iamrobothahaha/Analyzes_Floating_Solar_Anomalies

☀️ Floating Solar Thermal Anomaly Analysis

  

Detect anomalies on floating solar farms from drone thermal images. Upload DJI thermal photos → the app finds hot spots and bypass-diode faults, pins them on a satellite map using the GPS embedded in each photo, and exports an Excel report plus a Google Earth (KMZ) file.

<!-- TODO: add a screenshot or GIF here, e.g. [image] -->

▶ Try it live: https://huggingface.co/spaces/iamrobothahaha/AnalyzesFloatingSolar_Anomalies

How it works

Thermal image (DJI drone)
        │
        ├─► YOLOv8 (fine-tuned best.pt) ──► bounding boxes + confidence
        │
        └─► ExifTool ──► GPS lat/lon from EXIF/XMP metadata
                              │
        ┌─────────────────────┘
        ▼
  Interactive map (Folium) · Results table · Excel report · Google Earth KMZ

Features

Model

Metrics

<!-- TODO: fill in real numbers. Get them by running: yolo val model=best.pt data=path/to/data.yaml or from runs/detect/train/results.csv of your training run. -->

Usage

• —Single File Analysis — upload one image for a detailed view: original vs. annotated side-by-side, per-detection confidence, GPS map.
• —Multiple Files Analysis — batch process many images and get a full report: summary metrics, anomaly map, filterable table, Excel + KMZ downloads.
• —Confidence slider — only detections above the threshold are shown (default 0.40).

Run locally

# 1. Install Python dependencies
pip install -r requirements.txt

# 2. Install ExifTool (needed for GPS extraction)
#    - Windows: download from https://exiftool.org, rename to exiftool.exe,
#      and place it in the project root (the app auto-detects it there)
#    - Linux:   sudo apt install libimage-exiftool-perl
#    - macOS:   brew install exiftool

# 3. Start the app
streamlit run app.py

Optional — GPU acceleration (local machines with NVIDIA GPU):

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

The Hugging Face Space runs on CPU (free tier); GPU only applies when running locally.

Project structure

├── app.py              # Main Streamlit application
├── best.pt             # Fine-tuned YOLOv8 weights
├── requirements.txt    # Python dependencies
└── packages.txt        # System packages for HF Spaces (ExifTool)

Limitations

• —Trained on a limited dataset — detects 2 anomaly classes only; other defects (soiling, vegetation shading, string outages, cracked cells) are not yet recognised.
• —Reports bounding boxes and confidence, not actual temperatures — radiometric ΔT readout from DJI thermal data is planned.
• —GPS accuracy depends on the drone's metadata; images without GPS tags still get analysed but won't appear on the map.
• —Tested primarily with DJI drone thermal JPEGs.

Roadmap

• —[ ] Sample images folder so visitors can test without their own data
• —[ ] Severity scoring (Critical / Major / Minor) per detection
• —[ ] Radiometric temperature (ΔT) extraction from DJI XMP data
• —[ ] More anomaly classes + larger base model (YOLOv8s / YOLOv11)
• —[ ] PDF inspection report export
• —[ ] Cluster nearby detections into single faults (DBSCAN)

Credits

Built with Ultralytics YOLOv8, Streamlit, Folium, and ExifTool.