Browsing by Author "Mert, A."

Now showing 1 - 2 of 2

An Acoustic Signal-to-Conversion Integrated Convolutional Neural Network Model for Egg Crack Detection
(Taylor and Francis Ltd., 2025) Balci, Z.; Yabanova, İ.; Mert, A.
1. The presence of fractures or cracks in eggshells represent a significant risk in terms of food safety. Bacteria and viruses are likely to enter through these cracks, which increases the risk of food poisoning. Furthermore, deformations in the shell can compromise the integrity of the protective shell, rendering the egg more susceptible to environmental damage and accelerating deterioration. 2. In order to mitigate these risks, a convolutional neural network (CNN) integrated into an acoustic signal to image conversion was developed as a crack detection system. Mechanical and electronic sub-systems were designed to generate non-destructive acoustic excitation on the eggshell and capture the resulting sound with a high-sensitivity microphone. 3. The recorded 1 × 731-sample signals from 120 intact or cracked eggs were subjected to variational mode decomposition (VMD) to extract intrinsic mode functions (IMF). Subsequently, IMF were converted to greyscale images and classified using the proposed acoustic signal-to-image conversion and the lightweight CNN. 4. The proposed model showed the capability (100%) to distinguish between intact and cracked eggs, including invisible micro-cracks. © 2025 Elsevier B.V., All rights reserved.
Enhanced Photoacoustic Signal Processing Using Empirical Mode Decomposition and Machine Learning
(Taylor and Francis Ltd., 2025) Balci, Z.; Mert, A.
In this study, we propose a robust photoacoustic (PA) signal processing framework for a material independent defect detection using empirical mode decomposition (EMD) and machine learning algorithms. First, a database of the PA signals with 960 samples has been obtained from aluminium, iron, plastic and wood materials using a laser, microphone and data acquisition board-based PA apparatus. Second, the EMD based time and time-frequency domain techniques are proposed to extract robust cross-material feature space focusing on laser induced acoustic signal, and the decomposed intrinsic mode (IMF) with 14 extracted features are performed on totally 960 samples PA signals to evaluate k-nearest neighbour (k-NN), decision tree (DT) and support vector machine (SVM) classifiers. Inter- material and cross-material evaluations are performed, and the accuracy rates up to 100% for SVM and 97.77% for k-NN are yielded. © 2024 Informa UK Limited, trading as Taylor & Francis Group.