Ship Detection in Optical Satellite Images Based on Sparse Representation

Ship detection in remote sensing imagery has been widely applied in military and citizen applications, such as fishery management, vessel surveillance or marine safety and security. With the development of optical satellite, optical satellite imagery ship detection has caused a lot of attention. In this paper, we propose an offshore ship detection method based on sparse representation. First we employ histogram of oriented gradient (HOG) as the feature descriptor, then the HOG feature are extracted from training dataset. After feature extraction, all of samples are used to adaptively train a dictionary. Next, we encode HOG feature description of patches from test image by the dictionary. Finally, the sparse code and support vector machine (SVM) classification are employed in ship target validation and false alarms elimination. Experiments have shown better detection performance and stronger robustness of our method compared with other methods.