Dense Residual Network for X-ray Images Super-Resolution

In view of the information loss problem caused by image compression, data packet loss and other reasons during X-ray nondestructive testing, we proposed a novel dense residual network to further improve the quality of steel cord conveyor belt X-ray images reconstruction. Compared with previous works, the special connections provide multiple short paths for the forward propagation to improve reconstruction accuracy and for the backward propagation of gradient flow to accelerate the convergence speed. Extensive experiments confirm the effectiveness of our dense residual connections. Both subjective visual effect and objective quantitative evaluation are better than bicubic interpolation and super-resolution convolutional neural network (SRCNN). Especially for execution time, our method only takes 0.142s and it is fast enough for real-time application. Moreover, our method can be extended to natural images reconstruction.