Detection / segmentation
YOLO-family pipelines for defects, objects, and general industrial detection.
For line AOI, industrial inspection, and industrial camera boards, Shunter delivers mass-production-ready on-device AI vision algorithms with vision PCBA in one package: data capture, model train–deploy, board design, SMT, and bring-up testing—shortening the gap between “the algorithm demos” and “the line goes live.”
Algorithms, ISP / sensors, and board engineering are co-tuned by one team—avoiding split delivery where “the model is accurate, the image is unstable, and production is inconsistent.” Consultations typically get a response within 12 hours.
Algorithm chain
From data to board: full train–deploy chain
A mass-production industrial detection algorithm delivery path—six coordinated stages from assets to go-live.
-
01
Data capture
Scene samples
Annotation standards -
02
Model training
YOLO / MobileNet
Detection and classification pipelines -
03
Quantize & compress
INT8 quantization
Operator adaptation -
04
On-board deploy
NPU / SoC inference
Latency and power tuning -
05
Imaging bring-up
Co-validate with
ISP / sensors -
06
Mass-prod regression
Version control
Continuous iteration
Algorithm stack
Four algorithm layers working together
Lightweight backbones
MobileNet and similar for edge industrial control and low-power camera products.
Quantization & deploy
INT8 quantization, operator adaptation, on-board inference, and performance / power tuning.
Scenario customization
Iterate on customer line data and conditions into continuously upgradeable model assets.
Board-level chain
From selection to mass production: vision PCBA delivery
Built on Shunter R&D and HOVISION camera vision PCBA capability—industrial vision boards ready for trial production and volume.
-
01
Sensor selection
OV series
Scene and bandwidth match -
02
Schematic design
Power / clock
MIPI interface planning -
03
PCB layout
Up to 16-layer HDI
C-PHY signal integrity -
04
ISP bring-up
AWB / AE / HDR
Optional NPU integration -
05
SMT / assembly
Placement and assembly
AOI / X-Ray -
06
Functional test
Imaging and interface validation
Trial to volume
Board-level stack
Four engineering layers working together
Selection layer
OmniVision OV sensor family—selected for industrial, industrial-control, and camera routes.
Design layer
Schematic and high-speed layout focused on MIPI / C-PHY, power integrity, and thermal design.
Imaging layer
ISP bring-up with optional on-device AI / NPU so imaging and detection algorithms stay aligned.
Manufacturing layer
SMT, AOI/X-Ray, and functional fixture testing for trial-to-volume consistency.
Implementation path
Six steps from requirements to line acceptance
Typical industrial vision projects advance by stage so demo capability can enter firmware and mass-production lines.
-
1
Metrics and defect definition
Lock defect types, accuracy, takt, latency, and power budget.
-
2
Data and board selection
Capture real line samples; confirm sensor / SoC / board form in parallel.
-
3
Training and layout design
Model training evaluation and high-speed link layout proceed in parallel.
-
4
Quantization and first-board bring-up
INT8 on-board adaptation; first-board ISP / model co-validation.
-
5
Line bring-up and trial production
Integrate with AOI / industrial I/O; freeze SMT trial production and test fixtures.
-
6
Mass-production regression and iteration
Model and board version control, sampling regression, and ongoing engineering support.
Case study
Algorithm delivery: detection model pack + board form factors
Train–deploy and PCBA engineering land as mass-production detection model packs and industrial / AI camera boards.
Product capabilities
- Defect and object detection pretrained packs for AOI / IPC / industrial control, fine-tunable on customer data
- Mass-production boards on OmniVision OV series + ISP/SoC (Sigmastar / Rockchip)
- Optional on-board NPU for line-side inference and real-time takt
How this solution lands
- Pretrain + quantize: Ship on-device models that demo and trial-produce
- Select + layout: Mass-production design under MIPI bandwidth and thermal constraints
- Manufacturing test: SMT-to-functional-test closed loop to shorten customer import
Customer value
- Algorithms and boards from one team—less imaging/hardware split
- Same capability extends to warehouse vision and global-shutter industrial cameras
- Aligned with the HOVISION overseas delivery window
More form factors
Where the same industrial vision capability extends
Industrial AOI
Defect detection model packs with customer-sample retrain and on-board regression.
Industrial vision PCBA
Global shutter, multi-cam sync, GigE / USB3 / MIPI for AOI and robotics.
Industrial-control vision bring-up
Line capture and industrial I/O integrated for stable takt acceptance.