基于映射空间编码的高速运动轨道图像去模糊研究OA北大核心CSTPCD

This paper proposed a high-speed motion railway image deblurring algorithm based on Maximum A Posteriori(MAP)estimation and mapping space encoding,aiming to address the problem of image blurring caused by lens jitter or rapid camera movements in the railway defect detection system.First,the algorithm employed deep encoder-decoder and residual networks to encode the mapping relationship between clear and blurred images in the dataset,as well as the blur kernel.To preserve the frequency information during encoding,the algorithm introduced a dual-channel residual network with fast Fourier transform(FFT)channels on the traditional residual modules to compensate for the frequency loss caused by multiple feature extractions.Second,the algorithm utilized the Deep Image Prior(DIP)to parameterize the latent clear image and blur kernel,and subsequently leveraged the obtained prior blur kernel and clear image to invoke the mapping relationship within the encoding space.Finally,through alternating optimization of the latent clear image and blur kernel,the algorithm approximated an unknown mapping to achieve the deblurring of high-speed motion railway images in real scenes.The experimental results indicate that the frequency component intensity of the feature maps extracted by the dual-channel residual module is generally higher than that of the traditional residual module.Compared to implementing the algorithm using a traditional residual module,adopting a dual-channel residual module can result in an increase of 0.84 dB in Peak Signal-to-Noise Ratio(PSNR)and an improvement of 0.025 1 in Structural Similarity(SSIM).Compared with existing deep learning deblurring algorithms,the proposed deblurring algorithm shows superior deblurring performance for images acquired by the high-speed railway detection system,achieving an improvement of 1.84 dB in PSNR and 0.017 3 in SSIM over the best deblurring algorithm and significantly enhancing the quality of the acquired images.The proposed method can provide clear images for the next step to identify whether there are defects in the railway components.