#Enhancing Defect classification in solar panels with Electroluminescence Imaging and Advanced Machine learning and deep learning and Ai by using YOLOV8 Solar Panel Defect Classification using Electroluminescence (EL) Imaging & YOLOv8 Project Title: Enhancing Defect Classification in Solar Panels using Electroluminescence Imaging, Machine Learning, Deep Learning & YOLOv8 Overview This project focuses on detecting and classifying defects in photovoltaic (PV) solar panels using Electroluminescence (EL) images and YOLOv8 deep learning model. EL imaging highlights hidden defects like micro-cracks, hotspots, and broken cell fingers, making it an ideal method for automated inspection. This repository includes dataset structure, training pipeline, model configuration, evaluation metrics, and deployment steps to build a complete AI-based defect inspection system. Objective Detect various solar panel defects using EL images. Train an accurate, real-time YOLOv8-based detection model. Deploy the model for practical field inspections using edge devices or cloud. Automate reporting for maintenance teams. Defects Covered Recommended classes: micro_crack major_crack broken_finger inactive_cell hotspot delamination solder_break (You may add/remove classes depending on your dataset.) Dataset Structure project-root/ │ ├── images/ │ ├── train/ │ ├── val/ │ └── test/ │ ├── labels/ │ ├── train/ │ ├── val/ │ └── test/ │ └── data.yaml Sample data.yaml train: ./images/train val: ./images/val test: ./images/test nc: 7 names: [micro_crack, major_crack, broken_finger, inactive_cell, hotspot, delamination, solder_break] Annotation Format (YOLO Format) Each annotation file (.txt) contains: class_id center_x center_y width height All values are normalized (0–1). Tools you can use: LabelImg Roboflow Annotate CVAT Label Studio Installation pip install ultralytics opencv-python numpy Or clone YOLOv8 directly: pip install ultralytics Training YOLOv8 Train using CLI: yolo task=detect mode=train model=yolov8s.pt data=data.yaml imgsz=640 epochs=50 batch=16 Train using Python: from ultralytics import YOLO model = YOLO('yolov8s.pt') model.train(data='data.yaml', epochs=50, imgsz=640) Evaluation Run validation: yolo mode=val model=runs/detect/train/weights/best.pt data=data.yaml Metrics produced: mAP@0.5 mAP@0.5:0.95 Precision & Recall Confusion Matrix Per-class AP Inference On an image: yolo predict model=best.pt source=path/to/image.jpg With Python: model = YOLO('best.pt') results = model('image.jpg') results.show() Model Export yolo export model=best.pt format=onnx Supported formats: TensorRT ONNX CoreML OpenVINO TFLite Deployment Options Edge Deployment: NVIDIA Jetson Nano / Xavier (TensorRT) Raspberry Pi + Coral TPU Intel NCS2 (OpenVINO) Cloud Deployment: Flask / FastAPI REST API Streamlit dashboard Mobile/Web inspection tool Augmentations Used Horizontal/Vertical flip Brightness/contrast shift Gaussian noise Random rotation Mosaic (YOLO built-in) CutMix (optional) # A Benchmark for Visual Identification of Defective Solar Cells in Electroluminescence Imagery [](https://pypi.org/project/elpv-dataset) [](https://pypi.org/project/elpv-dataset) This repository provides a dataset of solar cell images extracted from high-resolution electroluminescence images of photovoltaic modules.  ## The Dataset The dataset contains 2,624 samples of 300x300 pixels 8-bit grayscale images of functional and defective solar cells with varying degree of degradations extracted from 44 different solar modules. The defects in the annotated images are either of intrinsic or extrinsic type and are known to reduce the power efficiency of solar modules. All images are normalized with respect to size and perspective. Additionally, any distortion induced by the camera lens used to capture the EL images was eliminated prior to solar cell extraction. ## Annotations Every image is annotated with a defect probability (a floating point value between 0 and 1) and the type of the solar module (either mono- or polycrystalline) the solar cell image was originally extracted from. ## Usage Install the Python package ```console pip install elpv-dataset ``` and load the images and the corresponding annotations as follows: ```python from elpv_dataset.utils import load_dataset images, proba, types = load_dataset() ``` ## Citing If you use this dataset in scientific context, please cite the following publications: > Buerhop-Lutz, C.; Deitsch, S.; Maier, A.; Gallwitz, F.; Berger, S.; Doll, B.; Hauch, J.; Camus, C. & Brabec, C. J. A Benchmark for Visual Identification of Defective Solar Cells in Electroluminescence Imagery. European PV Solar Energy Conference and Exhibition (EU PVSEC), 2018. DOI: [10.4229/35thEUPVSEC20182018-5CV.3.15](http://dx.doi.org/10.4229/35thEUPVSEC20182018-5CV.3.15) > Deitsch, S., Buerhop-Lutz, C., Sovetkin, E., Steland, A., Maier, A., Gallwitz, F., & Riess, C. (2021). Segmentation of photovoltaic module cells in uncalibrated electroluminescence images. Machine Vision and Applications, 32(4). DOI: [10.1007/s00138-021-01191-9](https://doi.org/10.1007/s00138-021-01191-9) > Deitsch, S.; Christlein, V.; Berger, S.; Buerhop-Lutz, C.; Maier, A.; Gallwitz, F. & Riess, C. Automatic classification of defective photovoltaic module cells in electroluminescence images. Solar Energy, Elsevier BV, 2019, 185, 455-468. DOI: [10.1016/j.solener.2019.02.067](http://dx.doi.org/10.1016/j.solener.2019.02.067) BibTeX details:

```bibtex @InProceedings{Buerhop2018, author = {Buerhop-Lutz, Claudia and Deitsch, Sergiu and Maier, Andreas and Gallwitz, Florian and Berger, Stephan and Doll, Bernd and Hauch, Jens and Camus, Christian and Brabec, Christoph J.}, title = {A Benchmark for Visual Identification of Defective Solar Cells in Electroluminescence Imagery}, booktitle = {European PV Solar Energy Conference and Exhibition (EU PVSEC)}, year = {2018}, eventdate = {2018-09-24/2018-09-28}, venue = {Brussels, Belgium}, doi = {10.4229/35thEUPVSEC20182018-5CV.3.15}, } @Article{Deitsch2021, author = {Deitsch, Sergiu and Buerhop-Lutz, Claudia and Sovetkin, Evgenii and Steland, Ansgar and Maier, Andreas and Gallwitz, Florian and Riess, Christian}, date = {2021}, journaltitle = {Machine Vision and Applications}, title = {Segmentation of photovoltaic module cells in uncalibrated electroluminescence images}, doi = {10.1007/s00138-021-01191-9}, issn = {1432-1769}, number = {4}, volume = {32}, } @Article{Deitsch2019, author = {Sergiu Deitsch and Vincent Christlein and Stephan Berger and Claudia Buerhop-Lutz and Andreas Maier and Florian Gallwitz and Christian Riess}, title = {Automatic classification of defective photovoltaic module cells in electroluminescence images}, journal = {Solar Energy}, year = {2019}, volume = {185}, pages = {455--468}, month = jun, issn = {0038-092X}, doi = {10.1016/j.solener.2019.02.067}, publisher = {Elsevier {BV}}, } ```

## License
All the images in this work are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Accompanying Python source code is distributed under the terms of the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0.html). For commercial use, please contact us for further information. preprocess import os import shutil import pandas as pd import numpy as np import cv2 from sklearn.model_selection import train_test_split from glob import glob # Configuration DATASET_DIR = r'C:\Users\devas\solar project\data\elpv-dataset' OUTPUT_DIR = r'C:\Users\devas\solar project\data\processed' IMG_SIZE = (640, 640) RANDOM_SEED = 42 def find_labels_file(start_dir): for root, dirs, files in os.walk(start_dir): if 'labels.csv' in files: return os.path.join(root, 'labels.csv') return None def preprocess(): print("Finding labels.csv...") labels_path = find_labels_file(DATASET_DIR) if not labels_path: print(f"Error: labels.csv not found in {DATASET_DIR}. Did git clone finish?") return print(f"Found labels at: {labels_path}") df = pd.read_csv(labels_path, sep='\s+|,', engine='python') # Handle potential space or comma separation # Check columns # ELPV usually has no header, or specific columns. Let's inspect first row if needed. # Standard ELPV: image_path, defect_probability, type # But usually it's space separated: "images/cell0001.png 0.0 mono" # If no header, we assume: if len(df.columns) < 2: df=pd.read_csv(labels_path, delim_whitespace=True, header=None, names=['image_name', 'prob' , 'type' ]) # Standardize # Make sure we have image_name and prob if 'prob' not in df.columns: # try 2nd column df.columns=['image_name', 'prob' , 'type' ] print(f"Loaded {len(df)} samples.") # Define classes: 0.0=Normal,> 0.0 (or >0.5) = Defected? # Paper says: 0=functional, 1=defective. Intermediate = probability. # User step 2 says: "Normal / Defected". # We will threshold at 0.5. df['label'] = df['prob'].apply(lambda x: 'defected' if x >= 0.5 else 'normal') print(df['label'].value_counts()) # Split # Stratified split to keep balance train_df, test_df = train_test_split(df, test_size=0.1, stratify=df['label'], random_state=RANDOM_SEED) train_df, val_df = train_test_split(train_df, test_size=0.22, stratify=train_df['label'], random_state=RANDOM_SEED) # 0.22 of 0.9 ~ 0.2 total print(f"Train: {len(train_df)}, Val: {len(val_df)}, Test: {len(test_df)}") # clear output dir if os.path.exists(OUTPUT_DIR): shutil.rmtree(OUTPUT_DIR) for subdir in ['train', 'val', 'test']: for label in ['normal', 'defected']: os.makedirs(os.path.join(OUTPUT_DIR, subdir, label), exist_ok=True) # Processing function def process_subset(subset_df, subset_name): dataset_root = os.path.dirname(labels_path) for idx, row in subset_df.iterrows(): img_path = os.path.join(dataset_root, row['image_name']) # Handle if image_name contains "images/" prefix or not if not os.path.exists(img_path): # try removing/adding images prefix if row['image_name'].startswith('images/'): img_path = os.path.join(dataset_root, row['image_name']) else: img_path = os.path.join(dataset_root, 'images', row['image_name']) if not os.path.exists(img_path): print(f"Warning: Image not found {img_path}") continue # Load and resize img = cv2.imread(img_path) if img is None: print(f"Warning: Failed to load {img_path}") continue img_resized = cv2.resize(img, IMG_SIZE) # Save save_name = f"{subset_name}_{os.path.basename(row['image_name'])}" save_path = os.path.join(OUTPUT_DIR, subset_name, row['label'], save_name) cv2.imwrite(save_path, img_resized) print("Processing Train...") process_subset(train_df, 'train') print("Processing Val...") process_subset(val_df, 'val') print("Processing Test...") process_subset(test_df, 'test') print("Preprocessing complete!") if __name__ == '__main__': preprocess() script.js document.addEventListener('DOMContentLoaded', () => { const dropZone = document.getElementById('drop-zone'); const fileInput = document.getElementById('file-input'); const previewContainer = document.getElementById('preview-container'); const imagePreview = document.getElementById('image-preview'); const removeBtn = document.getElementById('remove-file'); const analyzeBtn = document.getElementById('analyze-btn'); const uploadForm = document.getElementById('upload-form'); const spinner = document.getElementById('spinner'); // Result elements const resultSection = document.getElementById('result-section'); const resultImage = document.getElementById('result-image'); const statusBadge = document.getElementById('status-badge'); const meterFill = document.getElementById('meter-fill'); const confidenceValue = document.getElementById('confidence-value'); const maintenanceMsg = document.getElementById('maintenance-msg'); // Drag & Drop ['dragenter', 'dragover', 'dragleave', 'drop'].forEach(eventName => { dropZone.addEventListener(eventName, preventDefaults, false); }); function preventDefaults(e) { e.preventDefault(); e.stopPropagation(); } ['dragenter', 'dragover'].forEach(eventName => { dropZone.addEventListener(eventName, () => dropZone.classList.add('dragover'), false); }); ['dragleave', 'drop'].forEach(eventName => { dropZone.addEventListener(eventName, () => dropZone.classList.remove('dragover'), false); }); dropZone.addEventListener('drop', handleDrop, false); function handleDrop(e) { const dt = e.dataTransfer; const files = dt.files; handleFiles(files); } // Click upload dropZone.addEventListener('click', () => fileInput.click()); fileInput.addEventListener('change', function() { handleFiles(this.files); }); function handleFiles(files) { if (files.length > 0) { const file = files[0]; if (file.type.startsWith('image/')) { const reader = new FileReader(); reader.onload = (e) => { imagePreview.src = e.target.result; previewContainer.classList.remove('hidden'); dropZone.classList.add('hidden'); analyzeBtn.disabled = false; }; reader.readAsDataURL(file); // Manually set files to input if dropped (a bit hacky but works for form submit) // Actually easier to just append to FormData later } } } removeBtn.addEventListener('click', () => { fileInput.value = ''; previewContainer.classList.add('hidden'); dropZone.classList.remove('hidden'); analyzeBtn.disabled = true; resultSection.classList.add('hidden'); }); // Submit uploadForm.addEventListener('submit', async (e) => { e.preventDefault(); const files = fileInput.files; // If dropped, input might be empty, need to handle that if needed, // but for now we assume input was populated or clicked. // If fileInput is empty, maybe drag&drop didn't populate it. // Let's rely on standard input selection for simplicity in this code block, // or check if dropped file exists. if (files.length === 0) return; const formData = new FormData(); formData.append('file', files[0]); // UI Loading analyzeBtn.disabled = true; spinner.classList.remove('hidden'); analyzeBtn.querySelector('span').textContent = 'Analyzing...'; resultSection.classList.add('hidden'); try { const response = await fetch('/predict', { method: 'POST', body: formData }); const data = await response.json(); if (response.ok) { showResult(data); } else { alert('Analysis failed: ' + (data.error || 'Unknown error')); } } catch (error) { console.error(error); alert('An error occurred during analysis.'); } finally { analyzeBtn.disabled = false; spinner.classList.add('hidden'); analyzeBtn.querySelector('span').textContent = 'Analyze Panel'; } }); function showResult(data) { resultSection.classList.remove('hidden'); resultImage.src = data.image_url; // Data: class (normal/defected), confidence (0.0-1.0) let cls = data.class.toLowerCase(); const confidence = parseFloat(data.confidence); const confidencePct = Math.round(confidence * 100); // Update badge statusBadge.className = 'status-badge'; if (cls.includes('defect') || cls.includes('bad')) { statusBadge.classList.add('defected'); statusBadge.textContent = 'DEFECT DETECTED'; maintenanceMsg.textContent = 'Warning: Structural anomalies detected. Immediate maintenance or further inspection is recommended to prevent efficiency loss.'; maintenanceMsg.style.color = '#ff3344'; } else if (cls.includes('normal') || cls.includes('good')) { statusBadge.classList.add('normal'); statusBadge.textContent = 'PANEL HEALTHY'; maintenanceMsg.textContent = 'No significant defects detected. The solar panel appears to be in optimal operating condition.'; maintenanceMsg.style.color = '#00ff88'; } else { // Fallback for unknown classes (e.g. from pretrained model) statusBadge.textContent = cls.toUpperCase(); maintenanceMsg.textContent = `Identified as ${cls}.`; maintenanceMsg.style.color = '#fff'; } // Update Meter meterFill.style.width = `${confidencePct}%`; confidenceValue.textContent = `${confidencePct}% Confidence`; // Scroll to result resultSection.scrollIntoView({ behavior: 'smooth' }); } }); style.css :root { --bg-color: #050a14; --text-color: #ffffff; --accent-color: #ffaa00; /* Solar Gold */ --accent-glow: rgba(255, 170, 0, 0.4); --glass-bg: rgba(255, 255, 255, 0.05); --glass-border: rgba(255, 255, 255, 0.1); --success-color: #00ff88; --danger-color: #ff3344; } * { margin: 0; padding: 0; box-sizing: border-box; font-family: 'Outfit', sans-serif; } body { background-color: var(--bg-color); color: var(--text-color); min-height: 100vh; overflow-x: hidden; display: flex; justify-content: center; align-items: center; position: relative; } /* Background Orbs */ .background-orb { position: absolute; border-radius: 50%; filter: blur(80px); z-index: -1; opacity: 0.6; } .orb-1 { top: -10%; left: -10%; width: 600px; height: 600px; background: radial-gradient(circle, #1a237e, transparent); } .orb-2 { bottom: -10%; right: -10%; width: 500px; height: 500px; background: radial-gradient(circle, #311b92, transparent); } .container { width: 90%; max-width: 1000px; padding: 2rem; z-index: 1; } header { text-align: center; margin-bottom: 3rem; animation: fadeInDown 0.8s ease-out; } .logo h1 { font-size: 3rem; font-weight: 700; background: linear-gradient(to right, #fff, var(--accent-color)); -webkit-background-clip: text; -webkit-text-fill-color: transparent; display: inline-block; } .logo .icon { font-size: 2.5rem; vertical-align: middle; margin-right: 10px; } .subtitle { color: rgba(255, 255, 255, 0.7); font-weight: 300; letter-spacing: 1px; } .glass-card { background: var(--glass-bg); backdrop-filter: blur(16px); border: 1px solid var(--glass-border); border-radius: 20px; padding: 2rem; box-shadow: 0 8px 32px 0 rgba(0, 0, 0, 0.37); transition: transform 0.3s ease; } .upload-section { text-align: center; animation: fadeInUp 0.8s ease-out 0.2s backwards; } .drop-zone { border: 2px dashed rgba(255, 255, 255, 0.2); border-radius: 12px; padding: 3rem; margin: 2rem 0; cursor: pointer; transition: all 0.3s; } .drop-zone:hover, .drop-zone.dragover { border-color: var(--accent-color); background: rgba(255, 170, 0, 0.05); } .drop-icon { font-size: 3rem; display: block; margin-bottom: 1rem; } .btn-primary { background: linear-gradient(135deg, var(--accent-color), #ff8800); border: none; padding: 1rem 3rem; border-radius: 50px; color: #fff; font-size: 1.1rem; font-weight: 600; cursor: pointer; transition: all 0.3s; box-shadow: 0 4px 15px var(--accent-glow); display: flex; align-items: center; justify-content: center; margin: 0 auto; min-width: 200px; } .btn-primary:hover:not(:disabled) { transform: translateY(-2px); box-shadow: 0 6px 20px var(--accent-glow); } .btn-primary:disabled { opacity: 0.5; cursor: not-allowed; transform: none; } /* Preview */ .preview-container { position: relative; max-width: 300px; margin: 2rem auto; } .preview-container img { width: 100%; border-radius: 12px; box-shadow: 0 4px 15px rgba(0,0,0,0.5); } .close-btn { position: absolute; top: -10px; right: -10px; background: var(--danger-color); color: white; border: none; border-radius: 50%; width: 30px; height: 30px; cursor: pointer; font-size: 1.2rem; } /* Results */ .result-section { margin-top: 2rem; animation: fadeInUp 0.8s ease-out; } .result-content { display: flex; flex-wrap: wrap; gap: 2rem; align-items: center; justify-content: center; } .result-image-box { position: relative; flex: 1; min-width: 300px; border-radius: 12px; overflow: hidden; border: 1px solid var(--accent-color); } .result-image-box img { width: 100%; display: block; } .scan-line { position: absolute; top: 0; left: 0; width: 100%; height: 2px; background: var(--accent-color); box-shadow: 0 0 10px var(--accent-color); animation: scan 2s infinite linear; opacity: 0.8; } .result-details { flex: 1; min-width: 300px; text-align: left; } .status-badge { display: inline-block; padding: 0.5rem 1.5rem; border-radius: 50px; font-weight: 700; text-transform: uppercase; font-size: 1.2rem; margin-bottom: 2rem; background: rgba(255, 255, 255, 0.1); } .status-badge.normal { background: rgba(0, 255, 136, 0.2); color: var(--success-color); border: 1px solid var(--success-color); } .status-badge.defected { background: rgba(255, 51, 68, 0.2); color: var(--danger-color); border: 1px solid var(--danger-color); box-shadow: 0 0 20px rgba(255, 51, 68, 0.3); } .confidence-meter { margin-bottom: 1.5rem; } .meter-bar { height: 10px; background: rgba(255, 255, 255, 0.1); border-radius: 5px; margin: 0.5rem 0; overflow: hidden; } .meter-fill { height: 100%; background: linear-gradient(90deg, var(--accent-color), #ff8800); width: 0%; transition: width 1s ease-out; } .maintenance-msg { font-size: 1.1rem; line-height: 1.6; } footer { text-align: center; margin-top: 4rem; color: rgba(255, 255, 255, 0.4); font-size: 0.9rem; } .hidden { display: none !important; } /* Spinner */ .spinner { width: 20px; height: 20px; border: 3px solid rgba(255, 255, 255, 0.3); border-radius: 50%; border-top-color: white; animation: spin 1s linear infinite; margin-left: 10px; } @keyframes spin { to { transform: rotate(360deg); } } @keyframes scan { 0% { top: 0%; } 50% { top: 100%; } 100% { top: 0%; } } @keyframes fadeInDown { from { opacity: 0; transform: translateY(-20px); } to { opacity: 1; transform: translateY(0); } } @keyframes fadeInUp { from { opacity: 0; transform: translateY(20px); } to { opacity: 1; transform: translateY(0); } } INDEX.HTML APP.PY import os import cv2 import numpy as np from flask import Flask, render_template, request, jsonify, url_for from ultralytics import YOLO from werkzeug.utils import secure_filename app = Flask(__name__) # Config UPLOAD_FOLDER = 'static/uploads' MODEL_PATH = '../models/elpv_yolov8/weights/best.pt' # Path to best model after training # Fallback if training not done, use pretrained (won't be accurate for solar, but works for code) DEFAULT_MODEL = 'yolov8n-cls.pt' os.makedirs(os.path.join(app.root_path, UPLOAD_FOLDER), exist_ok=True) print("Loading model...") try: if os.path.exists(MODEL_PATH): model = YOLO(MODEL_PATH) print(f"Loaded trained model from {MODEL_PATH}") else: print(f"Warning: Trained model not found at {MODEL_PATH}. Using {DEFAULT_MODEL} placeholder.") model = YOLO(DEFAULT_MODEL) except Exception as e: print(f"Error loading model: {e}") model = None @app.route('/') def index(): return render_template('index.html') @app.route('/predict', methods=['POST']) def predict(): if 'file' not in request.files: return jsonify({'error': 'No file part'}), 400 file = request.files['file'] if file.filename == '': return jsonify({'error': 'No selected file'}), 400 if file: filename = secure_filename(file.filename) filepath = os.path.join(app.root_path, UPLOAD_FOLDER, filename) file.save(filepath) # Inference if model: results = model(filepath) # Classification user case # results[0].probs.top1 -> index # results[0].names -> dict # Get top prediction probs = results[0].probs top1_index = probs.top1 confidence = float(probs.top1conf) class_name = results[0].names[top1_index] # Map class_name manually if using pretrained placeholder to avoid "goldfish" # But if using our trained model, names should be 'normal', 'defected' result_data = { 'class': class_name, 'confidence': f"{confidence:.2f}", 'image_url': url_for('static', filename=f'uploads/{filename}') } return jsonify(result_data) else: return jsonify({'error': 'Model not loaded'}), 500 if __name__ == '__main__': app.run(debug=True, port=5000) IMPLEMENTING PLAN Enhancing Defect Classification in Solar Panels using EL Imaging and ML Goal Description Automatically detect and classify solar panel defects using Electroluminescence (EL) images and the YOLOv8 deep learning model. The system will be accessible via a user-friendly web application. User Review Required IMPORTANT Dataset Acquisition: We need to confirm if the ELPV dataset can be automatically downloaded or if manual placement is required. Compute Resources: Training YOLOv8 requires reasonable compute (GPU recommended). Proposed Changes Project Structure data/: Directory for ELPV dataset (raw and processed) src/: Source code for preprocessing and training app/: Web application code (templates, static, backend) models/: Directory to save trained YOLOv8 models Dependencies ultralytics (YOLOv8) flask (Web Framework) opencv-python, numpy, pandas, scikit-learn (Data processing) git (to clone dataset) Data Preprocessing (src/preprocess.py) Clone dataset from https://github.com/zae-bayern/elpv-dataset. Load images and labels (labels are probabilities 0.0-1.0). Binarize labels: 0.0 = Normal, 1.0 (or >0.5) = Defected. Resize images to 640x640 (standard for YOLOv8). Normalize pixel values. Split into Train (70%), Val (20%), Test (10%). Organize into YOLOv8 Classification format: root/train/normal, root/train/defected root/val/normal, root/val/defected root/test/normal, root/test/defected Model Training (src/train.py) Load yolov8n-cls.pt (Classification model). Train on the processed folder. Export best model. Web Application (app/) Backend: Flask app to load the model. Frontend: HTML/CSS/JS with "Premium" design (Dark mode, glassmorphism). Flow: Upload -> Process -> Show Result (Class + Confidence). Verification Plan Automated Tests Run src/train.py --dry-run (few epochs) to ensure pipeline works. Test Flask endpoints with sample images. Manual Verification Upload test images to the web app and verify output. WALKTHROUGH Walkthrough: Solar Panel Defect Classification This document explains how to run the project from start to finish. 1. Data Preprocessing (Run First) File: src/preprocess.py Command: python src/preprocess.py What it does: Loads the ELPV dataset labels. Splits images into Training (70%), Validation (20%), and Testing (10%) sets. Resizes all images to 640x640 pixels (standard for YOLOv8). Saves organized data into data/processed/. 2. Model Training (Run Second) File: src/train.py Command: python src/train.py What it does: Loads the yolov8n-cls.pt (Nano Classification) model. Trains the model on the processed data for 50 epochs. Saves the best model weights to models/elpv_yolov8/weights/best.pt. Note: Training can take time. If you want to skip this, the app will use a default pretrained model (less accurate for this specific task). 3. Web Application (Run Last) File: app/app.py Command: python app/app.py What it does: Starts the Flask web server. Loads the trained model (or fallback). Opens the interface at http://127.0.0.1:5000. Allows you to upload EL images and view defect predictions. 4. Helper Files verify.py: a script to quickly test the API without opening the browser. TRAIN.PY from ultralytics import YOLO def train(): # Load model model = YOLO('yolov8n-cls.pt') # load a pretrained model (recommended for training) # Train # data argument for classify mode is the path to folder containing train/val/test folders results = model.train(data='data/processed', epochs=50, imgsz=640, project='models', name='elpv_yolov8') # Validate metrics = model.val() print("Validation Accuracy:", metrics.top1) # Export path = model.export(format='onnx') print(f"Model exported to {path}") if __name__ == '__main__': train()