Focusing on machine vision+industrial AI dual technology engine, CY vision BU provides customizable, standardized and one-stop optical-mechanical soft computing vision solution centering on four scenes of industrial manufacturing positioning guidance, surface inspection, ID identification and high-precision measurement.
Industry Challenges
- Scarcity of defect samples – New projects often lack NG samples, extending delivery cycles.
- Limitations of traditional machine vision – Struggles with distortion and variable defects, leading to high miss rates and false positives.
- Complex multi-tool integration – Multiple software tools require manual image transmission and debugging, making deployment cumbersome.
- Scattered production lines & inconsistent hardware – Wasted computing resources across diverse inference devices.
- Frequent changeovers – Model tuning is time-consuming, slowing down iteration and deployment.
Solution & Core Advantages
From algorithms to deployment, from samples to accuracy — CY Industrial AI delivers end-to-end value for complex industrial scenarios.
Core capabilities – No-code visual annotation, toolchain orchestration, hyperparameter tuning, multi-algorithm integration.
Value – Enables non-algorithm engineers to build detection models quickly, reducing project training cycles by up to 70%.
Core capabilities – Multi-GPU / multi-stream parallelism, cloud-edge deployment, one-click start/stop of pipelines.
Value – Shared computing power across multiple production lines, reducing hardware costs by up to 40%.
Core capabilities – Batch simulation of defects via text-to-image / image-to-image, one-click dataset export.
Value – Minimizes on-site sample collection, accelerating new project launch by 50%+.
Full-scenario coverage
Image classification, semantic segmentation, anomaly detection, object localization, OCR (character detection & recognition), and toolchain integration covering all core industrial inspection tasks.
Few-shot learning + defect simulation
High-fidelity defect generation based on Stable Diffusion. Anomaly detection can start with only OK images, drastically shortening delivery timelines.
Breakthrough inference performance
Three deployment modes (thread-embedded, process-isolated, cloud inference for multi-station sharing). Supports full NVIDIA RTX series and Apple M-series chips.
Full-chain delivery with industrial know-how
Integrated delivery of optics, mechanics, electronics, software, and AI. One-stop service from annotation, training, tuning to deployment. 100% end-to-end capability, backed by years of production line experience.
Real-World Case: Keyboard Inspection
Customer requirements:
UPH > 300, CT < 5s, 0% miss rate, false positive rate < 2%.
Results achieved:
0% miss rate, 1.8% false positive rate, inference time 0.9s — fully replacing the system from a major international company (Company C).
The proprietary toolchain system automatically links image samples and algorithm tools, requiring no manual coding. The inference system supports concurrent multi-pipeline execution, tailored for industrial scenarios as a more streamlined alternative to NVIDIA Triton.
Industries & Applications
CY Industrial Vision AI has been rigorously validated across diverse complex processes:
Keyboard Surface Inspection
10+ defect types (hook deformation, missing, burrs, etc.). 0% miss rate, <2% false positives, inference 2.9s — far superior to Company C.
Image Classification
Semantic Segmentation
Metal Parts
Missing inserts, dents, scratches. Miss rate <0.05%, false positives <10%. Semantic segmentation for precise defect localization.
Image Classification
Semantic Segmentation
Diffusion Film Inspection
Max FOV 750mm, defects: black/white dots, foreign material, pressure marks. Miss rate <0.3%, false positives <8%.
Semantic Segmentation
Secondary Analysis
Sun Gear Inspection
Inner/outer diameter, end-face defects. Miss rate <2%, false positives <5%. Strong resistance to dirty backgrounds.
Semantic Segmentation
Complex Background
Pharmaceutical Pill Inspection
Cracks, chips, black spots. 0% miss rate and 0% false positives.
Object Detection
Semantic Segmentation
FPC AOI re-inspection
False call removal rate >95%, true defect miss rate 50 PPM. Significantly reduces labor costs.
Semantic Segmentation
Secondary Analysis
0%
Keyboard project missing detection rate
<2%
Comprehensive over-rejection rate
0.05%
Iron part missing detection rate
95%+
FPC false spot removal rate
30min
Model training time consumption
Flexible Deployment Architecture
CY Industrial AI delivers end-to-end value for complex industrial scenarios
Dataset Management
Defect generation / model training
AI Cloud Inference Server
TCP/HTTP · multi-station concurrent requests
Edge Inference Node
Nvidia GPU / Apple M
AI-Desktop/SDK
No-code model buildingC# API