Abstract: The underground environment is harsh,with dim lighting and high levels of dust and moisture,resulting in poor image quality for safety monitoring.To improve the clarity of underground safety monitoring images,an image enhancement al gorithm based on a multi-branch attention convolutional neural network is proposed.The algorithm consists of an attention sub network,noise subnetwork,enhancement subnetwork,and reinforcement subnetwork.Model compression is achieved through pruning and quantization,while adaptive adjustment of algorithm parameters is enabled by brightness and noise monitoring modules.Additionally,multimodal fusion techniques are employed to optimize algorithm performance.Results show that the pro posed algorithm has an average running time of 0.056 seconds,with an average Peak Signal to Noise Ratio (PSNR) of 32.56 dB,an average Structural Similarity Index (SSIM) of 0.91,and an average Mean Absolute Error (MAE) of 0.043.In image enhancement tasks for suspected gas leakage areas,equipment failure zones,and human facial regions,the proposed algorithm significantly outperforms both the Multi-Scale Retinex with Color Restoration (MSRCR) algorithm and traditional Convolutional Neural Network Image Enhancement (CNN-IE) algorithms in terms of fPSNR ,fSSIM ,and fMAE .In practical applications,the algo rithm achieves PSNR values above 33 dB across various scenarios,fSSIM ranging from 0.92 to 0.96,and fMAE below 0.025.The study demonstrates that this algorithm effectively enhances image quality,meets real-time requirements,provides robust techni cal support for mine safety monitoring,and contributes to improving the safety level of mining operations.

Key words: underground safety monitoring,image enhancement algorithm,multi-branch attention,adaptive parameter ad justment,multi-modal fusion