Learning Critical Features for Arbitrary-Oriented Object Detection in Remote-Sensing Optical Images
Peng Sun,Yongbin Zheng,Wenqi Wu,Wanying Xu,Shengjian Bai,Xiaoping Lu
DOI: https://doi.org/10.1109/tim.2024.3378265
IF: 5.6
2024-04-02
IEEE Transactions on Instrumentation and Measurement
Abstract:Arbitrary-oriented object detection in remote-sensing optical images is an emerging, yet challenging task due to the diversity of the visual appearance of objects. Although considerable progress has been made, existing methods often suffer from insufficient positive samples and critical feature misalignment caused by objects with considerable diversity. In this article, we propose a critical feature learning (CFL) method to address the above issues. First, to obtain a more accurate location, we introduce the spatial transform to model the rotating bounding box regression and construct a spatial transform selection (STS) strategy, which is used to implement dynamic sample selection metrics to guarantee sufficient positive samples. Then, we propose a scale-controlled smooth loss (SC-Loss) to change the form of regression loss function based on the statistics of proposals so that the training can focus more on high-quality samples. Finally, to capture critical features, we propose a critical feature sampling (CFS) module to perform location refinement of classification feature samples to extract accurate features. Comprehensive and extensive experimental results based on three benchmarks, DOTA, FAIR1M-1.0, and HRSC2016, demonstrate that the proposed method is superior to existing methods in terms of accuracy and effectiveness.
engineering, electrical & electronic,instruments & instrumentation