A Multi-Feature Extraction and Selection Framework for High-Accuracy Surface Defect Classification in Industrial Casting Parts

Authors : Mustafa Büber, Ali Yasar

Pages : 82-87

Doi:10.55213/kmujens.1817251

View : 46 | Download : 132

Publication Date : 2025-12-23

Article Type : Research Paper

Abstract :This study presents an automated quality inspection approach for detecting surface defects in industrial casting parts using image processing and machine learning techniques. A total of ten different feature extraction methods—Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), Scale-Invariant Feature Transform (SIFT), Gray Level Co-occurrence Matrix (GLCM), Gabor filters, color histogram, wavelet transform, Hu moments, Zernike moments, and Fourier transform—were applied to 300×300 grayscale images. To evaluate the effect of spatial resolution, features were extracted at five different cell sizes: 25×25, 50×50, 100×100, 150×150, and 300×300. Dimensionality reduction was performed using minimum Redundancy Maximum Relevance (mRMR) and Chi-square (χ²) feature selection techniques. The resulting feature sets were classified with six different algorithms in MATLAB Classification Learner, including Fine Tree, Fine KNN, Wide Neural Network, Bagged Trees, Fine Gaussian SVM, and Binary GLM. Experimental results demonstrated that the highest accuracy rate of 99.7% was achieved with the Wide Neural Network model trained on the complete feature set at a 25×25 cell size. These findings indicate that smaller cell sizes preserve critical surface details and enhance classification performance. The study highlights that the proposed methodology can serve as a highly accurate, efficient, and practical solution for automated defect detection in the casting industry, offering strong potential for real-world industrial applications.
Keywords : Döküm kusurları, Görüntü işleme, Makine öğrenmesi, Özellik çıkarımı, HOG, LBP

ORIGINAL ARTICLE URL