Super-resolution reconstruction of unmanned aerial vehicle image based on deep learning

Abstract

Due to the influence of weather, air flow and other environmental factors, unmanned aerial vehicle (UAV) aerial photography often leads to problems such as blur, defocus and so on. In order to solve the problem of low definition of UAV aerial images, this paper proposes a super-resolution reconstruction algorithm based on improved residual learning blocks (IRB) to reconstruct aerial images to achieve high definition. This method can effectively solve the problem of gradient explosion which is often encountered in deep learning network. At the same time, it can suppress the learning of useless information and make full use of important feature information. Through the establishment of UAV aerial data set for cross training of the network, the network can better adapt to the UAV aerial environment. The experimental results show that, compared with the existing neural network algorithm, the improved residual network algorithm can reconstruct UAV aerial image well, and the subjective visual effect of the reconstructed image is better while the edge information is greatly preserved.