Handwritten Telugu Vowel Character Classification Using Modified 25-Layer AlexNET With Transfer Learning

Due to remarkable performance, deep neural networks have gained popularity in computer vision and machine learning applications such as regression, segmentation, classification, detection, and pattern recognition over the past few years. This paper examines the categorization of handwritten telugu vowel characters using a modified 25-layer AlexNET and transfer learning approach. The six-class Handwritten Telugu vowel dataset from Kaggle is utilized in this experiment. To accommodate the required amount of vowel dataset classes, the 23rd and 25th layers of pre-trained AlexNET are adjusted. Training makes advantage of stochastic gradient descent using momentum optimizer. The proposed model accurately classified handwritten telugu vowel characters with a remarkable 99.72 % accuracy.