摘要
以水面无人艇的视觉感知系统的研发为背景,使用级联主成分分析网络的深度学习框架,进行了海面船只检测算法研究.输入海面船只的可见光图像,通过显著性检测确定疑似目标区域,对检测出的疑似目标区域使用PCANet模型进行特征提取,将结果输入支持向量机中,得到最终二分类结果.实验结果表明,所设计的算法可以成功地输出海面船只检测结果,并通过与CNN算法的对比,验证了PCANet方法的高效性和准确性,证明了PCANet在特征提取方面的优越性.
To develop the vision perception system of unmanned surface vehicles,this paper employs the principal component analysis net（PCANet）which is a simple deep learning baseline,and proposes an algorithm for ship detection on the sea surface.First,it inputs a surface vehicle image,and then determines the hypothesis area of the target.Second,it classifies the detected areas using PCANet model,together with a support vector machine（SVM）,and gets the final dichotomous result.The experimental results show that the designed algorithm can successfully output the result of ship detection on the sea surface,and verify that the PCANet method has fast training time and high accuracy,prove that PCANet has the advantages in feature extraction.
出处
《杭州电子科技大学学报（自然科学版）》
2017年第2期23-27,共5页
Journal of Hangzhou Dianzi University：Natural Sciences
基金
国家高技术研究发展计划(863计划)资助项目(2014AA09A510)
关键词
船只检测
深度学习
级联主成分分析
显著性检测
支持向量机
ship detection
deep learning
principal component analysis net
saliency detection
support vector machine
