Classification of noisy signals using fuzzy ARTMAP neural networks

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Noisy signal classification poses challenges for standard algorithms.
• Fuzzy ARTMAP neural networks (FAMNN) are effective but can struggle with generalization in noisy environments.
• Textured gray-scale image segmentation requires robust classification methods.

Purpose of the Study:

• To propose a modified fuzzy ARTMAP neural network (FAMNN) with enhanced generalization for noisy signal classification.
• To evaluate the performance of the modified FAMNN on textured gray-scale image segmentation tasks.
• To compare the modified FAMNN against the standard FAMNN under various noise conditions and dataset sizes.

Main Methods:

• Modification of the testing phase of the fuzzy ARTMAP neural network.
• Application of the modified FAMNN to textured gray-scale image segmentation.
• Experimental evaluation using diverse texture sets (aerial photos, Brodatz album), feature vectors, and noise types.

Main Results:

• The modified FAMNN exhibits superior generalization performance compared to the standard FAMNN in the presence of noise.
• Experimental results confirm the effectiveness of the proposed modification across various datasets and noise levels.
• Classification performance was analyzed for different network sizes, demonstrating consistent improvements with the modified approach.

Conclusions:

• The modified fuzzy ARTMAP neural network offers a significant improvement for noisy signal classification.
• The proposed method is particularly effective for textured image segmentation tasks.
• This research provides a more robust FAMNN for applications dealing with noisy data.