A Two-Stage Deep Feature Selection Method for Online Handwritten Bangla and Devanagari Basic Character Recognition

In this paper, we have proposed a two-stage deep feature selection (FS) approach for the recognition of online handwritten Bangla and Devanagari basic characters. At the beginning of the approach, we have checked the performance of nine pre-trained transfer learning models namely, DenseNet121, EfficientNetB0, NASNetMobile, VGG-16, VGG-19, ResNet50, InceptionV3, Xception, and MobileNetV2 for the recognition of the said handwritten characters. After that we have considered the best-performing model which is VGG-19 in our case. The obtained features from this model are then reduced using a two-stage FS approach. In the first stage, features are ranked using a filter method, called ReliefF. Then in the second stage, the ranked features are optimized by applying a nature-inspired meta-heuristic, called gray wolf optimization. The experimental outcomes reveal that not only the proposed approach reduces the feature dimension by 88.5% for Bangla and 91.9% for Devanagari, but also it increases classification accuracy for Bangla (reaches 100%) and retains it for Devanagari (which is 99.61%). © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.