VCON: Visual Convolutional Optimization Network for Fuzzy Comprehensive Evaluation in Japanese Language Teaching Quality

Traditional methods of teaching quality assessment often face challenges, such as subjectivity, limited evaluation dimensions, and an inability to capture teaching details accurately when confronted with complex and dynamic classroom environments and individual student differences. To address these challenges, this paper proposes a visual convolutional optimization network for fuzzy comprehensive evaluation in Japanese language teaching quality. First, dynamic key-frame extraction is achieved by calculating pixel-level differences between video frames and setting reasonable thresholds, effectively capturing key teaching information such as student engagement and comprehension from Japanese teaching videos. Second, the method improves a convolutional neural network (CNN) through up-sampling and feature fusion techniques to extract feature information from key frames, capturing students’ facial expressions, body language, and interactions with teachers. These features are then classified into multiple evaluation indicators, such as student engagement, teaching ability, and classroom atmosphere. Finally, the fuzzy comprehensive evaluation method is employed, combining weights determined by the analytic hierarchy process (AHP) and a membership matrix based on CNN feature extraction, to provide a comprehensive, objective, and accurate evaluation of Japanese teaching quality. Experimental results demonstrate that the proposed method achieves an accuracy rate of 98.1% in evaluating Japanese teaching quality, surpassing state-of-the-art methods and indicating its good effectiveness and practical value.