改进YOLOv5s算法的高光谱遥感图像目标检测

doi:10.13229/j.cnki.jdxbgxb.20240459

摘要/Abstract

摘要：

针对高光谱图像的光谱分辨率非常高，且包含的地物种类波段较多，使目标与背景之间的光谱差异非常微小，容易造成光谱混淆，使目标检测的准确度较低的问题，提出基于改进YOLOv5s算法的图像目标检测方法。建立特征金字塔并实行多尺度加权，利用特征金字塔中不同层间的权重，对特征加权融合，并将其引入注意力机制中，输出空间注意力机制光谱特征，将该特征值作为对比参照，对通道重新加权分配，获取通道注意力机制输出的光谱特征，将两个光谱特征维度相乘，得到校准后的高光谱图像特征，将其作为改进YOLOv5s算法的输入，有效区分图像中的微小光谱特征差异，避免光谱混淆，根据中心值计算检测框与真实框重叠区域，完成目标检测，保证检测精准度。实验证明：本文方法对高光谱遥感图像中的地物检测精准度较高，在检测1 057 p像素大小的图像时，帧率高达60 fps，综合性能表现优异。

关键词: 改进YOLOv5s算法, 高光谱遥感图像, 空间注意力, 目标检测

Abstract:

The spectral resolution of hyperspectral images is very high， and there are many bands of ground objects， so the spectral difference between the target and the background is very small， which is easy to cause spectral confusion， and the accuracy of target detection is low. Therefore， an image object detection method based on improved YOLOv5s algorithm is proposed. A feature pyramid is established and multi-scale weighting is implemented. The weights between different layers in the feature pyramid are used to weight and fuse the features and introduce them into the attention mechanism. The spectral features of the spatial attention mechanism are output， and the feature value is used as a comparison reference. The hyperspectral image features obtained after calibration are used as the input of the improved YOLOv5s algorithm to effectively distinguish the tiny spectral feature differences in the image， avoid spectral confusion， calculate the overlap area between the detection frame and the real frame according to the central value， complete the target detection， and ensure the detection accuracy. Experiments show that the proposed method has a high accuracy for detecting ground objects in hyperspectral remote sensing images. When detecting 1 057 p pixel images， the frame rate is as high as 60fps， and the comprehensive performance is excellent.

Key words: improving YOLOv5s algorithm, hyperspectral remote sensing images, spatial attention, object detection

中图分类号:

TP123

田丽,贾煜辉. 改进YOLOv5s算法的高光谱遥感图像目标检测[J]. 吉林大学学报(工学版), 2025, 55(5): 1742-1748.

Li TIAN,Yu-hui JIA. Improved YOLOv5s algorithm for target detection in hyperspectral remote sensing images[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(5): 1742-1748.

图/表 4

图1

图2

图3

表1

参考文献 20

[1]	汪西莉, 梁敏, 刘涛. 特征增强的单阶段遥感图像目标检测模型[J]. 西安电子科技大学学报, 2022, 49(3): 160-170.
	Wang Xi-li, Liang Min, Liu Tao. Feature enhanced single-stage remote sensing image object detection model[J].Journal of Xidian University,2022,49(3):160-170.
[2]	单慧琳, 王硕洋, 童俊毅, 等. 增强小目标特征的多尺度光学遥感图像目标检测[J]. 光学学报, 2024, 44(6): 0628006.
	Shan Hui-lin, Wang Shuo-yang, Tong Jun-yi, et al.Object detection in multi-scale optical remote sensing images to enhance the characteristics of small targets[J].Acta Optica Sinica, 2024, 44(6): 0628006.
[3]	王艳辉, 张福泉, 邹静, 等. 基于改进卷积神经网络的遥感图像目标检测方法[J]. 南京理工大学学报, 2023, 47(3): 330-336.
	Wang Yan-hui, Zhang Fu-quan, Zou Jing, et al. Method of remote sensing image target detection based on improved convolution neural network[J]. Journal of Nanjing University of Science and Technology,2023,47(3):330-336.
[4]	Liang Z, Zhu B, Zhu Y X. High resolution representation-based Siamese network for remote sensing image change detection[J]. IET Image Processing,2022,16(9):2506-2517.
[5]	钱晓亮, 曾银凤, 林生, 等. 融合自适应窗口显著性检测和改进超像素分割的高光谱异常检测[J]. 遥感学报, 2023, 27(12): 2748-2761.
	Qian Xiao-liang, Zeng Yin-feng, Lin Sheng, et al. Hyperspectral anomaly detection via combining adaptive window saliencydetection and improved superpixel segmentation[J]. Journal of Remote Sensing,2023,27(12):2748-2761.
[6]	刘洋, 张国军. 基于NSCT及RT的遥感图像阴影消除方法[J]. 计算机仿真, 2022,39(5): 465-469.
	Liu Yang, Zhang Guo-jun. Remote sensing image shadow elimination method based on nsct and retinex theory(RT)[J]. ComputerSimulation,2022,39(5):465-469.
[7]	Hao F, Ma Z F, Tian H P, et al. Semi-supervised label propagation for multi-source remote sensing image change detection[J].Computers & Geosciences, 2023,170(1):1052491.
[8]	姚婷婷, 李鹏飞, 高源, 等. 基于感知增强无锚框网络的遥感图像目标检测[J]. 中国科技论文, 2023, 18(11): 1178-1185.
	Yao Ting-ting, Li Peng-fei, Gao Yuan, et al. Object detection based on perception enhanced anchor-free network for remote sensing images[J]. China Sciencepaper,2023,18(11):1178-1185.
[9]	Yang N, Chang K P, Dong S Z, et al. Rapid image detection and recognition of rice false smut based on mobile smart devices with anti-light features from cloud database[J].Biosystems Engineering,2022,218:229-244.
[10]	吴琪, 樊彦国, 樊博文, 等.基于图正则化低秩协同表示的高光谱异常检测[J]. 激光与光电子学进展,2022,59(12):467-475.
	Wu Qi, Fan Yan-guo, Fan Bo-wen, et al. Graph regularized low-rank and collaborative representation for hyperspectral anomaly detection[J]. Laser & Optoelectronics Progress,2022,59(12):467-475.
[11]	熊娟, 张孙杰, 阚亚亚, 等.基于CAFPN和细化双头解耦的遥感图像目标检测[J]. 应用科学学报, 2023, 41(6): 989-1003.
	Xiong Juan, Zhang Sun-jie, Kan Ya-ya, et al. Remote sensing image object detection based on cafpn and refinement double-head decoupling[J]. Journal of Applied Sciences,2023,41(6):989-1003.
[12]	成倩, 李佳, 杜娟. 基于YOLOv5的光学遥感图像舰船目标检测算法[J]. 系统工程与电子技术, 2023, 45(5): 1270-1276.
	Cheng Qian, Li Jia, Du Juan. Ship target detection algorithm of optical remote sensing image based on YOLOv5[J]. Systems Engineering and Electronics,2023, 45(5): 1270-1276.
[13]	Mahanti N K, Pandiselvam R, Kothakota A, et al. Emerging non-destructive imaging techniques for fruit damage detection: image processing and analysis[J].Trends in Food Science & Technology, 2022, 120: 418-438.
[14]	李强, 武文波, 何明一. 基于MPSoC的遥感图像目标检测算法硬件加速研究[J]. 航天返回与遥感, 2022, 43(1): 58-68.
	Li Qiang, Wu Wen-bo, He Ming-yi. Accelerator of remote sensing image object detection based on MPSoC[J]. Spacecraft Recovery & Remote Sensing,2022,43(1):58-68.
[15]	Wang P K, Wu L, Qi J X, et al. Unmanned aerial vehicles object detection based on image haze removal under sea fog conditions[J]. IET Image Pcessing,2022,16(10):2709-2721.
[16]	Camacho I C, Wang K. Convolutional neural network initialization approaches for image manipulation detection[J]. Digital Signal Processing, 2022, 122: 103376.
[17]	Geet L, Chitta R J, Chen J L, et al. Convolutional neural network-assisted adaptive sampling for sparse feature detection in image and video data[J]. IEEE Intelligent Systems, 2023, 38(1): 45-57.
[18]	顾勇翔, 蓝鑫, 伏博毅, 等. 基于几何适应与全局感知的遥感图像目标检测算法[J]. 计算机应用, 2023, 43(3): 916-922.
	Gu Yong-xiang, Lan Xin, Fu Bo-yi, et al. Object detection algorithm for remote sensing images based on geometric adaptation and global perception[J]. Journal of Computer Applications, 2023, 43(3):916-922.
[19]	Du X J, Wu H L. Feature-aware aggregation network for remote sensing image cloud detection[J].International Journal of Remote Sensing, 2023, 44(5):1872-1899.
[20]	Xu D Y, Yang H, Xu W H, et al.Inverse design of pancharatnam-berry phase metasurfaces for all-optical image edge detection[J]. Applied Physics Letters,2022, 120(24): 2411011.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

像素大小/p	检测帧率/fps
像素大小/p	特征增强法	增强小目标特征法	改进卷积神经网络法	方法本文
157	48.6	60.6	60.0	120.2
257	48.4	60.8	48.0	120.8
357	48.2	60.2	48.8	120.6
457	48.0	48.2	48.6	120.0
557	24.8	48.4	48.6	100.0
657	24.4	48.2	48.4	100.0
757	24.2	48.0	48.2	100.0
857	24.2	48.0	48.2	80.0
957	24.0	48.0	48.0	80.0
1 057	22.8	48.2	48.0	60.0