Study On The Deformation Process Of Deep Supercooled Vacuum Melt Based On Image Sequence Super-resolution Reconstruction

Deep Supercooling is the process of cooling a liquid or gas by inhibiting nucleation,so that it will not turn solid when the temperature has dropped below its point of freezing.The subject of this study is the metal droplets in this process.Deep supercooling melt deformation in microgravity environment is an important research content in material science.Due to the limitation of the falling tube height and the large falling speed of the object,the high-speed camera equipped with the simulated environment can also obtain the low-resolution sequence images of the experimental materials.In order to accurately establish the deformation process of deep supercooled melt,super-resolution reconstruction using the obtained low-resolution sequence images is a widely used research method.Image super-resolution reconstruction has been widely used in remote sensing,medicine and other fields.In recent years,due to the rise of deep learning research and the successful application of convolutional neural network in image field,the super-resolution reconstruction technology based on deep learning has also made great progress.The super-resolution reconstruction of a single frame image is limited by the limited information contained in a single image,and the reconstruction effect is not ideal.However,the current mainstream super-resolution reconstruction algorithm for multi-frame images focuses on the pursuit of high PSNR(Peak Signal to Noise Ratio),so that the reconstructed image is relatively smooth,and the reconstruction result lacks a lot of high-frequency details,which is not conducive to the study of deformation texture details of deep supercooled melt.To solve these problems,a new super-resolution reconstruction model based on WGAN(Wasserstein Generative Adversarial Network)is proposed in this paper.Compared with the current super-resolution reconstruction algorithms for advanced neural networks,such as SRCNN,SRGAN,VESPCN,etc.,the reconstruction results not only improve the PSNR and SSIM index,but also present better detail texture.Compared with the single-frame algorithm SRGAN,while maintaining the high frequency details in the model in this paper,the reconstruction result PSNR is improved by about 1.5db,and the SSIM is improved by about 0.1.Compared with other single-frame algorithms,when PSNR is basically close,SSIM also increases by about 0.1.Compared with the current advanced multi-frame algorithm,the reconstruction results of the model in this paper have been improved to varying degrees in PSNR and SSIM indexes.Moreover,after local amplification,the high-frequency detail reconstruction is better with obvious improvement.Through the test of the open data set,the model proposed in this paper has effectively acquired the inter-frame information.Under the condition that the indicators of PSNR and SSIM have been improved,better high-frequency detail images have been reconstructed than the current advanced multi-frame algorithm.For the deep supercooled melt image of this subject,this model still achieved good results.