航天光学遥感器信噪比的人工神经网络评价

Assessment of signal-to-noise ratio of space optical remote sensor using artificial neural network

提出基于人工神经网络进行航天光学遥感器信噪比评价的方法,首先对航天遥感图像进行分析,从图像中将与景物结构和噪声有关的特征向量分别提取出来,作为ANN的输入。网络通过对大量信噪比已知的图像样本训练后,可完成对航天光学遥感器传输下来的任意一幅地面景物图像进行系统的信噪比测试,从而避免了采用特定景物目标进行测量中的诸多弊端,测量平均误差低于10%。

On the basis of artificial neural network （ANN）, a new method to assess the signal-to- noise ratio （SNR） of space optical remote sensor is proposed. Through analyzing the images of space remote sensor, the eigenvectors related to the structure of landscape and noise were abstracted respectively, and then these eigenvectors were used as the input of ANN. After being trained with simulated images whose SNR were known, the ANN could assess the SNR of unknown images. This method can avoid the defects that special views were needed, and the mean assessment error is less than 10%.