Comparison of the Effectiveness of ANN and CNN in Image Classification

Considering that both ANN and CNN networks are used in image classification, this paper aimed to evaluate the effectiveness of both types of networks in the binary classification of selected datasets containing color and monochrome samples with variable resolutions. Additionally, it was examined whether there are factors influencing the effectiveness of the applied ANN and CNN networks. Based on the obtained results, it can be unequivocally stated that CNN networks proved to be a decidedly better, almost ideal, tool for image classification compared to ANN networks. Factors that affected the effectiveness of ANN networks, such as sample resolution, the percentage content of samples labeled 0 and 1 in the dataset, and whether the sample was color or monochrome, did not affect the effectiveness of CNN networks. Moreover, it was observed that the choice of research metrics in the form of the number of correctly classified samples and the value of the AUC coefficient proved to be an effective and stable metric of classification quality. The results obtained encourage further exploration for new image classification tools or the development of existing ones.