MEM-YOLO11:一种多特征选择机制的无人机小目标检测模型

MEM-YOLO11:A UAV small object detection model with multi-feature selection mechanism

摘要

针对遥感图像目标具有多尺度、多旋转角度、背景复杂等特点,提出了一种多特征选择机制融合改进模型 MEM-YOLO11.首先,在backbone中设计了一种特征提取模块,结合了多尺度特征提取和边缘信息增强,提高了对特征信息的整合效果.其次,设计了一种高效的卷积网络模块,由原始图像生成更高分辨率的图像,增强了颈部网络中全局到局部的特征学习能力.并且,重新设计了一个多头部特征提取模块,减少标准卷积中广泛存在的空间和信道冗余,减少了参数量和计算量.最后,优化 MEM-YOLO11 模型的超参数,缓解分类和定位不相关问题.在 VisDrone2019数据集上精度、召回率、mAP50 分别提升7.6%、5.6%、6.5%;WiderPerson 数据集上精度、召回率、mAP50分别提升6.8%、4.2%、5.6%.

Abstract

To address the challenges of multi-scale objects,varied orientations,and complex backgrounds in remote sensing images,we propose MEM-YOLO11,an enhanced model integrating a multi-feature selection mechanism.First,a feature extraction module is designed within the backbone network,combining multi-scale feature extraction and edge information enhancement to improve feature integration.Second,an efficient convolutional network module is developed to generate higher-resolution images from raw inputs,strengthening the global-to-local feature learning capability in the neck network.Additionally,a reconstructed multi-head feature extraction module reduces spatial and channel redundancy inherent in standard convolutions,significantly decreasing parameters and computational costs.Finally,hyperparameters of MEM-YOLO11 are optimized to alleviate the discrepancy between classification and localization tasks.Evaluations demonstrate significant improvements:on the VisDrone2019 dataset,precision,recall,and mAP50 increase by 7.6%,5.6%,and 6.5%,respectively;on the WiderPerson dataset,gains of 6.8%,4.2%,and 5.6% are achieved for the same metrics.

关键词

小目标检测 / 多尺度特征提取 / 边缘增强 / 超参数

Key words

small object detection / Multi-scale feature extraction / edge enhancement / hyperparameter

参考文献

[1] Li LZ, Han L, Ye YX, et al. Deep learning in remote sensing image matching:a survey[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2025, 225: 88-112.

[2] 叶栢铖, 朱尤攀, 周永康, 等 . 轻量级目标检测算法综述[J]. 红外技术, 2025, 47(3): 289-298.

[3] Ye BC, Zhu YP, Zhou YK, et al. Review of lightweight target detection algorithms[J]. Infrared Technology, 2025, 47(3): 289-298.

[4] 王宁, 智敏. 深度学习下的单阶段通用目标检测算法研究综述[J]. 计算机科学与探索, 2025, 19(5): 1115-1140.

[5] Wang N, Zhi M . Review of one-stage universal object detection algorithms in deep learning[J]. Journal of Frontiers of Computer Science and Technology, 2025, 19(5): 1115-1140.

[6] 吴一全, 童康. 基于深度学习的无人机航拍图像小目标检测研究进展[J]. 航空学报, 2025, 46(3): 174-200.

[7] Wu YQ, Tong K . Research advances on deep learning-based small object detection in UAV aerial images[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(3): 174-200.

[8] Li JL, Zheng CH, Chen P, et al. Small object detection in UAV imagery based on channel-spatial fusion cross attention[J]. Signal,Image and Video Processing, 2025, 19(4): 302.

[9] Yi WG, Wang B . Research on underwater small target detection algorithm based on improved YOLOv7[J]. IEEE Access, 2023, 11: 66818-66827.

[10] Guan ST, Lin YM, Lin GY, et al. Real-time detection and counting of wheat spikes based on improved YOLOv10[J]. Agronomy, 2024, 14(9): 1936.

[11] Lv LJ, Li XJ, Mao FJ, et al. A deep learning network for individual tree segmentation in UAV images with a coupled CSPNet and attention mechanism[J]. Remote Sensing, 2023, 15(18): 4420.

[12] 王文润, 党建武, 王阳萍, 等 . 基于复杂背景的多尺度特征融合手-物交互检测方法[J]. 兰州交通大学学报, 2024, 43(5): 94-102.

[13] Wang WR, Dang JW, Wang YP, et al. A multi-scale feature fusion method for hand-object interaction detection based on complex background[J]. Journal of Lanzhou Jiaotong University, 2024, 43(5): 94-102.

[14] 倪康, 孙笠焜, 邹旻瑞. 基于边缘特征引导学习的SAR目标检测[J]. Journal of Data Acquisition & Processing, 2025, 40(3): 612-624.

[15] 颜豪男, 吕伏, 冯永安. 特征级自适应增强的无人机目标检测算法[J]. 计算机科学与探索, 2024, 18(6): 1566-1578.

[16] Yan HN, Lü F, Feng YA . Feature-level adaptive enhancement for UAV target detection algorithm[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(6): 1566-1578.

[17] 张浩, 左杭, 刘宝华. 视觉与二维激光雷达的目标检测方法[J]. 电子测量与仪器学报, 2022, 36(3): 79-86.

[18] Zhang H, Zuo H, Liu BH . Target detection method for visual and 2D laser radar[J]. Journal of Electronic Measurement and Instrumentation, 2022, 36(3): 79-86.

[19] 黄骏, 郭颖, 严舒. 基于多感受野与动态特征细化的遥感图像检测算法[J]. 激光与光电子学进展, 2024, 61(22): 2228004.

[20] Huang J, Guo Y, Yan S . Remote sensing image detection algorithm based on multi-receptive field and dynamic feature refinement[J]. Laser & Optoelectronics Progress, 2024, 61(22): 2228004.

[21] Cui YN, Ren WQ, Knoll A. Omni-kernel network for image restoration[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(2): 1426-1434.

[22] Yu ZP, Huang HB, Chen WJ, et al. YOLO-FaceV2:a scale and occlusion aware face detector[J]. Pattern Recognition, 2024, 155: 110714.

[23] 梁嘉杰, 李星星. 特定任务上下文解耦的遥感图像目标检测方法[J]. 计算机工程与应用, 2025, 61(2): 293-303.

[24] Liang JJ, Li XX . Task-specific context decoupling object detection method for remote sensing images[J]. Computer Engineering and Applications, 2025, 61(2): 293-303.

[25] Du D, Zhu P, Wen L, et al. VisDrone-DET2019:The vision meets drone object detection in image challenge results[EB/OL].[2025-04-17]. https://www.semanticscholar.org/paper/VisDrone-DET2019%3A-The-Vision-Meets-Drone-Object-in-Zhu-Wen/80edf721b4ab5f4f4e24db69131c360d3343eaed.

[26] Zhang SF, Xie YL, Wan J, et al. Wider Person:a diverse dataset for dense pedestrian detection in the wild[J]. IEEE Transactions on Multimedia, 2020, 22(2): 380-393.

基金资助

国家自然科学基金(11904085)

河南省科技攻关项目(202102210299)