Image Restoration For CCSDS Image Lossy Compression Based On Nerual Network

With the development of space technology,people obtain more and more information from the universe,and images play an important role as carriers of information.However,due to factors such as spacecraft storage capacity and limited downlink data,for the sake of obtaining more information,the original image needs to be compressed.The CCSDS lossy compression algorithm is widely used due to its simple implementation and good compression effect.However,with increasing compression ratio,the compressive image will be seriously degraded and the quality will be significantly reduced,which will not meet the requirements for using.For this reason,this paper focuses on CCSDS lossy compression image,trying to find recovery algorithms to recover the lossy images caused by compression so that high-loss lossy compressed images can be used.In this paper,firstly,according to directly observing and analyzing the residual values of the original image and the degraded image,it Compare the CCSDS lossy compression degraded image with the motion degraded and the Gaussian degraded image.It is determined that the CCSDS lossy compression degradation is closer to the Gaussian blur degeneration.Therefore,this paper chooses the BP neural network algorithm which is commonly used in Gaussian degraded image restoration to perform CCSDS lossy compression image.The experimental verification shows that the BP neural network based on Adam learning algorithm has a certain recovery effect on the lossy compressed degraded images.On the basis of this,this paper selects BEGA(Bee Evolutionary Genetic Algorithm)and MEA(Mind Evolutionary Algorithm)to further optimize the BP neural network.Because of the fast convergence of the previous BP iterations,we optimize the algorithm with it.The experimental results were evaluated from both subjective and objective evaluation:subjective evaluation criteria considered that the optimized MEA was basically the same as the traditional MEA and better than the BEGA;the objective evaluation criteria selected Peak Signal to Noise Ratio(PSNR)and structural similarity index(SSIM).It is considered that the optimized MEA-BP algorithm is obviously better than other algorithms in terms of PSNR or SSIM.It is demonstrated by experiments that the optimized MEA-BP has a beter recovery effect in CCSDS lossy compression degraded images.