Intelligent fault detection in solar photovoltaic panels through thermal imaging and machine learning
Reproducibility package for “Intelligent fault detection in solar photovoltaic panels through thermal imaging and machine learning”. Contents: • Analysis pipeline: image-grouped nested cross-validation (5 outer × 3 inner folds, 20 repetitions) with Optuna hyperparameter search, HOG features and SVM, KNN, random forest and decision tree classifiers • Patch-extraction script with healthy patches sampled to match the hotspot size distribution, and the patch metadata (coordinates, class, size, source image) • Comparison descriptors: dense SIFT and SURF bags of visual words, KAZE at detected keypoints, LBP, GLCM, Lab* colour statistics, raw pixels
Resource information
- Format
- zip
- Original repository
- 23758653
- DOI
- 10.5281/zenodo.23758653
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- mit-license
- Last record update
- Nov 5, 2026
About this resource
Reproducibility package for “Intelligent fault detection in solar photovoltaic panels through thermal imaging and machine learning”. Contents: • Analysis pipeline: image-grouped nested cross-validation (5 outer × 3 inner folds, 20 repetitions) with Optuna hyperparameter search, HOG features and SVM, KNN, random forest and decision tree classifiers • Patch-extraction script with healthy patches sampled to match the hotspot size distribution, and the patch metadata (coordinates, class, size, source image) • Comparison descriptors: dense SIFT and SURF bags of visual words, KAZE at detected keypoints, LBP, GLCM, Lab* colour statistics, raw pixels and a size-only control • Random seeds, fold assignments of all repetitions, Optuna trial records of the four main models, repetition-1 predictions, and a manifest with software versions and settings • Raspberry Pi 3 Model B and workstation benchmark: scripts, deployment models, and raw timing results The README gives the commands used and the software versions (Python 3.10.19, scikit-learn 1.7.2, OpenCV 4.10.0, Optuna 5.0.0). The original thermal images are not redistributed; they are available from Roboflow Universe (American University of Sharjah, Hotspot Detection in Solar Panel dataset, Version 2, public domain).