Dysgraphia, the shakes on the brain, impacts the children when it comes to learning how to write the typical way; children just won’t be able to know the traditional writing curriculum, and thus, written expression suffers. This weakness in writing leaves a student at an academic disadvantage, as well as their confidence. To address this, a new handwritten task data set was developed in this paper, and a wide variety of features were chosen to incorporate as much handwriting information as possible. The proposed R2CNN method integrates a multitask refinement network for high-quality detection of inclined boxes and a Text RPN to predict candidate text regions. We use a balance parameter for the loss function to solve the problem of the unbalanced training categories and prevent overfitting due to feature selection. This study aimed to discriminate dysgraphia using these extracted features involving handwriting and geometry. The feature learning phase of deep transfer learning makes it possible to extract and transfer key features to detect dysgraphia. Finally, to obtain better detection accuracy of dysgraphia, we apply the Enhanced Adaptive Butterfly Optimization Algorithm (EABOA) to optimize the model parameters. As the results showed, the handwritten images can help detect dysgraphia in children in this work. The overall proposed approach achieves an accuracy 99.2%, precision 95.3%, recall 99.1% and F1-score 97.16%, respectively. The results of the data collection process indicate that handwritten text samples may be used in this study to determine whether someone has dysgraphia.
Balamurugan et al. (Wed,) studied this question.