Classifying Graphs of Elementary Mathematical Functions Using Convolutional Neural Networks

• Joaquim Viana UFPA
• Helder Matos UFPA
• Marcelle Mota UFPA
• Reginaldo Santos UFPA

Resumo

The classification of images of elementary mathematical function graphs presents a significant challenge in computer vision; this is due to the varied shapes and formats of each functions curves. This classification is crucial for identifying function graphs, which have important applications in text and mathematical symbol recognition technologies, aiding visually impaired individuals by providing access to printed content. In educational environments, this identification helps obtain the analytical expression of drawn graphs, facilitating the extraction of information from educational materials. This article investigates various convolutional neural network (CNN) architectures to identify the most suitable model for classifying images of elementary mathematical function graphs. We compare our model with other renowned architectures, such as ResNet, MobileNet, and EfficientNet, using a custom dataset of function graphs. Our experiments show that the proposed architecture significantly outperforms networks of general purpose, achieving an accuracy of 98.51% in classifying elementary mathematical function graphs.