An Efficient Semantic Segmentation Model for Coal Dust Particle Characterization via Deep Super-Resolution

The particle size distribution (PSD) of coal dust measurement is effective means of preventing coal dust hazards. To address the issues of efficiency and accuracy present in traditional coal dust detection, this study proposes a lightweight semantic segmentation model based on deep super-resolution reconstruction to extract detailed features from coal dust samples during the mining process. The model uses partial convolution (PConv) to extract particle features and incorporates a multiscale attention mechanism to emphasize particle details. A deep super-resolution reconstruction strategy is introduced to capture edge features of the particles, while a quadtree decomposition localization module recursively divides the image into regions to effectively separate overlapping coal dust particles, thereby improving segmentation accuracy. Experimental results show that the particle recognition accuracy $({O}_{Z})$ under different fractal dimensions reaches 96.23%, providing an efficient and accurate solution for coal dust detection in coal mine safety video monitoring.