Performance comparison of ANN, CNN and RBF in multiclass image classification: batch vs. incremental learning

Multiclass image classification represents one of the key challenges in machine learning. This study compares the effectiveness of three neural network models—Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Radial Basis Function (RBF) networks—in the task of multiclass classification of monochromatic images, evaluated separately for incremental and batch learning approaches. The research involved analyzing classification performance across various image resolutions. To ensure a reliable assessment of the models, K-fold cross-validation and statistical analysis were employed, including the Friedman test and Dunn’s post-hoc test with Bonferroni correction. The results demonstrated that the CNN model achieved the highest classification accuracy in both training approaches. The ANN model performed better in batch learning than in incremental learning, while the RBF model exhibited significant instability in the incremental setting. Statistical analysis confirmed significant differences between the models, indicating the superiority of CNN over ANN and RBF. Furthermore, it was shown that reducing image resolution gradually decreases classification performance; however, no distinct critical threshold was observed where accuracy drops abruptly.