基于Faster R-CNN的海上舰船识别算法

Marine Ship Recognition Algorithm Based on Faster-RCNN

摘要: 针对现有的舰船目标自动识别方法容易受到物理噪声干扰、实时性差等问题，提出一种基于深度学习中Faster R-CNN (快速区域卷积神经网络)的海上舰船识别算法。首先建立了一套海上舰船图片的训练集与测试集；其次为了增强网络的泛化能力，在区域生成网络的第一个全连接层后增加了一个dropout层；最后为了减小过拟合，在分类时只使用了一个含有2048个神经元的全连接层。目前算法可以将海上舰船目标自动识别为航母、其他军舰、民船三类，在本文设定的测试集上准确率为90.4%，检测速度为每秒15帧左右。

Abstract: In order to solve the problem that the existing automatic ship-target-recognition methods are vulnerable to physical noise interference and poor real-time performance, an algorithm based on Faster R-CNN in depth learning is proposed. Firstly, a set of training set and test set of marine ship images are established; secondly, in order to enhance the generalization ability of the network, a dropout layer is added after the first full connection layer of region generating network; finally, in order to reduce over-fitting, only a full connection layer containing 2048 neurons was used for classification. At present, the algorithm can automatically identify ship targets as aircraft carriers, other warships and civilian ships. The accuracy of the test set in this paper is 90.4 per cent and the detection speed is about 15 frames per second.