Ship Detection in Optical Satellite Images Using Haar-like Features and Periphery-Cropped Neural Networks

Abstract

The ship detection field faces many challenges due to the large-scale and high complexity of optical remote sensing images. Therefore, an innovative ship detection method that is simple, accurate, and stable is proposed in this paper. The algorithm consists of the following two steps: 1) the AdaBoost classifier, combined with Haar-like features, is used to rapidly extract candidate area slices, and 2) according to the characteristics of ships, a periphery-cropped network is designed for ship verification. Furthermore, we analyze the characteristics of ocean images to improve the contrast between the target and the background. Thus, an RGB spectrum-stretching method is proposed. Finally, we evaluate our method using spaceborne optical images from the Jilin-1 satellite, Google satellites, and the public dataset NWPU VHR-10. Our experimental results indicate that the proposed algorithm achieves a high detection rate.

Keywords

• Object detection
• feature extraction
• CNN
• AdaBoost classifier