A remote sensing image object detection algorithm based on transfer learning

As a method to obtain and process remote sensing images by using remote sensing technology, remote sensing image analysis has been extensively used in various important fields. It plays a key role in environmental monitoring, resource management, urban planning, agricultural forestry, disaster monitoring, climate research, military reconnaissance, public safety, transportation, biomedicine, etc. To solve the problems of remote sensing images such as variable object scale, fuzzy object, and complex background, a remote sensing small sample object detection algorithm RE-FSOD based on feature reweighting was proposed. The model consists of a meta-feature extractor, a feature reweighting extractor, and a prediction module. The meta-feature extractor is composed of CSPDarknet-53, FPN and PAN, and it is responsible for extracting the meta-features of the data; The feature reweighting extractor is used to generate feature reweighting vectors, which are used to adjust meta-features to strengthen features that help detect new categories; The prediction module is composed of the prediction module of YOLOv3.On this basis, the positioning loss function is replaced by the CIOU loss function to improve the positioning accuracy of the model. Finally, training and testing were carried out on the NWPUVHR-10 remote sensing data set. The experimental results revealed that compared with the baseline method FSODM, the method was improved by about 19 %, 11 % and 8 % in the case of 3-shot, 5-shot and 10-shot, respectively.