2023, 44(12): 2804-2811.
LI L F, FAN X Y. Few-shot object detection with meta-learning and multi-scale feature fusion[J]. Journal of Chinese Computer Systems, 2023, 44(12): 2804-2811.
[13] KANG B Y, LIU Z, WANG X, et al. Few-shot object detection via feature reweighting[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 8419-8428.
[14] 李红光, 王玉峰, 杨丽春. 基于元学习的小样本遥感图像目标检测[J]. 北京航空航天大学学报, 2024, 50(8): 2503-2513.
LI H G, WANG Y F, YANG L C. Meta-learning-based few-shot object detection for remote sensing images[J]. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(8): 2503-2513.
[15] 金璐. 少样本条件下的红外空中目标识别与检测技术研究[D]. 上海: 中国科学院大学 (中国科学院上海技术物理研究所), 2020.
JIN L. Research on key technologies of infrared aerial target recognition and detection with few shots[D]. Shanghai: University of Chinese Academy of Sciences (Shanghai Institute of Technical Physics, Chinese Academy of Sciences), 2020.
[16] 何湘杰, 宋晓宁. YOLOv4-Tiny的改进轻量级目标检测算法[J]. 计算机科学与探索, 2024, 18(1): 138-150.
HE X J, SONG X N. Improved YOLOv4-tiny lightweight target detection algorithm[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(1): 138-150.
[17] LI B H, YANG B Y, LIU C, et al. Beyond max-margin: class margin equilibrium for few-shot object detection[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 7359-7368.
[18] HU J, SHEN L, ALBANIE S, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023.
[19] WANG Q, WU B, ZHU P, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 11534-11542.
[20] WOO S, PARK J, LEE J F, et al. CBAM: convolutional block attention module[C]//Proceedings of the 15th European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[21] DONG Z, WANG Q L, ZHU P F. Multi-head second-order pooling for graph transformer networks[J]. Pattern Recognition Letters, 2023, 167: 53-59.
[22] LI E Z, SAMAT A, ZHANG C, et al. First and second-order information fusion networks for remote sensing scene classification[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 8014405.
[23] WANG Z W, DONG W, ZHANG B B, et al. GSoANet: group second-order aggregation network for video action recognition[J]. Neural Processing Letters, 2023, 55(6): 7493-7509.
[24] DAI T, CAI J R, ZHANG Y B, et al. Second-order attention network for single image super-resolution[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 11057-11066.
[25] GAO Z L, XIE J T, WANG Q L, et al. Global second-order pooling convolutional networks[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 3019-3028.
[26] SUN B, LI B H, CAI S C, et al. FSCE: few-shot object detection via contrastive proposal encoding[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 7348-7358.