基于BP神经网络的云相态检测方法研究

利用MODIS中5个光谱波段上不同云相态的特性,提出了一种基于BP神经网络的云相态检测方法。首先,分析了所选波段上不同云相态的特性,利用5个波段上光谱图像的反射率、亮温值和亮温差值构成4组特征数据作为输入层,隐层和输出层分别采用优化的传输函数。然后,利用3层前馈型BP神经网络对所选波段MODIS数据进行了云相态检测。最后,将两组测试数据用该BP神经网络算法进行云相态检测的结果与相应MOD06云相态数据进行了对比分析,结果表明该方法能很好地识别云相态,检测平均准确率达到86.11%,计算结果与标准结果平均相关性达到0.874的高度相关,且无需在计算前进行复杂的云和晴空分离处理。

A Research on Cloud Phase Detection based on BP Neural Network

To improve the image quality of band 5 and band 27 which contain stripe noises acquired by Moderate Resolution Imaging Spectroradiometer (MODIS) level_1B,based on MODIS scanning characteristics,a method of using the max mean of each swath to judge the stripe noises was proposed.When destriping noises,according to the thought of single line stripe interpolation on band 5,an interpolate method of using the adjacent multi\|line stripe noises on band 27 was proposed.Finally,comparison diagram,mean diagram and numeric analysis between original data and processed data were compared to validate the effect of destriping noises.The results show that the method can judge all the stripe noises exactly on both bands,and can remove the stripe noises well.The process of destriping noises is easily and suitable for the complex remote sensing scenes.