Application of Image Classification Based on Improved LSTM in Internet Reading Therapy Platform

Abstract

Reading therapy is an effective approach for improving mental states or addressing disabilities associated with individuals’ dyslexia. In traditional approaches, this process is performed through the intervention and supervision of an expert, which incurs time and cost. However, by utilizing artificial intelligence technologies, the reading therapy process can be automated. This article focuses on presenting an internet-based automated platform for reading therapy. In this method, audio and visual features during the reading therapy are processed using deep learning techniques to identify the individual’s emotional state based on their reading status. In this state, two separate convolutional neural networks are used to describe the facial image features and speech characteristics of the individual. Then, the described features from these two models are merged to determine the individual’s mental states using LSTM layers. Finally, a reinforcement learning model is used to provide feedback and design subsequent exercises. This reinforcement approach leads to continuous improvement of the evaluated process and plays a significant role in enhancing the efficiency of the internet-based reading therapy system. The performance of the proposed method has been evaluated based on information from 20 volunteers. According to the results, the proposed method can effectively improve individuals’ mental states and compete with the conventional supervisor-based approaches. The performance of the proposed deep learning models in identifying emotional states has also been investigated. The results indicate that this model achieves a minimum improvement of 9.71% in emotional state recognition compared to previous research, with an average correlation coefficient of 0.64485.

Keywords

• Internet-based reading therapy
• reading therapy framework
• deep learning
• fuzzy logic
• reinforcement learning