Lightweight Real-Time Intelligent Inspection System for Digital Transmission Security

The whole process automation of patrol inspection can effectively improve the management and production efficiency of current digital transmission business. The rapidly developed image super-resolution (SR) technology is helpful to achieve this goal. Therefore, we specially design a lightweight effective multi-level dual residual attention network (MDRAN) for single image super-resolution network (SISR). Firstly, to fully extract and utilize the dependency information between different channels, a dual residual attention block (DRAB) is proposed. This module can adaptively adjust the weight proportion between features across various channels, so as to accurately extract high-frequency information containing rich details and texture information. Meanwhile, to maximize the usage of the characteristics from various levels, the DRAB is cascaded combining with residual skip connection, which can not only reduce the loss of information as the network depth increases, but also ease the difficulty of network training. The experimental results on multiple benchmark datasets show that the PSNR and SSIM indicators of MDRAN are the highest among the comparison models so as to the greatly reduced amount of model parameters. Furthermore, the visual effect of MDRAN is richer in texture and border.