Ryanflash/herislab-ca-training-data

---
language:
- en
license: mit
task_categories:
- image-classification
tags:
- thermal-imaging
- anomaly-detection
- autoencoder
- electrical-fault-detection
- infrared
size_categories:
- 10K<n<100K
---

# CA_Training_Data -- Convolutional Autoencoder (Track A)

Curated dataset for training and evaluating the Convolutional Autoencoder anomaly detection model.

## Approach

The autoencoder is trained **only on normal (no-fault) images**. At inference, high reconstruction error indicates an anomaly/fault.

## Structure

```
train/normal/ -- Normal images for autoencoder training
electric_motor/ -- 168 PNG (Electric Motor Thermal Fault Diagnosis, no_fault class)
induction_motor/ -- 20 BMP (Thermal Images of Induction Motor, Noload class)
pv_om_inspection/ -- 7,836 TIFF (PV System O&M Inspection, double-row + single-row)
pv_thermal_inspection/ -- 1,075 TIFF (PV System Thermal Inspection)
solar_modules/ -- 2,302 JPG (Infrared Solar Modules, No-Anomaly class)

test/normal/ -- Held-out normal images for threshold calibration
electric_motor/ -- 28 PNG
induction_motor/ -- 5 BMP

test/fault/ -- Fault images for evaluating anomaly detection
electric_motor/ -- 173 PNG (Electric Motor Thermal Fault Diagnosis, fault class)
induction_motor/ -- 344 BMP (Thermal Images of Induction Motor, 10 fault conditions)
```

## Total Counts

| Split | Normal | Fault | Total |
|-------|--------|-------|-------|
| Train | 11,401 | 0 | 11,401 |
| Test | 33 | 517 | 550 |

## Source Datasets

| Directory | Source Dataset | Domain |
|-----------|--------------|--------|
| electric_motor | Electric Motor Thermal image Fault Diagnosis DATASET | Electrical (primary) |
| induction_motor | Thermal Images of Induction Motor Dataset | Electrical (primary) |
| pv_om_inspection | Photovoltaic System O&M inspection | Solar PV (adjacent) |
| pv_thermal_inspection | Photovoltaic system thermal inspection | Solar PV (adjacent) |
| solar_modules | Infrared Solar Modules (No-Anomaly only) | Solar PV (adjacent) |

## Notes

- PV O&M files are prefixed `dr_` (double-row) and `sr_` (single-row) to avoid filename collisions
- Solar module images were filtered from module_metadata.json (anomaly_class == "No-Anomaly")
- Test/normal hold-out is ~14-20% of electrical equipment normal images
- Image formats are mixed (PNG, BMP, TIFF, JPG) -- preprocessing/normalization is required before training