Target Detection in Remote Sensing Images Based on Multi-scale Fusion Faster RCNN

Abstract

Target detection in remote sensing images is widely used in civil and military applications and plays an increasingly important role. In this paper, a novel multi-scale fusion Faster RCNN model is proposed to realize target detection in optical remote sensing images. First, the preprocessing operation is performed on the dataset by using VOC format for pretreatment and data enhancement. Second, based on traditional Faster RCNN architecture, the EfficientNetb0 and FPN is combined to extract multi-scale fusion image features, so as to obtain pyramid feature and improve the ability of network. Finally, the prepared dataset is used to train the multi-scale fusion Faster RCNN model, and then the prediction results are obtained by ROI Polling. The experimental results demonstrate that the proposed method in this paper can achieve higher accuracy as high as 84.4% of target detection in remote sensing images while reducing the size of the model.