海洋溢油合成孔径雷达图像特征提取及其关键度分析

Feature extraction and its criticality analysis for oil spill detection in synthetic aperture radar images

就SAR图像溢油检测的方法论而言,用于识别溢油和疑似现象的定性或定量的统计特征量选择,通常是任意的。对于不同的分类模型,所选用的特征量也不尽相同。主要是进行海洋SAR图像特征提取及其关键度分析。其目的是将"最小距离"判别法应用于海上溢油和疑似溢油的识别研究。首先,针对海洋SAR图像溢油检测常用的特征量,进行冗余处理;然后,引入关键系数,定量地研究特征量的关键度,提取显著特征量;藉以构造一个多维的特征矢量空间,以适于最小距离判别法在特征矢量空间中进行溢油和疑似溢油的识别研究。

In terms of the methodology of oil spill identifying in SAR images, it is usually arbitrary to select qualita- tive or/and quantitative features for classifying dark objects as oil spill or look alikes. The features seleeted in dif- ferent classification models are not the same. The feature extraction and the eritieality analysis are made in SAR images. Its aim is to apply the minimum distance method to discriminating oil spills from look-alikes. First, through correlation analysis, the redundancy is removed. Next,a criticality coefficient is introduced to quantitative- ly study the criticality of features. Then, distinguishing features are extraeted. Sequentially, the dimension of fea- ture veetor is reduced to fit for the application researeh of the minimum distance method.