Defect Detection in Tire X-Ray Images: Conventional Methods Meet Deep Structures (2402.18527v1)

Abstract: This paper introduces a robust approach for automated defect detection in tire X-ray images by harnessing traditional feature extraction methods such as Local Binary Pattern (LBP) and Gray Level Co-Occurrence Matrix (GLCM) features, as well as Fourier and Wavelet-based features, complemented by advanced machine learning techniques. Recognizing the challenges inherent in the complex patterns and textures of tire X-ray images, the study emphasizes the significance of feature engineering to enhance the performance of defect detection systems. By meticulously integrating combinations of these features with a Random Forest (RF) classifier and comparing them against advanced models like YOLOv8, the research not only benchmarks the performance of traditional features in defect detection but also explores the synergy between classical and modern approaches. The experimental results demonstrate that these traditional features, when fine-tuned and combined with machine learning models, can significantly improve the accuracy and reliability of tire defect detection, aiming to set a new standard in automated quality assurance in tire manufacturing.

• The paper introduces a hybrid defect detection method combining traditional feature extraction (LBP, GLCM, Fourier, Wavelet) with an RF classifier and enhanced YOLOv8.
• The study evaluates individual feature set contributions and dataset parameters, demonstrating superior precision and recall compared to deep learning models.
• The results highlight industrial relevance by achieving accurate, real-time defect detection in high-resolution tire X-ray images.

Enhanced Automated Defect Detection in Tire X-Ray Images through Traditional Feature Engineering and Machine Learning

Introduction

The paper presents an innovative method for automated defect detection in tire X-ray images, aiming to overcome the limitations of manual inspection with a system that combines traditional feature extraction techniques and advanced machine learning algorithms. The focus is on Local Binary Patterns (LBP), Gray-Level Co-occurrence Matrix (GLCM), Fourier, and Wavelet-based features alongside a Random Forest (RF) classifier and a comparison against the YOLOv8 deep learning model. This hybrid approach targets the challenges posed by the intricate patterns and textures of tire X-ray images, striving for a high-accuracy, real-time performance system suitable for industrial application.

Challenges in Automated Tire Defect Detection

The inspection of tires using X-ray imaging confronts several obstacles:

• High-Resolution X-Ray Samples Challenge necessitates a careful balance between computational demands and the preservation of critical details for accurate defect detection.
• Defect Characteristics Challenge involves detecting defects of various sizes and shapes, often with low contrast against the tire background.
• Tread Pattern Variability Challenge complicates defect detection due to the variability in tread patterns, which may resemble defects, alongside the issue of anisotropy affecting the learning efficacy of CNN architectures.
• Data Imbalance Challenge presents a bias risk towards defect-free samples, requiring strategies to ensure machine learning models accurately detect defects.

Methodology

The methodology section explores the specifics of the proposed approach:

• Feature Extraction: Explains the selection and expected performance of LBP, GLCM, Wavelet, and Fourier features in detecting textural anomalies and variations critical for identifying defects.
• YOLO with Augmented Features: Discusses the integration of engineered features into the YOLOv8 model for improved defect detection.
• Model Selection, Optimization, and Pre-Processing: Details the choice of the RF classifier, optimization techniques, and the pre-processing steps geared towards enhancing model performance.

Experiments and Results

The research includes a series of experiments assessing the effectiveness of the proposed method:

• Dataset Overview: Provides details of the dataset comprising 1054 tire X-ray scans annotated with defects.
• Comparing Engineered Features vs. Deep Models: Demonstrates the superior performance of the traditional feature-based detection method compared to YOLOv8, especially in precision and recall metrics.
• Ablation Study on the Feature Sets: Examines the contribution of individual feature sets, identifying combinations of GLCM, Fourier, and Wavelet features as the most effective.
• Effect of Dataset Generation Parameters: Analyzes the impact of window size, step size, and threshold settings on the per-window classification performance of RF models.

Discussion

The paper confirms the potential of integrating traditional feature extraction techniques with machine learning algorithms to improve automated defect detection in tire X-ray images. It highlights the effectiveness of carefully selected feature sets and optimized dataset parameters in surpassing the performance of deep learning models like YOLOv8 in certain benchmarks.

Conclusions

This research advocates for a hybrid approach to defect detection in tire manufacturing, blending classical feature engineering with advanced machine learning to meet the industry's need for a reliable, real-time performance system. Future work will likely focus on extending the applicability of this framework, promising significant advancements in industrial automation and quality assurance.

By harnessing both traditional and contemporary techniques, the paper sets a foundation for further exploration into efficient machine vision applications, offering a pragmatic solution to the challenges of tire defect detection.