基于深度学习的毫米波和亚毫米波成像仪的图像增强技术

风云四号卫星毫米波和亚毫米波成像仪(MMSI)数据根据采样方式分为过采样和非过采样数据。由于采样方式的影响,非过采样数据在采样过程中会有一定的信息损失。为解决采用简单的线性插值方法做精细化处理时提升精度有限问题,采用基于深度学习的方法增强MMSI亮温图像,设计卷积神经网络重建风云四号卫星MMSI的亮温图像和风云三号卫星微波成像仪亮温图像。实验结果显示:相比传统的双三次插值方法,在风云三号卫星微波成像仪亮温图像样本上峰值信噪比提升了1.13dB,结构相似度提升了0.01。实验结果表明:对于非过采样亮温数据,采用基于深度学习的方法增强图像具有更高的精度,同时可在其他微波探测仪数据中使用,具有很强的普适性。

Image Enhancement Technique of Millimeter and Sub-MillimeterSounding/Imager Based on Deep Learning

FY-4 millimeter and sub-millimeter sounding/imager (MMSI) data are divided into oversampling and non-oversampling data according to the sampling model. However, non-oversampling data have a certain information loss during the sampling process. The traditional method employs the linear interpolation to perform the fine processing, but the accuracy improvement is limited. In this paper, a method based on deep learning and convolution neural network is used to enhance the brightness temperature images of FY-4 MMSI and FY-3 microwave imager. The experimental results show that the peak signal-to-noise ratio is increased by 1.13 dB and the structure similarity is increased by 0.01 for FY-3 microwave imager, compared with the traditional bicubic interpolation method. The experimental results indicate that the image enhancement based on the deep learning method is more accurate than that based on the bicubic method. It can also be applied to other microwave detector data due to strong universality.