Picture and video storage method based on deep learning

The invention belongs to the technical field of picture and video storage, and particularly relates to a picture and video storage method based on deep learning, which comprises the following steps: firstly, writing picture and video data into a storage system, and attaching a corresponding label to the picture and video data; secondly, the storage system performs periodic sampling on the picture video data after the label is attached, generates an SRGAN Net Generator model, and performs model training and storage through the sampled data; carrying out compression processing on the picture video data in the sampling range; carrying out super-division conversion on the compressed picture video data through the trained model; and finally, carrying out structural similarity evaluation on the picture video data after super-division conversion, deleting the original picture video when the structural similarity evaluation meets requirements, and otherwise, keeping the original picture video. The storage system can