Experimental Study of Various Deep Learning Models in Handwritten Character Recognition

Applications requiring automated analysis of handwritten text heavily rely on handwritten digit recognition, an essential issue in computer vision. The MNIST dataset is a well-known standard in the area, and the present research investigates many deep-learning techniques for digit recognition. This study examines modern designs such as Convolutional Neural Networks (CNN)., VGG-16., VGG-19., ResNet-50, ResNet-9., GoogLeNet, and DenseN et-121., concentrating on their feature extraction and classification skills., building on the advantages of classic models. The categorical cross-entropy loss function was used to train each model., and the ADAM optimizer was used for optimization. The models' capacity to autonomously extract features from the input data was improved by combining ReLU activations with techniques like gradient descent and backpropagation. CNN and its variations have shown remarkable performance in image identification and classification challenges. This study compares different models to shed light on the advantages and disadvantages of each architecture in the context of digit classification. It assesses their accuracy., precision., recall., and F1 scores in handwritten digit recognition. CNN and its variations have shown remarkable performance in image identification and classification challenges. This study compares different models to shed light on the advantages and disadvantages of each architecture in the context of digit classification. It assesses their accuracy., precision., recall., and F1 scores in handwritten digit recognition.