融合注意力与特征金字塔的小尺度目标检测算法

doi:10.19678/j.issn.1000-3428.0066724

摘要/Abstract

摘要：

针对Faster R-CNN算法对于小尺寸目标以及遮挡或截断物体检测能力不足的问题，提出一种融合CBAM注意力机制和特征金字塔结构的改进Faster R-CNN算法。为重点聚焦特征图像局部高效信息，在特征提取网络中融入CBAM机制，减少无效目标的干扰，提升面对遮挡或截断物体的检测能力。引入特征金字塔网络结构，联结高层与底层特征数据，获得高分辨率、强语义数据，从而增强小目标物体的检测效果。为缓解梯度消失现象以及减少超参数规模，使用表达能力较强的倒残差VS-ResNet网络替换VGG16网络，VS-ResNet网络在原有ResNet 50基础上修改了部分层次结构，加入辅助分类器，设计倒残差和组卷积方式，使激活函数信息在高维环境中完整保留，提高检测准确率。采用重置候选框分值计算方法弥补非极大值抑制算法误消除重叠检测框的缺陷。实验结果表明，相比VGG16，VS-ResNet在CIFAR-10数据集上的正确率提高2.97个百分点，该算法在Pascal VOC 2012数据集上的目标检测mAP值为76.2%，比原始Faster R-CNN算法的mAP值提高了13.9个百分点。

关键词: 深度学习, 注意力机制, 特征金字塔, 小目标检测, 截断物体检测

Abstract:

A modified Faster R-CNN algorithm is proposed to address the problem of poor detection ability for small-scale objects and occluded or truncated objects, combining the CBAM mechanism and feature pyramid structure. To focus on the efficient use of local information in feature images, the CBAM mechanism is integrated into the feature extraction network to reduce the interference of invalid targets and improve the detection ability, notwithstanding occluded or truncated objects. This introduces a Feature Pyramid Network(FPN) structure to connect high- and low-level feature data, obtaining high-resolution and strong semantic data, thereby enhancing the detection effect of small objects. To alleviate the phenomenon of gradient vanishing and reduce the scale of hyperparameters, the commonly used VGG16 network is replaced with a strong expressive ability of the inverse residual VS-ResNet network. VS-ResNet modifies some hierarchical structures based on the original ResNet 50, adds auxiliary classifiers, designs inverse residual and group convolution methods, such that the activation function information is fully preserved in high-dimensional environments, and improves detection accuracy. The reset candidate box score calculation method is used to compensate for the defect of the Non-Maximum Suppression(NMS) algorithm in mistakenly eliminating overlapping detection boxes. The experimental results demonstrate that compared to VGG16, VS-ResNet has a 2.97 percentage points improvement in accuracy on the CIFAR-10 dataset. The target detection mAP value of the proposed algorithm on the Pascal VOC 2012 dataset is 76.2%, which is 13.9 percentage points higher than that of the original Faster R-CNN algorithm.

Key words: deep learning, attention mechanism, feature pyramid, small object detection, truncated object detection

圣文顺, 余熊峰, 林佳燕, 陈欣. 融合注意力与特征金字塔的小尺度目标检测算法[J]. 计算机工程, 2024, 50(1): 242-250.

Wenshun SHENG, Xiongfeng YU, Jiayan LIN, Xin CHEN. Small-Scale Object Detection Algorithm Integrating Attention and Feature Pyramids[J]. Computer Engineering, 2024, 50(1): 242-250.

https://www.ecice06.com/CN/Y2024/V50/I1/242

图/表 17

图1 Faster R-CNN算法框架

Fig.1 Framework of Faster R-CNN algorithm

图2 区域建议网络结构

Fig.2 Structure of region proposal network

图3 卷积块结构

Fig.3 Structure of convolution block

图4 辅助分类器示意图

Fig.4 Schematic diagram of auxiliary classifier

图5 CBAM注意力机制示意图

Fig.5 Schematic diagram of CBAM attention mechanism

图6 特征金字塔网络结构

Fig.6 Structure of feature pyramid network

图7 CF-RCNN整体结构

Fig.7 Overall structure of CF-RCNN

图8 VS-ResNet、ResNet 50、VGG16错误率对比

Fig.8 Comparison of error rates among VS-ResNet, ResNet 50, VGG16

图9 截断目标检测结果对比

Fig.9 Comparison of truncated object detection results

图10 算法优化前后检测结果对比

Fig.10 Comparison of detection results before and after algorithm optimization

参考文献 35

1	张珂, 冯晓晗, 郭玉荣, 等. 图像分类的深度卷积神经网络模型综述. 中国图象图形学报, 2021, 26(10): 2305- 2325.
	ZHANG K, FENG X H, GUO Y R, et al. Overview of deep convolutional neural networks for image classification. Journal of Image and Graphics, 2021, 26(10): 2305- 2325.
2	XU P F, LI F, WANG H P. A novel concatenate feature fusion RCNN architecture for sEMG-based hand gesture recognition. PLoS One, 2022, 17, 262810.
3	JIANG L, CHEN J A, TODO H, et al. Application of a fast RCNN based on upper and lower layers in face recognition. Computational Intelligence and Neuroscience, 2021, 2021, 1- 12.
4	DELWICHE S, BAEK I, KIM M S. Does spatial region of interest(ROI)matter in multispectral and hyperspectral imaging of segmented wheat kernels?. Biosystems Engineering, 2021, 212, 106- 114. doi: 10.1016/j.biosystemseng.2021.10.003
5	赵薇, 赵雪妮, 康凯, 等. 基于LDA_SVM的小麦质地检测方法研究. 中国粮油学报, 2023, 38(1): 146- 152.
	ZHAO W, ZHAO X N, KANG K, et al. Research on wheat texture detection method based on LDA_SVM. Chinese Journal of Cereals and Oils, 2023, 38(1): 146- 152.
6	翁昕. 目标检测网络SSD的区域候选框的设置问题研究[D]. 西安: 西安电子科技大学, 2017.
	WENG X. Research on the setting of regional candidate frame of SSD in target detection network[D]. Xi'an: Xidian University, 2017. (in Chinese)
7	SEONG J H, LEE S H, KIM W Y, et al. High-precision RTT-based indoor positioning system using RCDN and RPN. Sensors, 2021, 21(11): 3701. doi: 10.3390/s21113701
8	JI X Y, YAN Q Y, HUANG D, et al. Filtered selective search and evenly distributed convolutional neural networks for casting defects recognition. Journal of Materials Processing Technology, 2021, 292, 117064. doi: 10.1016/j.jmatprotec.2021.117064
9	LUO Q, LIU Z D. Research on face local attribute detection method based on improved SSD network structure. Advances in Multimedia, 2022,(3): 1- 11.
10	CATELANI M, CIANI L, GALAR D, et al. Risk assessment of a wind turbine: a new FMECA-based tool with RPN threshold estimation. IEEE Access, 2020, 8, 20181- 20190. doi: 10.1109/ACCESS.2020.2968812
11	QI J H, ZHANG J D, MENG Q Y, et al. Detection of auxiliary equipment in engine room based on improved SSD[C]//Proceedings of the 3rd International Conference on Modeling, Simulation, Optimization and Algorithm. Sanya, China: [s. n. ], 2022: 1-10.
12	ROCHA D A, FERREIRA F M F, PEIXOTO Z M A. Diabetic retinopathy classification using VGG16 neural network. Research on Biomedical Engineering, 2022, 38(2): 761- 772. doi: 10.1007/s42600-022-00200-8
13	FU H X, SONG G Q, WANG Y C. Improved YOLOv4 marine target detection combined with CBAM. Symmetry, 2021, 13(4): 623. doi: 10.3390/sym13040623
14	LI Y C, ZHOU S L, CHEN H. Attention-based fusion factor in FPN for object detection. Applied Intelligence, 2022, 52(13): 15547- 15556. doi: 10.1007/s10489-022-03220-0
15	WALIA I S, KUMAR D, SHARMA K, et al. An integrated approach for monitoring social distancing and face mask detection using stacked ResNet-50 and YOLOv5. Electronics, 2021, 10(23): 2996. doi: 10.3390/electronics10232996
16	WANG A L, WANG W Y, ZHOU H M, et al. Network intrusion detection algorithm combined with group convolution network and snapshot ensemble. Symmetry, 2021, 13(10): 1814. doi: 10.3390/sym13101814
17	屈景怡, 刘畅. 基于轻量化网络MobileNetV2的航班延误预测模型. 信号处理, 2022, 38(5): 973- 982.
	QU J Y, LIU C. Flight delay prediction model based on lightweight network MobileNetV2. Journal of Signal Processing, 2022, 38(5): 973- 982.
18	LIU M, CAI Z Q, CHEN J S. Adaptive two-layer ReLU neural network: I. best least-squares approximation[EB/OL]. [2022-12-08]. https://arxiv.org/abs/2107.08935.
19	PAZHANI A A J, VASANTHANAYAKI C. Object detection in satellite images by Faster R-CNN incorporated with enhanced ROI pooling(FrRNet-ERoI)framework. Earth Science Informatics, 2022, 15(1): 553- 561. doi: 10.1007/s12145-021-00746-8
20	SZOSTAK D, WŁODARCZYK A, WALKOWIAK K. Machine learning classification and regression approaches for optical network traffic prediction. Electronics, 2021, 10(13): 1578. doi: 10.3390/electronics10131578
21	孟月波, 石德旺, 刘光辉, 等. 多维度卷积融合的密集不规则文本检测. 光学精密工程, 2021, 29(9): 2210- 2221.
	MENG Y B, SHI D W, LIU G H, et al. Dense irregular text detection based on multi-dimensional convolution fusion. Optics and Precision Engineering, 2021, 29(9): 2210- 2221.
22	ZHU X L, HE Z L, ZHAO L, et al. A cascade attention based facial expression recognition network by fusing multiscale spatio-temporal features. Sensors, 2022, 22(4): 1- 10. doi: 10.1109/JSEN.2022.3147718
23	朱伟, 马立新, 张平, 等. 基于GoogLeNet和无人机图像的水稻秧苗形态识别. 华南农业大学学报, 2022, 43(3): 99- 106.
	ZHU W, MA L X, ZHANG P, et al. Morphological recognition of rice seedlings based on GoogLeNet and UAV image. Journal of South China Agricultural University, 2022, 43(3): 99- 106.
24	CHEN Q, ZHANG W Y, ZHU K, et al. A novel trilinear deep residual network with self-adaptive Dropout method for short-term load forecasting. Expert Systems With Applications, 2021, 182, 115272. doi: 10.1016/j.eswa.2021.115272
25	LIAO Z H, FAN N, XU K. Swin Transformer assisted prior attention network for medical image segmentation. Applied Sciences, 2022, 12(9): 4735. doi: 10.3390/app12094735
26	陈勇, 刘曦, 刘焕淋. 基于特征通道和空间联合注意机制的遮挡行人检测方法. 电子与信息学报, 2020, 42(6): 1486- 1493.
	CHEN Y, LIU X, LIU H L. Occluded pedestrian detection based on joint attention mechanism of channel-wise and spatial information. Journal of Electronics & Information Technology, 2020, 42(6): 1486- 1493.
27	LU Y M, HUANG X F, HUANG Y Z, et al. Sigmoid function model for a PFM power electronic converter. IEEE Transactions on Power Electronics, 2020, 35(4): 4233- 4241. doi: 10.1109/TPEL.2019.2935632
28	FENG T, LIU J G, FANG X A, et al. A double-branch surface detection system for armatures in vibration motors with miniature volume based on ResNet-101 and FPN. Sensors, 2020, 20(8): 2360. doi: 10.3390/s20082360
29	QIN A, ZHANG Y. YOLOv3 traffic sign recognition and detection based on FPN improvement. Scientific Journal of Intelligent Research, 2022, 2(11): 1- 10.
30	LLV X Y. CIFAR-10 image classification based on convolutional neural network. Frontiers in Signal Processing, 2020, 4(4): 1- 12.
31	王晓华, 叶振兴, 王文杰, 等. 多级特征融合下的高精度语义分割方法. 西安工程大学学报, 2021, 35(5): 43- 49.
	WANG X H, YE Z X, WANG W J, et al. High precision semantic segmentation based on multi-level feature fusion. Journal of Xi'an Polytechnic University, 2021, 35(5): 43- 49.
32	陈学仕, 苏通, 漆为民. 基于改进Faster RCNN的印刷电路板瑕疵检测算法. 江汉大学学报(自然科学版), 2022, 50(1): 87- 96.
	CHEN X S, SU T, QI W M. Printed circuit board defect detection algorithm based on improved Faster RCNN. Journal of Jianghan University (Natural Science Edition), 2022, 50(1): 87- 96.
33	WU S K, YANG J R, WANG X G, et al. IoU-balanced loss functions for single-stage object detection. Pattern Recognition Letters, 2022, 156, 96- 103. doi: 10.1016/j.patrec.2022.01.021
34	唐欣. 基于深度学习的小样本水稻害虫识别方法研究[D]. 合肥: 安徽大学, 2021.
	TANG X. Research on identification method of small sample rice pests based on deep learning[D]. Hefei: Anhui University, 2021. (in Chinese)
35	JIA D Y, HE Z H, ZHANG C W, et al. Detection of cervical cancer cells in complex situation based on improved YOLOv3 network. Multimedia Tools and Applications, 2022, 81(6): 8939- 8961. doi: 10.1007/s11042-022-11954-9

[1]	王炼红, 林飞鹏, 李潇瑶, 谌桂枝, 周莉. 融入课程知识图谱的KMAKT预测[J]. 计算机工程, 2024, 50(7): 23-31.
[2]	钱来, 赵卫伟. 基于对比学习和注意力机制的文本分类方法[J]. 计算机工程, 2024, 50(7): 104-111.
[3]	刘建敏, 林晖, 汪晓丁. 基于图注意力机制的无地图场景轨迹预测方法[J]. 计算机工程, 2024, 50(7): 144-153.
[4]	牛瑞婷, 严天峰, 高锐, 王映植. 低信噪比下基于深度学习TCNN-MobileNet的调制识别[J]. 计算机工程, 2024, 50(7): 204-215.
[5]	屠乃威, 焦猛, 阎馨. 复杂环境下输电线路鸟巢目标图像检测模型[J]. 计算机工程, 2024, 50(7): 216-226.
[6]	肖慈, 徐杨, 张永丹, 冯明文, 黄易仟. 结合注意力和低光增强的夜间语义分割[J]. 计算机工程, 2024, 50(7): 271-281.
[7]	张诗婧, 莫绪涛, 赵行, 董杨林. 基于球面折反射成像和YOLOv7的内螺纹缺陷检测[J]. 计算机工程, 2024, 50(7): 282-292.
[8]	张锡英, 孙守东, 于海浩, 边继龙. 基于空间传播的多视图三维重建[J]. 计算机工程, 2024, 50(7): 293-302.
[9]	贵向泉, 刘世清, 李立, 秦庆松, 李唐艳. 基于改进YOLOv8的景区行人检测算法[J]. 计算机工程, 2024, 50(7): 342-351.
[10]	徐明亮, 李芳媛, 马浩然, 何飞. 大规模神经记录的峰电位聚类算法(特邀)[J]. 计算机工程, 2024, 50(6): 1-34.
[11]	魏琢艺, 罗迈, 李文兵, 曾远松, 余伟江, 杨跃东. 基于多源域适应的单细胞智能分类[J]. 计算机工程, 2024, 50(6): 48-55.
[12]	李子杰, 周菊香, 韩晓瑜, 甘健侯, 鹿泽光, 王俊. 序列特征与学习过程融合的知识追踪模型[J]. 计算机工程, 2024, 50(6): 77-85.
[13]	程腾腾, 姚春龙, 于晓强, 李旭, 王庆丰. 基于多头注意力机制融合常识知识的共情对话生成[J]. 计算机工程, 2024, 50(6): 94-101.
[14]	更藏措毛, 黄鹤鸣, 杨毅杰. 融合多尺度特征与上下文信息的语音增强方法[J]. 计算机工程, 2024, 50(6): 138-147.
[15]	李永飞, 李铭洋, 常鑫, 曹可欣. 基于可解释性深度学习的物联网水质监测数据异常检测[J]. 计算机工程, 2024, 50(6): 179-187.

选择文件类型/文献管理软件名称

选择包含的内容