Journal of Multimedia Processing and Technologies Sun Lifang, Liu Yuan 16 1 2025 https://doi.org/10.6025/jmpt/2025/16/1/28-35 https://www.dline.info/jmpt/fulltext/v16n1/jmptv16n1_4.pdf Using a sliding window combined with heterogeneous multi-column convolutional neural networks, we found that this new evaluation tool has higher accuracy than traditional evaluation tools and can provide more comprehensive evaluation information. This new evaluation tool can effectively improve teaching efficiency and provide more comprehensive evaluations. Advanced computer vision analysis technology can quickly collect and process valuable classroom information and convey it to teachers and peers in real-time, effectively motivating their participation and promoting their learning, ultimately achieving the best educational outcomes.