Scholarlink

The digitization and processing of the Bangla script heavily relies on automated technologies that can identify basic characters written by hand in Bangla. These systems use advanced machine-learning techniques to identify and interpret handwritten text accurately. The enormous variances in writing styles present a significant problem for handwritten character identification, especially for complicated scripts like Bangla. The new deep convolutional neural network called the VashaNet-V2 model is presented in this paper. Its purpose is to solve the problem of handwritten Bangla letter recognition. In the VashaNet-V2 model, we utilized a comprehensive 19-layer DCNN architecture. This structure includes seven dropout layers, five convolutional layers, five max-pooling layers, one flattening layer, two dense layers, and an output layer. A mixed dataset with 22,500 images was used in the experiment for training and evaluating the model. To create the mixed dataset, we combined a primary dataset containing 7,500 images with the CMATERdb 3.1.2 dataset comprising 15,000 images. The character recognition model we proposed showed outstanding performance, attaining validation accuracy rates of 93.07% on the primary dataset, 94.13% on the CMATERdb 3.1.2 dataset, and 95.20% on the mixed dataset. By effectively handling the intricacies of the Bangla script, the VashaNet-V2 model offers a powerful tool for automated character recognition, contributing significantly to advancements in this field.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

VashaNet-V2: Bangla Handwritten Character Recognition Using a Novel Deep Convolutional Neural Network and an Extended Original Dataset

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mirza Raquib,
  • Mohammad Amzad Hossain,
  • Md. Jubayar Alam Rafi,
  • Iftehaz Newaz,
  • Md. Maruf Hossain,
  • Mohammad Rony,
  • Farida Siddiqi Prity,
  • Md. Bipul Hossain

摘要

The digitization and processing of the Bangla script heavily relies on automated technologies that can identify basic characters written by hand in Bangla. These systems use advanced machine-learning techniques to identify and interpret handwritten text accurately. The enormous variances in writing styles present a significant problem for handwritten character identification, especially for complicated scripts like Bangla. The new deep convolutional neural network called the VashaNet-V2 model is presented in this paper. Its purpose is to solve the problem of handwritten Bangla letter recognition. In the VashaNet-V2 model, we utilized a comprehensive 19-layer DCNN architecture. This structure includes seven dropout layers, five convolutional layers, five max-pooling layers, one flattening layer, two dense layers, and an output layer. A mixed dataset with 22,500 images was used in the experiment for training and evaluating the model. To create the mixed dataset, we combined a primary dataset containing 7,500 images with the CMATERdb 3.1.2 dataset comprising 15,000 images. The character recognition model we proposed showed outstanding performance, attaining validation accuracy rates of 93.07% on the primary dataset, 94.13% on the CMATERdb 3.1.2 dataset, and 95.20% on the mixed dataset. By effectively handling the intricacies of the Bangla script, the VashaNet-V2 model offers a powerful tool for automated character recognition, contributing significantly to advancements in this field.