Aiming at the problem that feature extraction stage of FSRCNN algorithm use only one convolution layer

leading to inaccurate feature extraction

change the network structure

increase another convolution layer for feature extraction. Aiming at placing the deconvolution layer at the rear of network model

the FSRCNN algorithm produces randomness in the process of image super-resolution reconstruction. The network structure is changed

and the deconvolution layer is placed in front of the feature extraction layer

so that the feature extraction is performed on the enlarged reconstructed image during the training process. After a large number of comparative experiments

it is proved that the FSRCNN model with improved network structure increases the complexity of the algorithm to a certain extent but improves the quality of image reconstruction. The problem of image super-resolution reconstruction can be better handled by using improved FSRCNN with the support of a certain hardware platform.