Research on Infrared Small Target Detection Algorithm Based on Improved YOLOv8
-
摘要:
针对复杂背景下红外小目标识别错误率高，模型回归损失较大的问题，提出一种改进的算法YOLOv8_SG（Small goals）。通过构建小目标检测层、引入SA注意力机制与WIoU_v3损失函数，使算法能够融合更深层特征，具有更大的感受野，并且降低了训练样本标注质量不均衡的影响，提高了预测框的位置精度，增强了对小目标的检测能力。实验结果表明，改进后的算法mAP由0.8514提升到0.8997，Box_loss效果较改进前提升了34.9%，该算法在小目标检测上具有更高的特征提取能力和更高的检测精度。
Abstract:Aiming at the problem of the high error rate of infrared small-target recognition and the large loss of model regression in complex backgrounds, an improved YOLOv8_SG (Small goals) algorithm was proposed by adding a small target detection layer and introducing the SA attention mechanism and WIoU_v3 loss function, which can fuse deeper features and have a larger receptive field. Moreover, the influence of the uneven labeling quality of the training samples was reduced, the position accuracy of the prediction box was improved, and the ability to detect small targets was enhanced. The experimental results show that the mAP of the improved algorithm increased from 0.8514 to 0.8997, and the overall loss effect of Box_loss increased by 34.9%. The proposed algorithm has a higher feature extraction ability and higher detection accuracy for small-target detection.
-
Keywords:
- target detection /
- attention mechanism /
- receptive field /
- loss function
-
-
表 1 优化网络结构实验结果
Table 1 Results of experiment with optimized network structure
Structural models YOLOv8_n YOLOv8_n_Small Enter the size/pixels 640×640 640×640 mAP@ 0.5 0.8514 0.8832 FPS 171.9 141.5 Parameter quantity/M 3.01 3.05 FLOPs/G 8.1 12.8 
下载: 导出CSV
表 2 注意力机制消融实验结果
Table 2 Results of attention mechanism ablation experiments
Structural models mAP@ 0.5 Precision Recall v8_n 0.8514 0.8499 0.7457 v8_n_S 0.8832 0.8971 0.7897 v8_n_S_MHSA 0.8292 0.8356 0.7383 v8_n_S_A2 0.8515 0.9092 0.7345 v8_n_S_SE 0.8795 0.8507 0.8222 v8_n_S_CA 0.8893 0.9293 0.7750 v8_n_S_CoTA 0.8886 0.8929 0.8190 v8_n_S_SA 0.8928 0.9045 0.8142
下载: 导出CSV
表 3 SA模块位置消融实验结果
Table 3 Results of SA module position ablation experiment
Backbone network mAP@ 0.5 Precision Recall v8_n 0.8514 0.8499 0.7457 + Floor 5 0.8521 0.8466 0.7615 + Floor 7 0.8546 0.8559 0.7922 + Floor 9 0.8928 0.9045 0.8142 + Floor 10 0.8933 0.8990 0.8156
下载: 导出CSV
表 4 IoU消融实验结果
Table 4 Results of IoU ablation experiment
Structural models mAP@0.5 Box_loss v8_n 0.8514 1.425 v8_n_S 0.8832 1.339 v8_n_S_SA 0.8928 1.350 v8_n_S_SA_FG 0.8878 1.191 v8_n_S_SA_FD 0.8846 1.182 v8_n_S_SA_FC 0.8917 1.163 v8_n_S_SA_FE 0.8799 1.271 v8_n_S_SA_FS 0.8726 2.061 v8_n_S_SA_W(v1) 0.8797 1.368 v8_n_S_SA_W(v2) 0.8931 0.861 YOLOv8_SG 0.8997 0.928
下载: 导出CSV
表 5 不同算法的实验结果
Table 5 Experimental results of different algorithms
Structural models Enter the size/pixels mAP@ 0.5 FPS Parameter quantity/M FLOPs/G YOLOv5_n 640×640 0.5156 201.4 1.83 4.36 YOLOv5_s 640×640 0.8110 172.3 7.06 16.50 YOLOv5_m 640×640 0.7260 166.5 21.05 50.60 YOLOv6_n 640×640 0.8437 173.9 4.65 11.39 YOLOv6_s 640×640 0.8760 116.3 18.54 45.28 YOLOv8_n 640×640 0.8514 171.9 3.01 8.10 YOLOv8_s 640×640 0.8461 141.4 11.13 28.60 YOLOv8_m 640×640 0.7874 97.5 25.80 79.10 YOLOv8_l 640×640 0.8026 54.8 43.63 165.40 YOLOv8_x 640×640 0.7617 33.7 68.15 258.10 YOLOv8_SG 640×640 0.8997 140.8 10.90 37.90
下载: 导出CSV
-
[1] 康鑫英, 张德育, 王君. 压缩SSD低空小目标检测算法研究[J]. 沈阳理工大学学报, 2022, 41(6): 1-5. DOI: 10.3969/j.issn.1003-1251.2022.06.001 KANG X Y, ZHANG D Y, WANG J. Research on detection algorithm of small target at low altitude based on compressed SSD[J]. Journal of Shenyang Ligong University, 2022, 41(6): 1-5. DOI: 10.3969/j.issn.1003-1251.2022.06.001
[2] PENG H, ZHANG Y F, YANG S, et al. Battlefield image situational awareness application based on deep learning[J]. IEEE Intelligent Systems, 2020, 35(1): 36-43. DOI: 10.1109/MIS.2019.2953685
[3] CHEN C Y, LIU M Y, TUZEL O, et al. R-CNN for small object detection[C]//Proceeding of Asian Conference on Computer Vision, 2016, 10115: 214-230.
[4] PANG J M, CHEN K, SHI J P, et al. Libra R-CNN: towards balanced learning for object detection[J]. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019: 821-830.
[5] 王君, 张德育, 康鑫英. 改进Faster-RCNN的低空小型无人机检测方法[J]. 沈阳理工大学学报, 2021, 40(4): 23-28. DOI: 10.3969/j.issn.1003-1251.2021.04.005 WANG J, ZHANG D Y, KANG X Y. Improved detection method of low altitude small UAV by faster-RCNN[J]. Journal of Shenyang Ligong University, 2021, 40(4): 23-28. DOI: 10.3969/j.issn.1003-1251.2021.04.005
[6] 李嘉新, 侯进, 盛博莹, 等. 基于改进YOLOv5的遥感小目标检测网络[J]. 计算机工程, 2023, 9(2): 1-11. LI J X, HOU J, SHENG B Y, et al. Remote sensing small object detection network based on improved YOLOv5[J]. Computer Engineering, 2023, 9(2): 1-11.
[7] 谢椿辉, 吴金明, 徐怀宇. 改进YOLOv5的无人机影像小目标检测算法[J]. 计算机工程与应用, 2023, 59(9): 198-206. XIE C H, WU J M, XU H Y. Small object detection algorithm based on improved YOLOv5 in UAV image[J]. Computer Engineering and Applications, 2023, 59(9): 198-206.
[8] 倪浩鹏. 基于深度学习的红外小目标检测关键算法研究[D]. 南京: 南京邮电大学, 2022. NI H P. Research on Critical Algorithms of Infrared Small Target Detection Based on Deep Learning[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2022.
[9] 张艳珠, 包慧哲, 于长海, 等. 基于无人机图像地面建筑物目标检测算法研究[J]. 沈阳理工大学学报, 2023, 42(4): 1-6, 14. DOI: 10.3969/j.issn.1003-1251.2023.04.001 ZHANG Y Z, BAO H Z, YU C H, et al. Research on the algorithm for detecting objects on the ground based on images by UAV[J]. Journal of Shenyang Ligong University, 2023, 42(4): 1-6, 14. DOI: 10.3969/j.issn.1003-1251.2023.04.001
[10] 商俊燕. 基于深度学习的遥感图像微小目标检测方法研究[J]. 计算机测量与控制, 2022, 30(10): 57-62. SHANG J Y. Research on remote sensing image micro target detection method based on deep learning[J]. Computer Measurement & Control, 2022, 30(10): 57-62.
[11] 贺顺, 谢永妮, 杨志伟, 等. 基于IHBF的增强局部对比度红外小目标检测方法[J]. 红外技术, 2022, 44(11): 1132-1138. http://hwjs.nvir.cn/article/id/0f2609dc-79df-467e-ac1d-4d5f888850d1 HE S, XIE Y N, YANG Z W, et al. IHBF-based enhanced local contrast measure method for infrared small target detection[J]. Infrared Technology, 2022, 44(11): 1132-1138. http://hwjs.nvir.cn/article/id/0f2609dc-79df-467e-ac1d-4d5f888850d1
[12] 张丹, 王军锋. 多尺度加权图像块局部对比度的红外小目标检测[J]. 计算机工程与设计, 2023, 44(5): 1459-1470. ZHANG D, WANG J F. Infrared small target detection based on multi-scale weighted local contrast[J]. Computer Engineering and Design, 2023, 44(5): 1459-1470.
[13] 蒲家鹏, 王雪梅, 高宏伟. 基于改进YOLOv4的航拍图像目标检测方法研究[J]. 沈阳理工大学学报, 2023, 42(3): 46-53. PU J P, WANG X M, GAO H W. Research of aerial image object detection method based on improved YOLOv4[J]. Journal of Shenyang Ligong University, 2023, 42(3): 46-53.
[14] ZHANG Q L, YANG Y B. SA-Net: shuffle attention for deep convolutional neural networks[J]. IEEE International Conference on Acoustics, 2021, 30(1): 2235-2239.
[15] 刘浩翰, 孙铖, 贺怀清, 等. 基于改进YOLOv3的金属表面缺陷检测[J]. 计算机工程与科学, 2023, 45(7): 1226-1235. DOI: 10.3969/j.issn.1007-130X.2023.07.010 LIU H H, SUN C, HE H Q, et al. Metal surface defect detection based on improved YOLOv3[J]. Computer Engineering & Science, 2023, 45(7): 1226-1235. DOI: 10.3969/j.issn.1007-130X.2023.07.010
[16] HAN Z Q, ZHANG C, FENG H Z, et al. PFFNET: a fast progressive feature fusion network for detecting drones in infrared images[J]. Drones, 2023, 7(7): 424. DOI: 10.3390/drones7070424
[17] SHUANG J D, ZHANG B F, ZHANG B, et al. An improved bounding box regression loss function based on CIOU loss for multi-scale object detection[C]// IEEE 2nd International Conference on Pattern Recognition and Machine Learning (PRML), 2021: 92-98.
[18] 刘雄彪, 杨贤昭, 陈洋, 等. 基于CIoU改进边界框损失函数的目标检测方法[J]. 液晶与显示, 2023, 38(5): 656-665. LIU X B, YANG X Z, CHEN Y, et al. Object detection method based on CIoU improved bounding box loss function[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(5): 656-665.
[19] 陈益方, 张上, 冉秀康, 等. 基于改进YOLOv8的SAR图像飞机目标检测算法[J]. 电讯技术, 2023, 9(2): 1-8. CHEN Y F, ZHANG S, RAN X K, et al. Aircraft target detection algorithm based on improved YOLOv8 in SAR image[J]. Telecommunication Engineering, 2023, 9(2): 1-8.
[20] 许晓阳, 高重阳. 改进YOLOv7-tiny的轻量级红外车辆目标检测算法[J]. 计算机工程与应用, 2023, 9(2): 1-13. XU X Y, GAO C Y. Improved YOLOv7-Tiny lightweight infrared vehicle target detection algorithm[J]. Computer Engineering and Applications, 2023, 9(2): 1-13.
[21] 孙晓亮, 郭良超, 张文龙, 等. 复杂背景下红外弱小运动目标检测半仿真数据集[EB/OL]. [2023-08-15]. 2021. https://doi.org/10.11922/sciencedb.j00001.00231. SUN X L, GUO L C, ZHANG W L, et al. A dataset for small infrared moving target detection under clutter background[EB/OL]. [2023-08-15]. 2021. https://doi.org/10.11922/sciencedb.j00001.00231.
[22] 马学森, 马吉, 蒋功辉, 等. 基于注意力机制和多尺度特征融合的绝缘子缺陷检测方法[J]. 南京大学学报: 自然科学, 2022, 58(6): 1020-1029. MA X S, MA J, JIANG G H, et al. Insulator defect detection method based on attention mechanism and multi-scale feature fusion[J]. Journal of Nanjing University: Natural Science, 2022, 58(6): 1020-1029.
计量
- 文章访问数: 114
- HTML全文浏览量: 12
- PDF下载量: 44
