摘要
Abstract
With the improvement of unmanned aerial vehicle(UAV)image resolution and the development of object detec-tion technology,object detection based on UAV images has become a research hotspot.However,UAV images have prob-lems such as complex backgrounds,low color contrast,high object overlap,large scale variation,and a high proportion of small objects,which easily lead to severe noise interference,unbalanced sample distribution,and great difficulty in feature extraction,significantly increasing the implementation difficulty of detection tasks.To address these problems,this paper improved the you only look once version 5 small(YOLOv5s)model algorithm.An improved convolutional block attention module(CBAM)was added to focus on object features through a sequential structure from channel to spatial,featuring both light weight and high efficiency;multi-scale feature fusion was innovated to combine shallow location information and deep semantic information,and a tiny object detection layer was newly added to the detection head to enhance the feature extrac-tion capability of small objects;the K-means clustering algorithm(K-means)was utilized to optimize the prior box param-eters to improve model adaptability.The experimental results show that the mean average precision(mAP)of the improved model increases by 2.8%,and the model has broad application prospects in fields such as military,rescue,and transporta-tion,which can provide a new direction for the technological development of UAV image object detection.关键词
深度学习/目标检测/无人机(UAV)影像/特征提取/第五版轻量型单次检测网络(YOLOv5s)Key words
deep learning/object detection/unmanned aerial vehicle(UAV)image/feature extraction/you only look once version 5 small(YOLOv5s)分类
天文与地球科学
