|
Abstract:
Object detection in optical aerial images is a fundamental problem in the field of remote sensing and shows great importance in the application. The performance of the early hand-craft-feature algorithm is limited, while deep learning is the primary method for object detection at present. However, due to the characteristics of the remote sensing image itself, it is difficult for the existing detection algorithms to perform well on these images. In this paper, we first describe the characteristics and challenges of the object detection task in aerial images. We then summarize the typical detection methods, including early hand-craft feature extraction methods and current deep learning methods, especially for the deep learning algorithm enhancement upon the characteristics of the aerial images. Then the commonly used detection datasets are introduced, and the performances of the existing methods are compared. Finally, we summarize the current deficiencies and analyze the trends of future studies.
|