BDNet: Bengali Handwritten Numeral Digit Recognition based on Densely connected Convolutional Neural Networks

Images of handwritten digits are different from natural images as the orientation a digit, well similarity features digits, makes confusion. On other hand, deep convolutional neural networks achieving huge success in computer vision problems, especially image classification. Here, we propose task-oriented model called Bengali numeral digit recognition based on densely connected (BDNet). BDNet is used to classify (recognize) digits. It end-to-end trained using ISI dataset. During training, untraditional data preprocessing and augmentation techniques so that works The has achieved test accuracy 99.78% (baseline was 99.58%) dataset numerals. So, gives 47.62% error reduction compared previous state-of-the-art models. Here have also created 1000 numerals model, it giving promising results. Codes, our own available at C :https://github.com/Sufianlab/BDNet.