LiQung Technology brings EL / IR / PL imagery, spectral signals, and microscopic imaging fully into the deep-learning fold —
integrating YOLOv8 / U-Net / Vision Transformer architectures —
reducing manual re-inspection effort by over 70% with defect detection accuracy ≥98%.
Electroluminescence, Infrared Thermography, and Photoluminescence imagery of solar cells and LED wafers, fed into a YOLOv8 + U-Net dual model. Automatic classification of 12 defect classes (Crack / Finger Break / Black Spot / Dim Cell, etc.) at ≥98% accuracy and <120 ms per-image inference.
Combining AM0 / AM1.5g spectral references with CNN+LSTM models — automated grading of IV curves, quantum efficiency, and spectral matching, producing IEC 60904 Class A/B/C reports. Outlier detection auto-flags anomalies.
Microscopic imagery (10×~1000×) combined with Vision Transformer for surface defect recognition: scratches, contamination, coating delamination, wafer-edge chipping. Multi-view stitching and sub-pixel measurement, repeatability σ < 0.5 px.
Time-series data (intensity, wavelength, temperature, current) fed into an LSTM-Autoencoder to detect light source decay, laser focus drift, CCD dark-current rise. Provides 24–72 h advance warning, cutting unplanned downtime by 60%.
Production-line image / spectral / signal collection with standardised labelling.
PyTorch + GPU-accelerated training, hyperparameter tuning.
Production-line validation of accuracy, false negatives, and false positives.
MES / SCADA / SECS-GEM API integration.
Edge device real-time inference + cloud re-training.
PyTorch 2.x · TensorFlow 2.x
YOLOv8 / Detectron2
U-Net / SegFormer
ViT / Swin Transformer
ONNX Runtime / TensorRT
NVIDIA Jetson / RTX series
Intel OpenVINO
AWS / Azure / GCP
Docker / Kubernetes
MQTT / OPC UA
PostgreSQL / InfluxDB
MinIO / S3 object storage
MLflow / Weights & Biases
Label Studio annotation
Grafana / Prometheus
We deliver a working PoC within 30 days of contract signing.
