基于深度卷积神经网络的船舶识别方法研究  

Research on ship recognition method based on deep convolutional neural network

作　　者：赵亮[1] 王晓峰[1] 袁逸涛 

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《舰船科学技术》2016年第8期119-123,共5页Ship Science and Technology

基　　金：国家海洋公益专项资助项目(201305026)

摘　　要：为解决目前船舶识别率较低的问题,基于深度卷积神经网络算法,提出一种在深度卷积神经网络基础上的改进算法。利用卷积神经网络对船舶图片进行深度特征提取,结合HOG算法得到准确的边缘特征,结合HSV算法得到颜色特征,通过SVM分类器对船舶进行分类。算法主要包括2个阶段:训练阶段实现卷积神经网络的预训练,将得到特征归一化,PCA降维,通过HOG算法得到边缘特征,最后训练SVM分类器;测试阶段则对算法的准确性进行核实。实验结果表明,该方法平均识别正确率达到93.6%,可以很好地实现船舶识别。In order to solve the problem of low recognition rate of ships, a new algorithm based on the deep convolutional neural network is proposed.Using the convolutional neural network to extract the depth of the ship image, the HOG algorithm is used to get the accurate edge feature, combining the HSV algorithm to get the color characteristics, and the ship is classified by SVM classifier. The algorithm mainly consists of two stages： the training stage to achieve the pre training of convolutional neural network, will get the feature normalization, PCA dimension reduction, through the HOG algorithm to get edge features, and finally trained SVM classifier.In the test stage, the accuracy of the algorithm is verified. Experimental results show that the average recognition accuracy of the proposed method is 93.6%, which can be very good to achieve the recognition of the ship.

关 键 词：深度卷积神经网络 船舶识别 边缘梯度方向直方图 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]