摘要
针对海浪、海雾会造成无人艇视觉系统采集的视频图像模糊,影响目标特征提取和识别,且单一特征提取不能有效识别水面多类目标问题,在对无人艇采集的视频序列图像进行海雾去除和电子稳像预处理的基础上,提出一种多类水面目标组合特征提取和识别方法。首先分割清晰化后的图像,实现目标背景分离;然后提取不同目标几何特征、不变矩特征和纹理特征;最后采用基于组合特征和主分量分析降维的分级BP神经网络进行目标识别。通过实测、网络搜索和3D建模获得礁石、岛屿与船只3大类目标样本库,在MATLAB7.9下进行仿真研究,结果表明:多类水面目标组合特征提取和识别方法能有效实现无人艇视觉系统对海面3大类常见目标的分类识别,识别正确率达到85%以上。
Sea waves and fog can cause video images captured by visual systems in a unmanned surface vehicle degradation and fuzzy,and influence targets feature extraction and recognition of targets.One single feature extraction is hard to identify multi-types of surface targets effectively.A method to extract and to identify features of multi-types surface targets for unmanned surface vehicle is proposed based on the fact that the video image sequence is preprocessed by removing sea fog and electronic image stabilization.At first,targets and background are separated by segmenting images.The geometrical feature,the moment invariant feature and texture feature in different targets are extracted.Then the grading BP neutral network with principal component analysis and dimension reduction is used to identify targets.A sample library of three types of targets such as reef,islands and ships are obtained through real measurements,network searching and 3D modeling.Simulations with MATLAB 7.9 show that the proposed method recognizes three types of common surface targets effectively,and the recognition accuracy is above 85 ％.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2014年第8期60-66,共7页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(51109047)
关键词
无人艇
视觉系统
水面目标
组合特征
识别
unmanned surface vehicle
visual system
surface targets
combination feature
recognition
