Dysgraphia Disorder Detection and Classification Using Deep Learning Technique

Dysgraphia, a neurological condition, impedes children’s acquisition of standard writing abilities, leading to subpar written expression. Inadequate or underdeveloped writing proficiency can adversely affect a child’s educational progress and self-esteem. To address this issue, our study involved compiling a novel dataset of handwritten operations and extracting an array of features to encapsulate the various dimensions of handwriting characteristics. This research presents the Rotational Region Convolutional Neural Network (R2CNN) as a novel approach to tackle this issue. The R2CNN framework integrates a multitask refinement network for accurate tilted box detection and a text region proposal network (Text RPN) to identify potential text areas. To address the imbalance in the training categories and mitigate the overpopulation problem through feature elimination, a balance parameter is incorporated into the loss function. This research focused on identifying dysgraphia by analyzing these extracted features, which included both handwriting and geometric elements. The feature-learning stage of deep transfer learning effectively extracts and applies characteristics to identify dysgraphia. Research findings indicate that this study can use handwritten images to detect dysgraphia in children. The results of the data-gathering process show that this investigation can leverage samples of handwritten text to recognize dysgraphia among young individuals.