Abstract: In response to the public safety hazards caused by small UAVs (unmanned aerial vehicles) flying indiscriminately at low altitude in cities, a target detection model based on improved YOLOv7 is proposed to address the existing problems of low accuracy and poor real-time detection of small moving targets. The RepVGG lightweight network is introduced into the backbone network to reduce computational complexity and achieve real-time detection requirements. Meanwhile, a C3m module is proposed to solve the issues of small UAV targets with limited presence and limited feature information in images, so as to improve the detection accuracy. On this basis, the CBAM convolutional attention mechanism is inserted between the backbone network and the feature fusion network to improve the sensitivity of the network to multi-scale UAV targets and reduce the influence of noise. Finally, extensive experiments are conducted on a self-built urban low-altitude UAV dataset. The results demonstrate that the improved YOLOv7 model, compared to the original YOLOv7 model, reduces the number of parameters by 6×106, decreases the floating-point operations by 77.9%, increases the detection accuracy (mAP@0.5) to 95.1%, improves the detection speed (FPS) by 43.7%, thus achieving real-time and high-precision detection of UAV.

Key words: YOLOv7, UAV detection, lightweight network, attention mechanism

摘要： 面向城市低空小型无人机（unmanned aerial vehicle，UAV）乱飞引发的公共安全隐患，针对目前存在的动小目标检测精度低、实时性差等问题，提出了一种基于改进YOLOv7的目标检测模型。该模型在主干网络中引入RepVGG轻量化网络，降低计算复杂度以满足实时性要求；提出C3m模块以解决无人机小目标在图像中占比小、特征信息有限等问题，从而改善检测精度；在此基础上，将CBAM卷积注意力机制插入到主干网络和特征融合网络之间，提高网络对多尺度无人机目标的敏感度，降低噪声影响。在自构建的城市低空无人机数据集上进行消融实验，结果表明改进后的YOLOv7模型与原始YOLOv7模型相比，参数量减少了6×106，浮点运算量降低了77.9%，检测精度mAP@0.5提高至95.1%，检测速度FPS提高了43.7%，实现了对复杂场景下小型无人机目标的实时高精度检测。

关键词: YOLOv7, 无人机检测, 轻量化网络, 注意力机制