Convolutional and Backpropagation Neural Networks On Image Classification
Jianqiao Song · International journal of high school research · 2023
Nowadays, with the rapid development of Artificial Intelligence (AI) in our everyday lives, machine learning algorithms have been applied in a wide variety of fields, performing tasks that are unfeasible for conventional algorithms.Neural networks, inspired by the design of the biological nervous system, have become increasingly popular in fields that require recognizing relationships between vast amounts of data, such as facial recognition, stock market prediction, and signature verification.Different types of models each have their unique architectures.There are differences between the computational complexities of the models, influencing their performance of specific tasks.However, determining which architecture best suits a specific application, such as image classification, takes time and effort.This article compares the performance of two types of artificial neural networks when classifying images: the Convolutional Neural Network and the Fully Connected Neural Network.The results from experimentation lead to a better understanding of the two fundamental models and how their training and validation accuracies vary individually, both reaching a terminal point ultimately.It was clear that CNN had an advantage in terms of better accuracy but more time-consuming.It was also found that the number of convolutional layers does not necessarily improve the accuracies.