SMC-SRGAN-Lightning super-resolution algorithm based on optical micro-scanning thermal microscope image

Due to the low spatial resolution of the existing optical micro-scanning thermal microscope imaging system, the acquired micro-scanning infrared images have inferior image quality and low contrast. Deep learning methods, represented by SRGAN, have shown promising results in super-resolution. However, this method still has artifacts, blurriness, low spatial resolution, and slow reconstruction speed. Therefore, we propose the SMC-SRGAN-Lightning super-resolution algorithm based on optical micro-scanning thermal microscope images in this study. Firstly, we enhance the model's attention to features and improve the details and clarity of the reconstructed images. Removing the BN layer in residual blocks, replacing the ReLU with SMU, and introducing the CBAM to construct the SMC module. Secondly, we incorporate the attention mechanism SEnet into the Bottleneck structure of MobileNetV2. Reducing the channels in the first 1 x 1 convolution layer to 1/4 and creating the SE-MobileNetV2 module. It can enhance the model's focus on essential features, computational efficiency, and accuracy. Finally, to validate the effectiveness of our method, we compare it with four other super-resolution algorithms on public datasets and images obtained from the optical micro-scanning thermal microscope imaging system. Experimental results indicate that our method improves image clarity, preserving details, and textures. Comprehensively considering super-resolved image quality and time costs, our method is superior to other methods.