Upload README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,61 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
colorTo: red
|
| 6 |
sdk: docker
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: EL Defect Detection System
|
| 3 |
+
emoji: 🔬
|
| 4 |
+
colorFrom: blue
|
| 5 |
colorTo: red
|
| 6 |
sdk: docker
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# 🔬 EL Defect Detection System
|
| 11 |
+
|
| 12 |
+
Production-grade electroluminescence (EL) defect detection for solar PV modules.
|
| 13 |
+
|
| 14 |
+
## Features
|
| 15 |
+
|
| 16 |
+
- **Upload** any EL image (full module, single cell, any brightness/size)
|
| 17 |
+
- **Automatic grid detection** segments modules into individual cells
|
| 18 |
+
- **Deep learning** (U-Net with ResNet encoder) detects defects per cell
|
| 19 |
+
- **Quantitative analysis**: crack length (mm), dark area (%), severity classification
|
| 20 |
+
- **PASS/FAIL decision** with configurable quality thresholds
|
| 21 |
+
- **Visual overlays**: color-coded defect masks on original images
|
| 22 |
+
- **Downloadable reports** (JSON + overlay images)
|
| 23 |
+
|
| 24 |
+
## Defect Types Detected
|
| 25 |
+
|
| 26 |
+
| Color | Defect | Description |
|
| 27 |
+
|-------|--------|-------------|
|
| 28 |
+
| 🔴 Red | Dark/Inactive | Area disconnected from cell circuit |
|
| 29 |
+
| 🔵 Blue | Crack | Micro-crack in silicon |
|
| 30 |
+
| 🔷 Cyan | Cross Crack | Crack at ribbon edge (high importance) |
|
| 31 |
+
| 🟢 Green | Busbar | Metal busbar connection (feature) |
|
| 32 |
+
|
| 33 |
+
## Architecture
|
| 34 |
+
|
| 35 |
+
```
|
| 36 |
+
input image → preprocessing (CLAHE) → grid detection → cell extraction
|
| 37 |
+
→ U-Net inference → mask cleaning → crack/dark analysis → PASS/FAIL
|
| 38 |
+
```
|
| 39 |
+
|
| 40 |
+
## Technical Details
|
| 41 |
+
|
| 42 |
+
- **Model**: U-Net with ResNet34 encoder (segmentation_models_pytorch)
|
| 43 |
+
- **Loss**: 0.5 × Dice + 0.5 × Weighted CrossEntropy
|
| 44 |
+
- **Dataset**: E-SCDD (snt-ubix/e-scdd) — 30 classes remapped to 5
|
| 45 |
+
- **Preprocessing**: CLAHE, percentile normalization, adaptive denoising
|
| 46 |
+
- **Grid Detection**: Projection profiles + FFT periodicity + peak detection
|
| 47 |
+
- **Crack Analysis**: Skeletonization + distance transform + false positive filtering
|
| 48 |
+
- **Dark Detection**: Adaptive threshold (0.6 × mean intensity)
|
| 49 |
+
|
| 50 |
+
## Running Locally
|
| 51 |
+
|
| 52 |
+
```bash
|
| 53 |
+
pip install -r requirements.txt
|
| 54 |
+
streamlit run src/app/app.py
|
| 55 |
+
```
|
| 56 |
+
|
| 57 |
+
## Training
|
| 58 |
+
|
| 59 |
+
```bash
|
| 60 |
+
python src/train.py --encoder resnet34 --epochs 100 --batch_size 8
|
| 61 |
+
```
|