Research on UAV remote sensing multispectral image compression based on CNN

Abstract

UAV remote sensing multispectral image has become more and more popular because of its high temporal and spatial resolution. However, multispectral image has the characteristics of large number of bands, large amount of data, spatial and spectral redundancy. These characteristics bring great challenges to image storage and transmission. According to the characteristics of multispectral images, an end-to-end multispectral image compression framework based on CNN is adopted. The whole framework is composed of self-encoder, quantization structure, entropy coding and rate distortion optimization. The innovation of this paper is to propose a new multi-source data preprocessing method, which uniformly converts the DN value of multispectral image into reflectivity, and the multispectral image compression framework uses 1 * 1 convolution to reduce the inter spectral redundancy of the image, self-encoder to reduce the dimension of the image, Gaussian mixture entropy coding to estimate the code rate, rate distortion optimization to jointly optimize the code rate and distortion. The experimental results show that under the same bit rate, the image compression effect of this method model is significantly better than the traditional image compression method, and the quality of the reconstructed image is significantly improved.