基于BP神经网络的船舰目标识别分类  

Ship recognition based on BP network

作　　者：梁锦雄[1] 王刻奇 

机构地区：[1]广州城市职业学院信息技术系,广东广州510405 [2]中山大学南方学院电子通信与软件工程系,广东广州510970

出　　处：《舰船科学技术》2015年第3期206-209,共4页Ship Science and Technology

摘　　要：随着各国在海洋领域竞争的日益激烈,船舰目标自动识别技术正逐渐成为研究热点。本文利用BP神经网络对航空母舰、驱逐舰、护卫舰、客船、集装箱、民用货船6种船只类型进行分类,首先对船舰图像进行中值滤波,去除随机噪声和椒盐噪声,然后利用OTSU法将灰度图像分割成背景和目标两部分,接着对目标区域提取了Hu不变矩、边缘梯度方向直方图、周长-面积比3个特征。为了使边缘梯度方向直方图也具有旋转和尺度不变性,本文提出了一种变换方法:将直方图循环右移,直至其最大值到达直方图最右端。最后利用BP神经网络对船舰图像进行了训练和测试。测试结果表明,本文的分类算法对船舰目标的分类精度达到84%左右,有效实现了常见船舰类型的识别分类。With the increasing competition in the field of ocean,automatic ship target recognition technology is becoming a research hotspot of society. Based on BP neural network,we successfully classified six types of ships： aircraft carriers,destroyers,frigates,passenger ship,container and civilian ships. For the recognition procedure,median filtering was used to remove the random noise as well as salt and pepper noise,then gray-scale image were segmented into the background and the target parts using OTSU method,after that,three feature were extracted： Hu invariant moment,the edge gradient direction histogram,the ratio of target perimeter and area. In order to make the edge gradient direction histogram scale and rotation invariant,we proposed a new transform method here： the histogram was shifted to the right until its maximum becomes the most upright one. Finally BP neural network were used for ship training and testing.Test results show that the classification algorithm for ship target recognition reaches classification accuracy as high as 84%,it gives good result for common ship recognition.

关 键 词：BP神经网络 HU不变矩 边缘梯度方向直方图 周长-面积比 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]