| Summary: | Deep Learning has been utilised in numerous fields, such as medicine, engineering, business, marketing, forensics, and industry. Pattern recognition, sentence recognition, document analysis, and handwritten recognition are where Convolution Neural Networks (CNN) are widely applied in deep learning. Deep learning architectures have recently been used to improve performance and produce significant outputs such as handwritten image recognition among schoolchildren. Therefore, deep learning is an essential requirement, especially for CNN, to evaluate various models that may be utilized as an indicator in recognizing schoolchildren's handwriting images. Hence, this study aims to assess the different deep learning models, specifically CNN models, in terms of their capacity to classify schoolchildren's handwritten characters with high accuracy and low error rates using a CNN network. To measure the performance of handwritten recognition, four CNN models are compared, namely AlexNet, GoogLeNet, CNN-2, and ResNet-50. The result obtained using AlexNet yielded an accuracy rate of 92.95%, which is the best compared to the other models focusing on Stochastic Gradient Descent with Momentum (SGDM) as an optimizer. Then, the experimental findings were compared to the previous work, which used various CNN models with an Adaptive Moment Estimation (ADAM) optimizer. The result proved that the model has the potential to be used in recognizing the handwritten characters of schoolchildren. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
|
|---|