Fuzzy Edge-Detection as a Preprocessing Layer in Deep Neural Networks for Guitar Classification.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Deep neural networks excel at hierarchical classification.
• Fuzzy image preprocessing effectively manages image uncertainty.
• Classifying guitar types based on body models is a relevant computer vision task.

Purpose of the Study:

• To evaluate fuzzy edge-detection methods combined with CNNs for guitar classification.
• To compare fuzzy edge-detection (Sobel, Prewitt, morphological gradient) against classical methods.
• To assess the impact of preprocessing on raw vs. normalized images and color spaces.

Main Methods:

• Implementation of two convolutional neural network (CNN) architectures.
• Application of fuzzy Sobel, fuzzy Prewitt, and fuzzy morphological gradient edge-detection.
• Comparison with classical edge-detection techniques (Sobel, Prewitt, morphological gradient) and grayscale/RGB images.

Main Results:

• Fuzzy preprocessing highlighted essential image features, outperforming classical methods.
• The proposed CNN models achieved favorable results compared to a pre-trained model.
• Fuzzy preprocessing reduced training times by over 20% compared to RGB images.

Conclusions:

• Fuzzy edge-detection combined with CNNs offers an effective approach for guitar classification.
• Fuzzy preprocessing enhances feature extraction and improves classification performance.
• This method provides a computationally efficient alternative for computer vision tasks.