RIASSUNTO
Along with the increasingly prominent role of ship automatic detection in the search and rescue of distressed ships, combating illegal fishing, marine traffic control and marine warfare, this paper proposes a ship detection method based on shape and texture features. Firstly, this article extracts the ship candidate area; combines the feature extraction algorithm to extract 5 shape characteristics such as unevenness, eccentricity, and symmetric regularity index, and then uses the LBP(Local Binary Pattern) rotation invariant uniform pattern algorithm to extract the LBP value of the ship candidate area and combine the LBP values For statistical histograms, the statistical histograms are finally connected into a multi-dimensional vector; shape features and LBP multi-dimensional vectors are combined into feature vectors, and the feature vectors are used to complete the training of the LightGBM classifier, and the trained LightGBM classifier is used to complete the final ship Detection. The detection algorithm in this paper is based on the 768 pixel * 768 pixel ship image provided by the Airbus Ship Detection Challenge, and finally forms a 46-dimensional feature vector with a recall rate of 93.3% and an accuracy rate of 92.9%. The results show that this method can not only effectively exclude the influence of clouds, small islands, and light, but also has a high recognition rate, and effectively realize accurate and rapid detection of ships at sea.