特定任务上下文解耦的遥感图像目标检测方法

doi:10.3778/j.issn.1002-8331.2309-0119

摘要/Abstract

摘要： 针对典型目标检测模型在遥感图像检测任务中因检测目标小而密集、尺度差异大、方向随机而背景复杂，导致漏检率高、边框回归精度差等问题，提出一种特定任务上下文解耦和快速部分卷积的遥感图像检测方法Faster-
YOLO-TSCDH。将检测方法改进为特定任务上下文解耦检测方法，将分类任务和回归任务分开处理，分别融合不同空间特征和语义特征的特征图，降低不同任务的相互干扰，提高检测精度和鲁棒性。提出一种快速部分卷积多层次聚合模块，改进特征提取阶段的跨阶段分部卷积模块，强化特征提取能力，同时减轻解耦头带来参数量和运算量暴增的问题。采用一种对锚框质量动态评估的边框回归损失Wise-IoU，减少过高质量或过低质量锚框对边框回归的负面影响，提高边框回归的整体性能。实验结果表明，在DOTAv2和AI-TOD两个公共遥感图像数据集进行目标检测任务时的平均精度均值（mAP@IoU=0.5）达到65.4%和51.3%，相较基准模型提升了3到5个百分点，证明了改进方法的可行性和有效性。

关键词: 深度学习, 目标检测, 遥感图像, 小目标, 解耦检测, 特征提取

Abstract: A remote sensing image detection method FasterYOLO-TSCDH based on task-specific context decoupling and fast partial convolution is proposed to address the issues of high miss rate and poor bounding-box regression accuracy caused by small and dense detection objects, large scale differences, random directions, and complex backgrounds in typical object detection models in remote sensing image detection tasks. This paper improves the detection method to a task-specific context decoupling detection method, separating the classification task and regression task, and fusing feature maps of different spatial and semantic features separately to reduce mutual interference between different tasks and improve detection accuracy and robustness. The paper proposes a fast partial convolution multi-level aggregation module, which improves the cross stage partial convolution module in the feature extraction stage, strengthens the feature extraction ability, and reduces the problem of parameter and computational bulk caused by decoupling heads. It adopts a dynamic evaluation of the quality of anchor frames using Wise-IoU to reduce the negative impact of high or low quality anchor on bounding-box regression and improve the overall performance of bounding-box regression. The experimental results show that the proposed method achieves a mAP@IoU=0.5 of 65.4% and 51.3% for object detection tasks on two common remote sensing image datasets, DOTAv2 and AI-TOD, which is 3 to 5 percentage points higher than the baseline model. The paper proves the feasibility and effectiveness of the improved method.

Key words: deep learning, object detection, remote sensing image, small object, decoupled detection, feature extraction

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.

LIANG Jiajie, LI Xingxing. Task-Specific Context Decoupling Object Detection Method for Remote Sensing Images[J]. Computer Engineering and Applications, 2025, 61(2): 293-303.

参考文献

[1] REDMON J, DIVVALS S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[2] GE Z, LIU S, WANG F, et al. YOLOx: exceeding YOLO series in 2021[J]. arXiv:2107.08430, 2021.
[3] TIAN Z, SHEN C, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 9627-9636.
[4] LI C, LI L, JIANG H, et al. YOLOv6: a single-stage object detection framework for industrial applications[J]. arXiv:2209.
02976, 2022.
[5] ZHUANG J, QIN Z, YU H, et al. Task-specific context decoupling for object detection[J]. arXiv:2303.01047, 2023.
[6] HOWARD A G, ZHU M, CHEN B, et al. Mobilenets: efficient convolutional neural networks for mobile vision applications[J]. arXiv: 1704.04861, 2017.
[7] HAN K, WANG Y, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1580-1589.
[8] DING X, ZHANG X, MA N, et al. RepVGG: making VGG-style convnets great again[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 13733-13742.
[9] CHEN J, KAN S H, He H, et al. Run, don’t walk: chasing higher FLOPS for faster neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 12021-12031.
[10] TONG Z, CHEN Y, XU Z, et al. Wise-IoU: bounding box regression loss with dynamic focusing mechanism[J]. arXiv:2301.10051, 2023.
[11] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6)：1137-1149.
[12] WANG C Y, LIAO H Y M, YE I H. Designing network design strategies through gradient path analysis[J]. arXiv:2211.04800, 2022.
[13] XIA G S, BAI X, DING J, et al. DOTA： a large-scale dataset for object detection in aerial images[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 3974-3983.
[14] WANG J, YANG W, GUO H, et al. Tiny object detection in aerial images[C]//Proceedings of the 25th International Conference on Pattern Recognition, 2021: 3791-3798.
[15] TAN M, LE Q. Efficientnet: rethinking model scaling for convolutional neural networks[C]//Proceedings of the International Conference on Machine Learning, 2019: 6105-6114.
[16] WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 7464-7475.