Handwritten Gujarati Numerals Classification Based on Deep Convolution Neural Networks Using Transfer Learning Scenarios

In recent years, handwritten numeral classification has achieved remarkable attention in the field of computer vision. Handwritten numbers are difficult to recognize due different writing styles individuals. a multilingual country like India, negligible research attempts have been carried out for Gujarati numerals recognition using deep learning techniques compared other regional scripts. The digit dataset is not available publicly and requires large amount labeled data training models. If number annotated sufficient enough train Convolutional Neural Networks (CNN) from scratch, transfer can be applied. However, issue arises by that how fine-tune pre-trained convolutional neural network while target model. this paper, we addressed these problems three scenarios classify images zero nine. We presented ten CNN architectures including LeNet, VGG16, InceptionV3, ResNet50, Xception, ResNet101, MobileNet, MobileNetV2, DenseNet169 EfficientNetV2S find best performing model freezing fine-tuning weight parameters. implemented models self-created with 8000 nine digits augmentation. Exhaustive experiments performed various performance evaluation matrices. showed promising results among all 98.39% accuracy, 97.92% testing 97.69% f1-score, 97.15% AUC. Our on https://github.com/Parth-Goel/gujarati-handwritten-digit-dataset/ .