Title Research of image classification based on the interaction of wavelet transforms and artificial neural networks
Translation of Title Bangelių transformacijų ir dirbtinių neuroninių tinklų sąveika paremto vaizdų klasifikavimo tyrimas.
Authors Ismayilov, Mahammad
Full Text Download
Pages 61
Keywords [eng] convolutional neural network (cnn) ; haar wavelet decomposition ; image classification, multi-level wavelet fusion ; explainable artificial intelligence (xai)
Abstract [eng] In medical image analysis, image classification is an important task in the field of artificial intelligence, where small visual details can influence diagnostic decisions. Therefore, convolutional neural network models have shown strong performance in this field, but they may still be sensitive to noise and to the loss of important details during downsampling. The main goal of this research is to develop and evaluate a proposed WaveletFusion architecture that integrates multi-level Haar wavelet decomposition with convolutional neural network-based feature extraction. The model uses approximation and directional detail subbands to preserve both structural and texture-related information from brain MRI images. The WaveletFusion architecture was evaluated on two datasets that contain brain tumor MRI images: the multi-class Brain Tumor MRI and Br35H :: Brain Tumor Detection 2020 datasets. Several experimental and ablation analyses were performed, including a Grad-CAM explainability analysis to better understand the model‘s performance. The results show that the WaveletFusion 7-level model achieved the best overall performance. It improved classification results, reaching 0.90 of test accuracy and macro-F1 on the Brain Tumor MRI dataset and 0.96 of test accuracy and macro-F1 on the Br35H :: Brain Tumor Detection 2020 dataset. The results demonstrate that combining multi-level wavelet decomposition with convolutional neural network learning can improve classification performance and support more informative feature representation for brain MRI image classification.
Dissertation Institution Kauno technologijos universitetas.
Type Master thesis
Language English
Publication date 2026