摘 要：针对无人机航拍图像中存在的目标边缘信息模糊、复杂背景干扰以及小目标分辨率低、难以辨识等问题，提出了一种新型改进RT-DETR的无人机航拍图像目标检测方法，记作MCS-DETR。首先，在主干网络中设计多尺度边缘增强模块，通过构建边缘信息增强机制，并结合深度卷积操作和特征融合策略，实现在不同尺度上提取特征信息，增强模型对图像边缘信息的感知能力。接着，在尺度内特征交互机制中融入卷积加法标记混合器，以优化特征交互过程，提升模型对全局上下文关键信息的捕获能力。最后，在原有特征融合方式的基础上进一步改进创新，提出小目标增强金字塔网络，增强模型对小目标细节特征的提取能力。实验结果表明，与RT-DETR模型相比，改进后的MCS-DETR算法在Visdrone2019-DET-Test上，参数量减少了20.7%，准确率、召回率、mAP@50和mAP@50:95分别提升了2.4%、3.1%、2.8%和2.0%，能够有效改善复杂场景下无人航拍图像目标检测的误检、漏检问题。

Abstract: In response to the challenges in unmanned aerial vehicle (UAV) aerial imagery, such as blurred object edge infor-mation, complex background interference, and the low resolution and difficulty in identification of small objects, this paper pro-poses a novel improved RT-DETR-based object detection method for UAV aerial images, named MCS-DETR. Initially, a multi-scale edge information enhancement module is designed within the backbone network. By constructing an edge information enhancement mechanism and integrating it with deep convolutional operations and feature fusion strategies, the module aims to extract feature information at different scales and enhance the model’s perception of image edge information. Subsequently, a convolutional additive token mixer is incorporated into the intra-scale feature interaction mechanism to optimize the feature in-teraction process and improve the model’s capability of capturing global contextual key information. Finally, based on the origi-nal feature fusion method, a small object enhancement pyramid network is proposed to enhance the model’s ability to extract detailed features of small targets. Experimental results indicate that, compared to the RT-DETR model, the MCS-DETR algo-rithm reduces the parameter size by 20.7% on the Visdrone2019-DET-Test dataset. Additionally, it achieves increases in accura-cy, recall, mAP@50, and mAP@50:95 of 2.4%, 3.1%, 2.8%, and 2.0%, respectively, effectively improving the issues of missed and erroneous detections in complex scenes for UAV aerial image object detection.