Solar-Panel-Defect-Detection

🔹 Solar Panel Defect Detection using YOLOv9c-seg

---

📘 Overview

This project focuses on detecting and segmenting surface defects in solar panels using the YOLOv9c-seg model.
The primary objective is to automate the inspection process and improve the efficiency and accuracy of identifying defective panels, reducing manual labor and maintenance downtime.

---

🧠 Objectives

• Develop a robust defect detection model for solar panels using YOLOv9c-seg.
• Segment defective regions with high accuracy to support maintenance planning.
• Evaluate model performance across multiple defect classes.
• Deploy an interactive application for visualizing defect detection results.

---

🗂️ Dataset

• Source: Custom dataset prepared using Roboflow.
• Total Images: 1,500+ labeled images.
• Classes: 6 defect categories — Bird-drop, Defective, Dusty, Electrical-Damage, Non-Defective, Physical-Damage.
• Structure:

train/
  images/
  labels/
val/
  images/
  labels/
test/
  images/
  labels/

---

⚙️ Tools & Technologies

Category	Tools / Libraries
Programming	Python
ML Framework	YOLOv9c-seg, PyTorch
Data Handling	Pandas, NumPy
Visualization	OpenCV, Matplotlib
Deployment	Streamlit
Version Control	Git, GitHub

---

🚀 Implementation Steps

1. Data Preprocessing: Data cleaning, augmentation, and annotation via Roboflow.
2. Model Training: Used YOLOv9c-seg for image segmentation and trained with early stopping.
3. Evaluation: Measured performance using mAP@50, Precision, and Recall metrics.
4. Deployment: Built a Streamlit web app to visualize real-time defect detection and segmentation results.

---

📂 Project Structure

solar-panel-defect-detection/
│
├── app/
│   ├── app.py                  # Streamlit web application
│   ├── requirements_app.txt    # (Optional) app-specific dependencies
│
├── train/
│   ├── Train_Solaryolov9.ipynb      # Model training notebook (Google Colab)
│   ├── requirements_train.txt  # (Optional) training dependencies
│
├── yolov9_best.pt              # Trained YOLOv9 model weights
├── requirements.txt            # Combined environment dependencies
├── README.md                   # Project documentation (this file)
└── data.yaml                   # Data yaml file

---

💾 Clone the Repository

Clone the repo

git clone https://github.com/arshiakauser/Solar-Panel-Defect-Detection.git  

Navigate into the project folder

cd solar-panel-defect-detection

---

⚙️ Environment Setup (Combined)

You can set up one environment for both training and running the app.

🧩 1. Create and Activate Environment

#Create a virtual environment
python -m venv solar_env

#Activate environment
# Windows:
solar_env\Scripts\activate
# macOS/Linux:
source solar_env/bin/activate

📦 2. Install Dependencies

Install all dependencies from the combined file:

pip install -r requirements.txt

Example requirements.txt:

ultralytics==8.2.77
streamlit
opencv-python
pillow
numpy
matplotlib
torch
torchvision
torchaudio
jupyter
pandas

---

🚀 Model Training (YOLOv9)

• The model was trained in Google Colab using YOLOv9.
• If you wish to retrain:

To Run Locally:

cd train
jupyter notebook Train_Solaryolov9.ipynb

• Update dataset paths and parameters as needed.
• Train the model.
• Export the best model as yolov9_best.pt and move it to the project root.

Dataset Structure:

• Ensure your dataset is structured as:

dataset/
├── train/
│   ├── images/
│   └── labels/
├── val/
│   ├── images/
│   └── labels/
├── test/
│   ├── images/
│   └── labels/
└── data.yaml

---

📊 Model Performance

Metric	Bounding Box(B)	Segmentation Mask (M)
Precision	0.786	0.779
Recall	0.715	0.689
mAP@50	0.776	0.736
mAP@50-95	0.583	0.501

---

💡 Insights & Learnings

• Data augmentation improved model generalization on unseen solar panel samples.
• Learned optimization of YOLOv9c parameters for better segmentation accuracy.
• Integrated computer vision models with a user-friendly dashboard using Streamlit.

---

🧩 Challenges Faced

• Managing annotation quality for small defect regions.
• Balancing dataset across multiple defect classes.
• Ensuring smooth inference during real-time deployment.

---

🖥️ Deployment

Run Streamlit App Locally:

cd app
streamlit run app.py

App Features:

• Upload and preview solar panel images
• Real-time defect detection using trained YOLOv9 model
• Class-wise count of detected defects
• Download annotated output (optional)
• User-friendly and responsive Streamlit UI

🧮 Example Output

Original Image	Detected Output

Detected Classes Example:

• Bird Drop: 17
• Defective: 6

App Link: 🔸 http://localhost:8501/

Demo Video: 🔸

---

🏁 Results

The YOLOv9c-seg model achieved 77% mAP@50, efficiently detecting multiple types of defects in solar panels with accurate segmentation. The Streamlit dashboard enables easy visualization and inspection support for maintenance engineers.

---

🧑‍💻 Author

• Name: Arshia Kauser Shahzaan
• LinkedIn: linkedin.com/in/arshiakauser
• GitHub: github.com/arshiakauser
• Email: 🔸(Optional — add your professional email)

---

🪪 License

This project is open-source and available under the MIT License.

---

🌟 Acknowledgements

• Dataset preparation support via Roboflow
• YOLOv9c-seg implementation by Ultralytics community
• Streamlit for enabling interactive AI app deployment