A Hybrid CNN-SVM Model for High-Accuracy Defect Detection in PV Modules Using Infrared Images

Abstract

This paper presents a novel approach for classifying infrared solar modules using a hybrid CNN-SVM model. The proposed method involves several key steps: preprocessing using histogram equalization to enhance image contrast, data augmentation to increase the diversity of the training set, feature extraction using a Convolutional Neural Network (CNN), and final classification with a Support Vector Machine (SVM) classifier. To evaluate the effectiveness of this approach, we used a comprehensive infrared solar modules dataset comprising 20,000 images. The hybrid model achieved an overall accuracy of 92.67%, with a precision of 90.85%, recall of 93.10%, and F1 score of 92.46%, demonstrating significant improvements over existing state-of-the-art methods. Comparative analysis with recent studies further validates the effectiveness of our approach. This work underscores the potential of combining deep learning with traditional machine learning techniques for enhanced solar module inspection and quality assurance.