Classification of Tire Wear Pattern Using Gabor Feature Extraction and Probabilistic Neural Networks

Vehicle accidents in Indonesia often occur due to several factors, one of which is the condition of the vehicle that is not in prime condition, including tires. Artificial neural network technology can be utilized in solving this problem by designing a digital image processing system that classifies tire wear patterns. By using the feature extraction method with Gabor filters and probabilistic artificial neural network algorithms as the classification algorithm, it is expected to improve the accuracy of tire wear pattern classification compared to previous studies using the Gray Level Co-occurence Matrix feature extraction method with a backpropagation classifier. Classification of 4 categories of wear, namely normal wear patterns, alignment error wear patterns, underinflation wear patterns and overinflation wear patterns. Gabor filters with a 90° orientation angle, 4 wavelengths and six Gabor filter feature parameters are the best combination in this study. The results of the proposed evaluation test show that the accuracy of the probabilistic artificial neural network with Gabor filter feature extraction produces an accuracy of 95%. The results obtained are better when compared to previous studies that used the Gray Level Co-occurrence Matrix feature extraction method with a Backpropagation classifier to detect tire patterns which was only 80%.