Few-shot Object Detection on Remote Sensing Images Based on Feature Reweighting
Aimed at the characteristics of variable target size,fuzzy target,and complex background,a few-shot object detection model based on feature reweighting is proposed.The model consists of three parts of element feature extractor,feature reweighting extractor and prediction module.The element feature extractor is composed of CSPDarknet-53,feature pyramid network(FPN)and path aggregation network(PAN),which is responsible for extracting the element features of data.The feature reweighting extractor is used to generate the feature reweighting vectors,and adjust the element features to enhance the helpful features for detecting new classes.The prediction module is composed of the prediction module of YOLOv3.On this basis,the positioning loss function is re-placed by CIOU to improve the positioning accuracy of the model.Finally,the training and testing are carried out on the NWPU VHR-10 remote sensing data set.The experimental results show that compared with the baseline method FSODM,the mean average precision(mPA50)improves by 19％,11％and 8％at the conditions of 3-shot,5-shot and 10-shot respectively.
万方数据
维普
