Abstract
Aiming at the problem of getting low resolution image and blurred image during unmanned aerial vehicle (UAV) inspection, a super-resolution reconstruction algorithm based on very deep network super resolution (VDSR) is proposed. The algorithm model is composed of deep convolution neural network and residual structure, which is improved on the basis of super resolution convolutional neural network (SRCNN). By deepening the network to 20 layers, the receptive field can be expanded, and the residual structure can be combined to obtain better reconstruction effect. The experimental results show that the proposed super-resolution method based on VDSR has richer texture and more realistic visual effect, with 2.95 dB and 0.037 improvement in PSNR and SSIM compared with the super-resolution methods based on Bicubic Interpolation, Sparse Coding and SRCNN. The proposed algorithm further promotes the theoretical research of inspection image super-resolution, and can be effectively applied to the practical application of power inspection.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call