Research on Portable Intelligent System Based on Lightweight Super-Resolved Image Recognition Algorithm

Recently, with the increase of business demand, the whole business field of digital transmission needs to accelerate the construction of a one-stop data collaborative application system, in which the portable intelligent system based on image super-resolution recognition technology is an important part of the system. However, it is too expensive to build and improve the performance of this portable intelligent system in terms of hardware. Remarkably, the image super-resolution technology based on deep convolution neural network has made outstanding progress and dominated the current research on super-resolution technology. However, the improvement of performance is often at the cost of a sharp increase in the number of parameters, which limits the practical application of super-resolution methods. Therefore, we design a lightweight dual-path attention network (LDAN) for single image super-resolution. Specifically, a dual-path attention block (DAB) is carefully designed for assigning more weights to high-frequency information. LDAN is constructed in the manner of stacking several DABs combining global residual skip connection. The experiment demonstrates that compared with the lightweight methods, i.e., FSRCNN, DRRN, LapSRN, and IDN, the proposed LDAN greatly reduces the number of parameters, while the qualitative and quantitative results of the super-resolved images are significantly better.