Convolutional Neural Network (CNN) Algorithm Development for Quality Measurement in Tire Product Classification: A Systematic Literature Review

Convolutional Neural Networks (CNN) have shown potential in tire defect classification, but challenges like image quality variations and model limitations persist. This Systematic Literature Review (SLR) employs the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology and a qualitative approach. It uses the Population, Intervention, Comparison, Outcome, and Context (PICOC) framework to analyze 30 articles from Google Scholar, published between 2018 and 2024 in Q1-Q4 journals. The review examines article distribution by publisher, evaluates the impact of key pre-training parameters (such as data split ratio, dropout rate, optimization algorithm, learning rate, and pre-trained models on CNN performance), and identifies challenges in defect classification. MDPI (Multidisciplinary Digital Publishing Institute) is the most frequent publisher, with the highest number of publications occurring in 2021. AlexNet and VGG16 are the most common models, and an 80:20 data split ratio, 0.5 dropout rate, and a 0.001 learning rate are frequently used. The Adam (Adaptive Momentum) optimization algorithm is widely adopted, but challenges remain in detecting small defects. Further improvements in CNN models, lighting techniques, and larger datasets are needed to enhance defect detection. This review highlights current trends and future research opportunities in tire quality measurement.