🔥 Helioscan – AI-Driven Detection of Solar Panel Anomalies via Thermal Imaging

Model: YOLOv9
Domains: Renewable Energy | Thermal Vision | Computer Vision | FastAPI | YOLO

---

🌍 About the Project

Photovoltaic (PV) systems are a cornerstone of sustainable energy solutions. Yet, their performance can suffer due to the absence of affordable and automated real-time inspection tools. This project presents a full AI pipeline designed to automatically detect thermal defects in solar panels using YOLOv9, delivered through a REST API powered by FastAPI and containerized with Docker.

Goal: Identify and classify 8 distinct thermal fault types in PV modules using labeled grayscale drone imagery.

---

📦 Dataset Details

• Raw Images: 2,723
• Post-Augmentation Samples: ~7,500
• Total Annotations: 7,772
• Label Format: YOLOv9 (Roboflow-generated)
• Resolution: 640×640
• Color Format: Grayscale

---

🔄 Augmentation Techniques

• 50% probability of horizontal and vertical flips
• Random rotations at 90° angles
• Shear transformations (±15°)
• Adjustments to hue, brightness, and exposure (±15%)

---

🧪 Fault Categories

• Single Hotspot
• Multiple Hotspots
• Single Diode Fault
• Multiple Diode Faults
• Single Bypassed Substring
• Multiple Bypassed Substrings
• Open Circuit (String)
• Reversed Polarity (String)

---

⚙️ Workflow Summary

1. Conduct research on thermal faults and define taxonomy
2. Collect and annotate data using Roboflow
3. Preprocess (resize, grayscale, align)
4. Apply data augmentation
5. Train YOLOv9 model and evaluate performance
6. Export final model as Th_G_v9.pt
7. Deploy API using FastAPI + Docker

---

📊 Performance Metrics

Metric	Result
Precision	74%
Recall	76%
mAP@0.5	78%
mAP@0.5:0.95	61%
Fitness Score	63%

The YOLOv9 model performed well on unseen thermal images.

---

🧠 API Info (FastAPI)

Endpoint: POST /predict/
Input Format: Grayscale thermal image (form-data)
Response Includes:

• 🖼️ JPEG with detection boxes
• 📦 HTTP headers: X-Anomaly-Count and X-Detection-Result

---

🐳 Running the Project with Docker

You can run the API in an isolated and reproducible environment using Docker.

🔧 1. Build the Docker Image and Run the Container

From the root directory of the project, run:

docker build -t solar-api .

docker run -p 8000:8000 solar-api


### 🔁 Sample curl Request

```bash
curl -X POST "http://localhost:8000/predict/" \
  -H  "accept: image/jpeg" \
  -F "file=@sample.jpg" \
  --output result.jpg