Research On Remote Sensing Image Compression Based On Wavelet Transform And Neural Network

With the rapid development of computer technology and earth observation technology,remote sensing images have been widely used in military reconnaissance,food production estimation,resource detection,and disaster prevention.However,the speed of updating remote sensing images and the continuous improvement of resolution make the memory occupied by images increase rapidly.To solve this problem,it is necessary to perform more efficient compression processing on the images.Wavelet analysis,as a representative of the emerging mathematics analysis,has good time-frequency localization capability and can be embodied in human vision.Therefore,it is widely used in image compression.In order to obtain a higher compression ratio and compression quality in the compression process and have the advantage of having wavelets at the same time,wavelet transforms are usual y combined with other compression coding.Neural network is also one of the current research hotspots.Due to its adaptability,fault tolerance and strong learning ability,highly parallel processing capability and artificial intel igence associative memory,many classical network models can be effective.Applied in video compression coding.How to combine the wavelet transform and the neural network in the image compression process so as to have the common advantages of the wavelet and the neural network at the same time has been a subject of great concern and research.This article takes 512*512 regions around Jiangxi University of Science and Technology as research objects,and carries out zero tree EZW encoding and multi-tree SPIHT encoding respectively.The BP neural network and the encoding method of this article are respectively compressed,and their size after compression,PSNR(peak SNR),MSE(mean squared error),and compression time are compared and analyzed.In this paper,the wavelet transform and the neural network are used to assist the coding.Firstly,the image to be compressed is subjected to lifting wavelet decomposition,the low frequency coefficients obtained by image decomposition are retained,the high frequency coefficients are encoded by the neural network vector quantization,and final y the inverse transformation is performed.The reconstructed highfrequency coefficients and retained low-frequency coefficients are used for image reconstruction.(1)Using EZW coding compression ratio,PSNR,MSE and compression time are respectively 8.4,25 d B,210,7S.The values after compression using SPIHT encoding are 11.7,24 d B,207,and 11 S,respectively.It can be seen from the results.SPIHT is still slightly better than EZW in compression when the compression ratio is greater than the EZW algorithm.In the compressed reconstructed image,it can be clearly seen that the SPIHT compression is less than the EZW image distortion and the box effect.SPIHT constructs a different type of zerotree collection.This setting makes the encoding more flexible.(2)The compression ratio,PSNR,MSE,and compression time of BP neural network were 9.7,22 d B,220,and 30 S,respectively.The methods used in this paper were 14,29.4d B,74.5,and 25 S,respectively.Using this method combined with wavelet and neural network not only improves compression coding of neural networks in compression ratio and compression time,but also achieves better compression quality.(3)Comparing the coding method used in this paper with the wavelet-based coding EZW and SPIHT,the compression method of this paper is better than the single wavelet transform in the case of higher compression rate,and the visual effect after decompression is better,and the peak value at the same time The signal-to-noise ratio is higher and the mean-square error is lower,that is,the compression quality is better,but the compression time is longer due to the complexity of the algorithm.