CMA-Net: Multiscale Cascading Interaction and Dual-Path Fusion for Super-Resolution Reconstruction in Coal Mine IoVT

The perception reliability of Internet of Video Things (IoVT) systems in underground coal mine environments is critically compromised by the compounded effects of uneven illumination, coal dust occlusion, and multi-scale degradation, which collectively induce severe structural and informational distortions. To address these challenges, we propose CMA-Net, a lightweight Super-Resolution(SR) network based on multi-scale cascaded cross-fusion and dual-path feature modulation. A cascaded cross fusion module is designed as a feature refinement branch to mitigate uneven lighting and enhance the representation of key information. An architecture comprising a multiscale feature refinement module and adaptive local and global modulation networks is designed as another branch, which not only enables adaptive feature enhancement in dusty areas but also effectively enhances edge details in multiscale features. Finally, the method is optimized using a Mean Absolute Error loss function, which helps preserve the accuracy of pixel-level structures. In the ×2 Super-Resolution task, the method achieved a PSNR of 36.3334 and an SSIM of 0.9771 on the CMUID test set,and a PSNR of 33.9999 and an SSIM of 0.9407 on the Set5 test set. CMA-Net contains only 0.78M parameters, providing a model foundation for deployment on resource-constrained IoVT edge devices. It enhances the perception capabilities of smart mines and provides robust support for accurate object detection.