How Teledyne FLIR and Neurala Collaborate to Make Deep Learning Implementation Faster, Easier, and more Cost Effective

The Challenge and Opportunity

Deep Learning is a powerful tool for machine vision engineers and OEM system designers looking to quickly automate complex and subjective decision making. However, this technology can be prohibitive for non-experts due to the need for multiple software tools, large datasets, specialized skills of developers, and their associated costs.

To overcome these challenges, FLIR Systems, Inc. has collaborated with Neurala to provide an end-to-end development and deployment solution specifically for non-experts. Our cost-effective solution also requires much smaller datasets.

Faster, Easier, and Cost Effective

Typical development and deployment of deep learning for machine vision systems entail several specialized resources and multiple steps. To name a few:

• Access to a highly skilled workforce with machine learning expertise.
• Multiple tools for image management, annotation, neural network training, and accuracy verification as well as managing the transition between these tools
• A separate CPU or GPU to run the trained neural network

FLIR and Neurala’s end-to-end development and deployment solution replaces these steps with a fully supported one-stop solution.

• Easy to Use: Neurala VIA (visual inspection automation) SAAS software walks you through all the necessary steps to manage and annotate images, train the model, and produce an effective deep learning inspection neural network model in minutes - you don’t need to be a machine vision or AI expert.
• Fully Supported End-to-End Solution: FLIR and Neurala have jointly developed a seamless workflow that replaces reliance on multiple software tools; delivering a dependable one stop solution.
• Minimal Data Requirements: Train with dozens, rather than thousands of images, so you can get started quickly, measure performance and update your model over time as you collect additional data.
• Cost Effective: By deploying the Neurala VIA model directly on the Firefly DL camera, there is no need for a host or peripheral hardware. In terms of software, overall development and deployment costs of other commercially available solutions often run into tens of thousands of dollars; while our solution achieves similar results with 3x savings.
• Saves Time: With Neurala’s VIA and FLIR Spinnaker SDK, you can rapidly prototype, develop, validate and deploy models without the need to setup a traditional AI development environment. Neurala’s VIA software produces a neural network that can be exported to a FLIR Firefly DL camera using our FLIR Spinnaker SDK, making deployment quick and simple.

Example Use Cases

The combined FLIR and Neurala solution is suitable for a wide range of classification use cases; from defect detection in semiconductor fabs to agricultural produce inspection at fruit processing plants.

• Surface Defect Detection: Identify cracks and defects on a circuit board inspection line at a mobile components’ manufacturing plant.
• Part Kitting Verification: Flag incomplete packages or missing components in a kit being inspected as part of a final inspection line before shipment.
• Natural Product Inspection: Quality inspection of produce and better grading of fruits based on bruising, colour, size etc.

FLIR + Neurala’s Brain Builder Dashboard GUI.

FLIR + Neurala’s Evaluation Dashboard GUI to Validate Accuracy

FLIR Firefly DL Camera Specifications

More information

FLIR Firefly DL: https://www.flir.com/products/firefly-dl/  

Contact FLIR at: [email protected]  

Neurala: https://info.neurala.com/via-and-flir-firefly-dl

Contact Neurala at: [email protected]

*Specifications are subject to change without notice.