A Method for Fast Ship Detection and Recognition in Sea-Sky Background Based on Multi-scale Feature Cluster

This paper approaches the problem of detecting and recognizing ship targets in color images based on sea-sky background. We propose a method based on multi-scale feature cluster robust to the changes in scale, orientation and illumination of the image and can fast detect and recognize ship targets. For ship detection, a "group feature" detecting method that can adapt to the multi-scale of the target and improve the adaptive of the canny edge detection is proposed that combines the canny edge detection algorithm, image pyramid algorithm and minimum spanning tree clustering algorithm. For ship recognition, a "probability tree" recognition method that can recognize a ship target in multi-angle, multi-target approach is proposed, combining with the probability tree classifier and principal component analysis algorithm. The recognition results are calculated by the total probability formula.