Transfer Learning-based Ensemble Classifier for Multi-Class Texture Classification in Remote Sensing

Abstract

Automated texture classification has become an important area of research in computer vision and industrial inspection systems, where accurate surface analysis is required to maintain product quality and detect defects during manufacturing. In earlier industrial practices, surface inspection was performed manually by human experts who visually examined materials to identify defects such as holes, cracks, and foreign objects. Although manual inspection was widely used, it was often slow, inconsistent, and prone to human error. Later, traditional automated systems based on classical image processing techniques were introduced, but these methods relied heavily on handcrafted features and simple classification approaches, which limited their ability to capture complex texture patterns. As industrial datasets became larger and more diverse, these limitations created a need for intelligent systems capable of extracting meaningful visual features and performing reliable multi-class classification. To address these challenges, this study presents a transfer learning–based texture classification framework that combines deep feature extraction with machine learning techniques. In the proposed approach, a pretrained Visual Geometry Group 19 (VGG19) is used to extract high-level visual features from industrial texture images. The extracted feature vectors are then used to train multiple machine learning classifiers including Linear Discriminant Analysis Classifier (LDAC), Quadratic Discriminant Analysis Classifier (QDAC), Support Vector Classifier (SVC), and Extra Trees Classifier (ETC). Among these models, the proposed VGG19 with ETC classifier achieved the highest classification accuracy of 97.31%, demonstrating its strong capability in distinguishing between different texture categories such as good, hole, and objects.