Few-Shot Optical Remote Sensing Object Detection Based on Shape Matching

Traditional object detection methods suffer from excessively high false alarm rates in scenarios with scarce training samples. To address this issue, this paper proposes a few-shot optical remote sensing object detection method based on shape matching. This method first obtains the sliced mask image of the object within the predicted bounding box in the pre-detection process. Then, the shape context algorithm is utilized to perform template matching between the template image and the sliced mask image, thereby eliminating false alarm objects and enhancing the object detection performance. To examine the capability of the proposed method, two public datasets, namely FAIR1M datasets and MAR20 datasets, are selected for experimentation. The experimental results show that our method achieves exceptional effectiveness, yielding F1 scores of 0.705 and 0.866 on the FAIR1M and MAR20 datasets, respectively.