In order to improve the resolution and detection ability of low-illuminance multi-band spectral density imag-es,a super-resolution reconstruction method based on optimized deep learning is proposed.Firstly,the super-resolu-tion feature sampling model of low-illumination multi-band spectral density image is constructed,and the vector pixel reconstruction of low-illumination image is realized by image compression sensing method.Secondly,the noise reduc-tion filtering of low-illumination image is carried out by fuzzy degree identification and matching filtering method,and the compression spectral dimension detection model of low-illumination multi-band spectral density image is construct-ed.Through image denoising,compression reconstruction and spectral feature recombination,a regularization con-straint model is established to recover the spectral information of the image.Finally,according to the correlation fea-ture distribution of all spectral data in the same space region,an optimized deep learning algorithm is adopted to real-ize the feature allocation and structure recombination of the low-illumination image,and the super-resolution recon-struction of the low-illumination image is realized.In super-resolution reconstruction of low-illumination images,this method can complete the details of the image,and its denoising and defuzzing capabilities are good,and the key infor-mation of the image can be effectively retained.The signal-to-noise ratio is 26 dB,and the structural similarity is higher than 0.94,which is superior to the comparison method,and has good application value.
optimizing deep learninglow illumination imagesuper-resolution reconstructionimage denoising
NETL
NSTL
万方数据
