Qualcomm Dragonwing Visual Inspection brings Edge Impulse AI to factory quality control

Edge Impulse is best known as a machine learning operations platform for building, training and deploying AI models on edge devices. At Micro Electronics USA, the company showed how that model-centric workflow is being extended into a more complete industrial solution with Qualcomm Dragonwing Visual Inspection.

The idea is straightforward. Many manufacturers want to use computer vision for quality control, but the model is only one part of the system. A working factory inspection setup also needs cameras, edge compute, application software, user management, site management, device management, a human-machine interface (HMI), inspection history and a process for reviewing errors. Edge Impulse is trying to package more of that workflow into a vertical solution rather than leaving customers to build everything around the model themselves. Edge Impulse describes the product as a complete AI-based computer vision solution for smart quality management.

Why visual inspection is the first vertical

In the interview, Tyler Hrycak explained that Edge Impulse still supports broad AI development across time-series, sensor and vision data. The change here is not a replacement of the core platform. It is an additional layer for specific vertical use cases.

Visual inspection is a logical first target. Manufacturing teams already use cameras and inspection systems, but many still struggle to move AI from pilot projects into production. Defect detection, assembly verification and label inspection all need reliable models, but they also need a production workflow that non-machine learning specialists can operate.

That is where Qualcomm Dragonwing Visual Inspection fits. Edge Impulse lists key use cases including optical character recognition (OCR), label inspection, line clearance, defect detection and assembly verification. It also highlights no-code model selection, human-in-the-loop continuous learning and remote management as part of the solution.

What the demo shows

The demo centred on printed circuit board (PCB) inspection. A camera captured small boards with deliberate defects, including scratched microcontrollers and missing terminal blocks. The system then displayed a live camera feed, object detection bounding boxes, defect classification and system information through a web-based HMI.

The hardware shown was the Advantech ICAM-300. Advantech describes this as a perception AI camera based on Qualcomm QCS6490, with a 5 MP global shutter image sensor, autofocus lens, LED illumination, low-power AI processing and an IP66 compact housing. That makes it a realistic fit for harsh industrial environments where AI needs to run close to the camera rather than in a remote cloud system.

The local HMI is important. It means inspection results can be viewed directly from the deployed system, including timestamps, detected objects, inference time and processing time. In practice, that gives operators a clearer way to inspect, review and understand model outputs.

Cloud training, local inference

Edge Impulse has traditionally trained models in the cloud, and Tyler confirmed that this demo still uses cloud training. The inference, however, runs on the edge device. That split is typical for many industrial AI workflows. Training can happen centrally, while production inference runs locally for lower latency, reduced bandwidth and better operational control.

Edge Impulse’s Qualcomm page also highlights a broader workflow around data preparation, AI experimentation, optimisation and deployment on Dragonwing processors. It also points to Qualcomm AI Hub integration for profiling and improving model performance on Dragonwing hardware.

Tyler also discussed future on-premise options, where training and storage could run on a larger local appliance while smaller edge devices handle inference. That would be especially relevant for manufacturers with strict data control requirements or sites that cannot rely on cloud connectivity.

Why this matters for engineers

The strongest part of Qualcomm Dragonwing Visual Inspection is not just the model. It is the fact that Edge Impulse is addressing the surrounding production system. Engineers can use the platform to reduce the amount of custom integration needed between model development, edge deployment, inspection review and device management.

That matters because many factory AI projects fail somewhere between a successful model demo and a scalable production deployment. By narrowing the scope around visual inspection, Edge Impulse can provide more of the tooling needed for a real use case rather than a generic AI workflow.

For manufacturers looking at defect detection, assembly verification or inspection automation, Qualcomm Dragonwing Visual Inspection is a useful sign of where edge AI is heading. The industry is moving from “can we train a model?” to “can we deploy, manage and improve this system in production?” That is a much more valuable question for engineering teams.