基于全维动态卷积的无人机航拍图像轻量化目标检测

LIGHTWEIGHT TARGET DETECTION FOR UAV AERIAL IMAGE BASED ON OMNI-DIMENSIONAL DYNAMIC CONVOLUTION

针对传统无人机航拍图像目标检测中出现的模型体积大、检测准确率低等问题,提出一种轻量化的无人机航拍图像目标检测算法。以YOLOv5s为基础,增加小目标检测层,采用全维动态卷积替换普通卷积,减少了参数量。使用跨层跨尺度的加权特征融合,并引入FasterNet模块,加强特征提取能力。使用动态标签分配策略,显著提升检测精度。实验结果表明,提出的算法在准确率和模型体积方面优于原YOLOv5s算法,可以更高效地完成无人机航拍图像的目标检测任务。

A lightweight target detection algorithm for UAV aerial images is proposed to address the problems of large model size and low detection accuracy in traditional UAV aerial image target detection.Based on YOLOv5s,a small target detection layer is added,and a omnidimensional dynamic convolution is used to replace the ordinary convolution,which reduces the number of parameters.Using crosslayer and crossscale weighted feature fusion,and introducing FasterNet module,the feature extraction capability is strengthened.A dynamic label assignment strategy is used to significantly improve the detection accuracy.The experimental results show that the proposed algorithm outperforms the original YOLOv5s algorithm in terms of accuracy and model volume,and can more efficiently accomplish the task of target detection in UAV aerial images.