No-reference video quality assessment for underground drilling sites based on spatiotemporal domain dynamic aggregation

No-Reference Video Quality Assessment (NRVQA) is a key technique for evaluating the video quality of underground drilling sites in coal mines and enabling remote monitoring. Existing NRVQA methods are mostly designed for general ground scenes and are difficult to achieve satisfactory performance in underground drilling environments where composite image distortions are caused by coal dust and equipment vibration. To address this problem, an NRVQA method for underground drilling sites based on spatiotemporal domain dynamic aggregation was proposed. Video features of drilling site surveillance videos were extracted from two dimensions, namely spatial and motion. The spatial feature extraction branch was based on the Swin Transformer architecture and introduced a local perception enhancement module to strengthen the representation capability of texture and edge details under coal dust interference. The motion feature extraction branch embedded a DeformConv3D deformable convolution module into ResNet to accurately capture the dynamic characteristics of drilling rig motion trajectories and coal dust diffusion. A spatiotemporal dynamic aggregation module was designed to dynamically allocate the weights of spatial and motion features, enabling discriminative representation of different distortion types and degrees. The Coal-DB dataset was constructed and ablation experiments and comparative experiments were conducted. The results showed that the proposed method achieved Spearman rank correlation coefficient, Pearson linear correlation coefficient, Kendall rank correlation coefficient, and root mean square error values of 0.904 3, 0.902 3, 0.753 6, and 4.684 0, respectively, which were superior to the baseline model and mainstream video quality assessment methods such as VSFA and StableVQA. The predicted video quality scores of this method were closer to the subjective scores.