摘要
针对侧扫声纳图像目标边缘检测困难的问题,利用二维离散小波变换对侧扫声纳(SSS)声图进行多分辨率分析,对大尺度分解的小波系数进行非极大值抑制,并重构小尺度上的低频分量。联合各尺度上的低频分量,构建SSS声图像素点处特征向量,构成其特征空间,对特征空间进行主成分分析,压缩其维数,并对压缩后的特征向量进行K-均值聚类分析,提取类间边缘线。利用含有沉船的SSS声图,并在其均质区域内加入目标与声影进行验证实验。该方法在实验中边缘检测准确率为0.90,表明该方法的有效性。
Aimed at edge-detection in side-scan sonar （SSS） images, multi-resolution analysis is used on the SSS image based on 2-D discrete wavelet transform. After non-maxima suppression on wavelet coefficients on large level, low frequency component is reconstructed on small level. Combined with low frequency components on different levels, a feature vector is extracted for each pixel from the components, and then the feature space is constructed. The dimensionality of the space is reduced using principle component analysis. The feature vectors are then clustered into disjoint clusters using K-Mean clustering. Edges are extracted from clusters. The image containing a shipwreck is used to verify the algorithm, with a target and its shadow added in homogeneous region of the SSS image. In the experiment, the accuracy rate of the edge detected is 0.90, indicating the validity of the algorithm.
出处
《海洋测绘》
2014年第3期63-66,共4页
Hydrographic Surveying and Charting
关键词
侧扫声纳
边缘检测
聚类分析
多分辨率分析
side-scan sonar
edge-detection
clustering
multi-resolution analysis
