基于高光谱“图谱”全尺度信息的海上溢油检测与油种识别

Oil spill detection plays an important role in the research field of ocean remote sensing.The most difficulty is how to discriminate oil spill from suspected oil spill. This research put forward a new mothed of the oil spill detection by using the "spectrum-image"information based on the complete muti-scale analysis technology.On the basis of using the spctrum ,textrue,shape features of the target,we also quantitativly descripted the directional features of the targets,and then wind and flow field information were introduced into the model in order to eliminate the disturbance of suspeted oil spill to the greatest extent to improve accuracy of oil spill recognition. Research Content:1)Study on how to choose the useful bands and textures of the oil spill; 2) Study on how to select the muiti-scale spectral and multi-scale texture features of the oil spill; 3) Quantitatively descripting the geometric features and directional features of the oil spill and suspected oil spill; 4) Optimization of feature set composed of the multi-scale spectral features, multi-scale texture features and geometric features based on rough set mothod ; 5)Design and optimization on the Ridgelet neural network and experimental verification based on the actual data. Researching significance:to break through the difficulty in oil spill detection ,and to improve the accuracy of recognition ; to provide technical support for the operational monitoring of surveillance department; to deepen basic research on the target recognition by hyperspectral remote sensing.

海洋溢油检测是海洋遥感的重要研究领域。检测中的最大难点是溢油与疑似溢油以及不同油种的准确区分。 本研究提出一种基于高光谱"图谱"全尺度信息的海上溢油精细化探测新方法。在提取目标的全尺度光谱、纹理和形状特征基础上，定量刻画目标方向特征，并将海洋风场与流场信息引入模型，最大限度地消除疑似溢油的干扰，提高溢油识别的准确性。 研究内容：（1）海上溢油敏感波段选择与图像纹理特征选择；（2）基于全尺度分解的海上溢油与疑似目标的可区分性光谱与纹理特征选择；（3）溢油与疑似目标几何形态与方向特征的定量刻画；（4）基于粗集方法的全尺度光谱与纹理特征和形态特征集的约简与优化；（5）溢油脊波神经网络分类器的设计与优化及实验验证。 研究意义：突破海上溢油检测的难点，提高溢油检测的准确性，为海监部门溢油业务化监测提供技术支持，同时深化高光谱遥感目标识别的应用性基础研究。

海面溢油遥感检测是国内外研究的热点和难点，尤其是溢油和疑似油膜的识别是亟待解决的难题。该课题借助高光谱遥感图-谱合一的优势开展了以下研究：1）高光谱图像曲波与小波串联去噪方法研究；2）油膜空间分布形态结构和光谱的差异性研究；3）基于粗集方法的溢油敏感波段选择与图像纹理特征选择方法的研究；4）神经网络分类器的设计与优化及实验验证。该研究方法消除了疑似溢油的干扰，提高溢油识别的准确性。