Abstract: To standardize miners’ behaviors in coal mines and prevent production accidents, this paper establishes a violation behavior recognition model for underground miners based on a Convolutional Neural Network. To further evaluate the performance of the CNN model in identifying miners’ non-compliant behaviors, a comparative analysis is conducted with a traditional Support Vector Machine based recognition model. The comparison aims to validate the reliability and accuracy of the CNN model in detecting violation behaviors in underground coal mining environments. Experimental results demonstrate that the CNN recognition model outperforms the SVM approach in efficiency, achieving an average accuracy of 96.15%. Compared to the SVM model, the average accuracy and operational efficiency of the CNN model are improved by 4.31% and 3.1 seconds.