A Deep Learning Model for Kannada Handwritten Optical Character Recognition using MobileNetV2 and CNN

Handwritten character recognition remains a complex and vital problem in the domains of computer vision and pattern recognition, particularly for regional languages such as Kannada, which feature intricate character structures and varying handwriting styles. This study presents a deep learning–based approach for the recognition of handwritten Kannada characters using a transfer learning architecture built upon MobileNetV2. The model uses custom dense layers for character classification after extracting high-level visual features using pretrained ImageNet weights. To enhance generalization, a large dataset of handwritten Kannada characters was used, preprocessed using data augmentation and normalization techniques. To address data imbalance, class weighting and adaptive learning techniques were used to train and optimize the model. The suggested MobileNetV2-based model outperforms traditional Convolutional Neural Networks (CNN) techniques in terms of recognition performance and training efficiency, as evidenced by evaluation metrics such as accuracy, precision, recall, and F1-score.