梯度纹理直方图与多层感知器船舶快速检测

Abstract

Aiming at the issues of low ship detection rate caused by the failure of background modeling in the dynamic complex environment of traditional ship detection methods, a rapid ship detection algorithm based on gradient texture histogram features and multilayer perceptron was proposed. The feature fusion between gradient and texture histogram of the target was performed using multilayer perceptron, constructing the feature space for ship targets. Firstly, the region proposal model based on binarized normed gradient feature was trained to quickly generate a small number of ship candidate windows with high recall rate and then the gradient texture histogram features were extracted from each candidate window. Secondly, a multilayer perceptron was designed as a ship classifier to distinguish the gradient texture histogram features. Experimental results show that the proposed algorithm has an average precision of 90.0% and an average time of 20.4 ms/frame in multiple maritime scenes, which effectively realizes rapid ship detection in maritime scenes.