| from transformers import pipeline | |
| from PIL import Image | |
| # Load a pre-trained image classification model (can be any model suitable for image classification) | |
| # Here, we use a simple CNN model like ResNet50 from Hugging Face, which can be fine-tuned for this task. | |
| model = pipeline("image-classification", model="google/vit-base-patch16-224-in21k") | |
| # Class labels for your solar panel damage types | |
| damage_classes = ['cracked', 'dusted', 'shaded', 'overheated'] | |
| def predict_damage(image_path): | |
| # Open and preprocess the image | |
| image = Image.open(image_path) | |
| # Get prediction from the model | |
| result = model(image) | |
| # Return the damage prediction text (i.e., class with highest confidence) | |
| predicted_class = result[0]['label'] | |
| confidence = result[0]['score'] | |
| # Check the predicted label and map it to the respective damage type | |
| if predicted_class in damage_classes: | |
| return f"This panel is {predicted_class} (Confidence: {confidence:.2f})" | |
| else: | |
| return "Unable to determine the damage type" | |