Comparison of LeNet-5, AlexNet and GoogLeNet models in handwriting recognition
Bartosz Michalski, Małgorzata Plechawska–Wójcik
Abstract
The aim of the study was to compare the accuracy of handwriting recognition and the time needed to classify data from the test sets. The Lenet-5, AlexNet and GoogLeNet architectures were used for the research. They are all models of convolutional neural networks. The research was carried out with the use of image databases, handwritten digits MNIST and handwritten letters EMNIST. After the tests, it was found that the GoogLeNet model showed the highest accuracy, and the LeNet-5 the lowest. However, the LeNet-5 model needed the least time to complete the task, and GoogLeNet the most. On the basis of the obtained results, it was found that increasing the complexity of the model positively influences the accuracy of object classification, but significantly increases the demand for computer re-sources.