What does this research mean for the field?

A hybrid hierarchical deep learning framework utilizing multi-level feature fusion achieves 96.8% accuracy in detecting learning disabilities from handwriting images, outperforming standard baseline models. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This research aims to create a deep learning framework that identifies learning disabilities by analyzing handwritten text images.

May 9, 20260 citationsOpen Access

A Hybrid Hierarchical Deep Learning Framework with Multi-Level Feature Fusion for Handwriting-Based Learning Disability Detection

KSK Soumya PCP Joseph Charles SKS Kavitha

Key Points

This research aims to create a deep learning framework that identifies learning disabilities by analyzing handwritten text images.
Integrated ResNet50 and InceptionV3 with a custom CNN for multi-level feature fusion.
Preprocessed and augmented 138,500 handwriting samples for training and evaluation.
Utilized a Softmax-based meta-classifier for final feature classification.
Achieved an accuracy of 96.8% in classifying images across normal and various handwriting styles.
Outperformed baseline models: AlexNet (88.6%), VGG16 (91.2%), InceptionV3 (94.3%), ResNet50 (95.1%).

Abstract

Abstract Objectives: To develop a hybrid hierarchical deep learning framework aimed at easing early identification of learning disabilities based on analysis of images of handwritten text. Method: The proposed framework integrates the transfer learning models ResNet50 and InceptionV3 with a custom-designed convolutional neural network (CNN) via a multi-level feature fusion strategy. After preprocessing, normalization, augmentation, and learning of handwriting images with different architectures, the feature vectors are fused and classified using a Softmax-based meta-classifier. Findings: It has been found that the classification model was evaluated on 138500 handwriting samples from the Normal, Reverse and Corrected handwriting classes with an accuracy of 96.8%. It was then compared against baseline models such as the AlexNet (88.6%), VGG16 (91.2%), InceptionV3 (94.3%) and ResNet50 (95.1%). Novelty: This study proposes a hybrid hierarchical architecture for the detection of handwriting features related to learning disabilities via multi-level feature fusion, stacked ensemble learning, and Grad-CAM interpretability. Keywords: Learning Disability Detection, Handwriting Analysis, Deep Learning, Transfer Learning, Feature Fusion, Hybrid CNN Framework

اسأل الذكاء الاصطناعي

Bookmark

View Full Paper