Aiming at the problem that conventional remote sensing image target models are difficult to deploy and run on low-power hardware,this paper proposes a lightweight remote sensing image multi-category target detection model.In different layers of the fea-ture extraction network,the convolution kernel group with SE channel attention module and the convolution kernel group with the bot-tleneck structure are used for feature extraction,and then a multi-scale enhancement network containing channel attention is used to output feature map parameters of three scales for final detection.Using RSOD and Google Earth as data sources,the data set is con-structed,and the training set is enhanced with samples.The experimental results show that the model proposed in this paper can quickly and accurately detect multi-category targets in different environments,and the model after training occupies less memory and has a low number of operating parameters,and can be deployed in low-power hardware terminals to quickly and accurately detect ob-jects in remote sensing images.
NETL
NSTL
万方数据
