一种基于ACE-Yolov5的小目标检测模型

A Small Object Detection Model Based on ACE-Yolov5

摘要

小目标的分辨率较低,在复杂的背景下有时难以定位和识别,从而导致漏检和误检的问题,因此小目标检测的精度还有很大的提升空间.为了提升小目标检测精度,一种基于ACE-Yolov5的小目标检测模型被提出,在Yolov5的基础上引入了小尺度特征地图的专用检测层,融合了上下文使用模块和改进的ECA注意力机制.首先,上下文使用模块主要由多分支的空洞卷积层组成,利用不同采样率的空洞卷积多个比例捕捉特征图的上下文信息.第二,将ECA进行改进,实现对输入特征图进行通道特征加强,并大大降低参数量.实验结果表明,在Tsinghua-Tencent 100K数据集和Traffic Camera Object Detection数据集上,此ACE-Yolov5模型检测性能优于其他模型,与Yolov5相比,mAP50分别提升了5.1%和2.2%.

Abstract

Due to the attributes of the small objects, there is still much room for improvement in the positioning accuracy of small object detection. In order to improve the small object detection accuracy, ACE-Yolov5 has been proposed based on the Yolov5. We incorporate the Contextual Use Module (CUM) and improved Efficient Channel Attention (ECA) based on the addition of a fourth detection layer in the baseline. First, the CUM mainly consists of a multi-branch null convolution layer, which captures the contextual information of the feature map using multiple scales of null convolution with different sampling rates. Second, the ECA is combined with the C3 module so that the ECA has the C3 attribute, which enables channel feature enhancement of the input feature map and greatly reduces the number of parameters. The experimental results show that, on the Tsinghua Tencent 100K dataset and the Traffic Camera Object Detection dataset, the ACE-Yolov5 model proposed in this paper outperforms other models in terms of detection performance. Compared with Yolov5, the mAP50 improved by 5.1% and 2.2%, respectively.

关键词

小目标检测 / Yolov5 / 上下文 / 注意力机制

Key words

small object detection / Yolov5 / context / attentional mechanisms

参考文献

[1] Ghenescu V, Mihaescu R E, Carata S V, et al. Face detection and recognition based on general purpose DNN object detector[C]// 2018 International Symposium on Electronics and Telecommunications (ISETC). IEEE, 2018: 1-4.

[2] Dou J, Fang J, Li T, et al. Boosting cnn-based pedestrian detection via 3d lidar fusion in autonomous driving[C]// Image and Graphics: 9th International Conference, ICIG 2017, Shanghai, China, September 13-15, 2017, Revised Selected Papers, Part II 9. Springer International Publishing, 2017: 3-13.

[3] Esmaeilzadeh S, Yang Y, Adeli E. End-to-end Parkinson disease diagnosis using brain MR-images by 3D-CNN[J]. arXiv preprint arXiv:1806.05233, 2018.

[4] Zhang H, Luo C, Wang Q, et al. A novel infrared video surveillance system using deep learning based techniques[J]. Multimedia Tools and Applications, 2018, 77: 26657-26676.

[5] Chen C, Liu M Y, Tuzel O, et al. R-CNN for small object detection[C]// Computer Vision-ACCV 2016: 13th Asian Conference on Computer Vision, Taipei, Taiwan, November 20-24, 2016, Revised Selected Papers, Part V 13. Springer International Publishing, 2017: 214-230.

[6] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.

[7] Girshick R. Fast r-cnn[C]// Proceedings of the IEEE international conference on computer vision, 2015: 1440-1448.

[8] Ren S, He K, Girshick R, et al. Faster r-cnn: Towards real-time object detection with region proposal networks[J]. Advances in Neural Information Processing Systems, 2015, 28: 1137-1149.

[9] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]// Computer Vision-ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part I 14. Springer International Publishing, 2016: 21-37.

[10] Redmon J, Farhadi A. Yolo9000: better, faster, stronger[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 7263-7271.

[11] Xiaolin F, Fan H, Ming Y, et al. Small object detection in remote sensing images based on super-resolution[J]. Pattern Recognition Letters, 2022, 153: 107-112.

[12] Zhu X, Lyu S, Wang X, et al. TPH-Yolov5: Improved Yolov5 based on transformer prediction head for object detection on drone-captured scenarios[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 2778-2788.

[13] Benjumea A, Teeti I, Cuzzolin F, et al. Yolo-Z: Improving small object detection in Yolov5 for autonomous vehicles[J]. arXiv preprint arXiv:2112.11798, 2021.

[14] Snyder C, Do M. Data for STREETS: A Novel Camera Network Dataset for Traffic Flow[DS]. Urbana-Champaign: University of Illinois at Urbana-Champaign, 2019. https://doi.org/10.13012/B2IDB-3671567_V1.

[15] Jung H K, Choi G S. Improved yolov5: Efficient object detection using drone images under various conditions[J]. Applied Sciences, 2022, 12(14): 7255.

[16] Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018.

[17] Xiao J S, Zhao T, Yao Y T, et al. Context augmentation and feature refinement network for tiny object detection[J]. Journal of Computer Research and Development, 2023, 60(2): 465-474.

[18] Wang Q, Wu B, Zhu P, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11534-11542.

[19] Zhu L, Wang X, Ke Z, et al. Biformer: Vision transformer with bi-level routing attention[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 10323-10333.

[20] Yang L, Zhang R Y, Li L, et al. Simam: A simple, parameter-free attention module for convolutional neural networks[C]// International Conference on Machine Learning. PMLR, 2021: 11863-11874.