Application of Convolutional Neural Network Integrating Attention Mechanism in Detecting Negative Behavior of College Students in Classroom
Lin Deng · 2024
With the rapid development of artificial intelligence technology, the application of Convolutional Neural Networks (CNNs) that incorporate attention mechanisms is increasingly becoming a research hotspot in various fields. This study mainly explores the design and implementation of CNN models with integrated attention mechanisms for detecting negative behaviors in university classroom settings. The research context is based on the prevalent issue of student inattentiveness during classes, which includes behaviors such as using mobile phones, dozing off, whispering with peers, etc. These behaviors not only affect the learning outcomes of students themselves but also disrupt the order of classroom teaching. To foster a positive learning attitude among students, real-time monitoring and timely correction of negative behaviors in the classroom are particularly crucial. Addressing this issue, the study constructed a CNN model integrating both spatial and temporal attention mechanisms, which can effectively extract the spatiotemporal features of student behavior and identify negative behaviors more accurately. The study utilized video data of 300 students from three different majors at a university during classroom sessions, creating training and testing datasets after data augmentation and preprocessing. During the model training process, adaptive weight adjustment strategies and multi-layer feature fusion techniques were introduced to effectively solve the problems of class imbalance and insufficient feature representation. The experimental results showed that the accuracy of the model reached 92.7%, which is an 8.3 percentage point improvement over the commonly applied learning analytics algorithms. Moreover, the model not only has a high recognition rate but also performs excellently in terms of real-time detection of negative behaviors, with an average response time of no more than 300 milliseconds, significantly superior to the current widely used behavior recognition methods. Therefore, the model proposed in this study greatly enhances the efficiency and accuracy of classroom negative behavior detection, holding significant theoretical implications and practical value for effectively managing classroom order and optimizing the educational environment.