#VISIONSTR Sneak Peek, Part II: Revolutionizing Defect Detection with Our EROS SWIR Machine Learning Conveyor Demo at VISION 2024
EROS - 5GigE Camera (Emergent Vision)

#VISIONSTR Sneak Peek, Part II: Revolutionizing Defect Detection with Our EROS SWIR Machine Learning Conveyor Demo at VISION 2024

As we continue our countdown to VISION 2024, we're thrilled to unveil another cutting-edge demonstration that showcases the synergy of advanced imaging and machine learning: the EROS SWIR HE-25000-SBL & HE-5300-SI Machine Learning Conveyor Demo.


Advanced Defect Detection Using Machine Learning

This demo features a powerful combination of:

  • 2x 5GigE HE-25000-SBL Cameras: High-resolution cameras based on the Sony IMX540 sensor, ideal for capturing detailed images at high speeds.
  • 2x 5GigE SWIR HE-5300-SI Cameras: Short-Wave Infrared (SWIR) cameras based on the Sony IMX992 sensor, capable of revealing material properties not visible in the standard visible spectrum.

By integrating these cameras, we achieve a comprehensive imaging solution that captures both visible and SWIR spectra, enhancing defect detection capabilities.

Machine Learning at Its Best

Utilizing machine learning libraries such as PyTorch, we've trained models to detect and report defects on objects like blister packs in real time. Here's how it works:

  • Data Capture: The cameras capture high-resolution images of products on a conveyor belt.
  • Direct Image Transfer: Using GPU Direct, images are transferred directly from the cameras to the GPU, bypassing the CPU to reduce latency.
  • Real-Time Processing: The GPU runs our in-house trained machine learning models to analyze images and detect defects instantaneously.
  • Results: Defective items are identified in real time, allowing for immediate action.

eCapture Pro Plug-In for Rapid Prototyping

This demo leverages our eCapture Pro plug-in feature for rapid, high-performance prototyping. Key advantages include:

  • Zero Copy and GPU Direct: Enhances data transfer efficiency, reduces CPU load, and minimizes latency.
  • Flexible Development: Allows for quick deployment and testing of machine learning algorithms.
  • Seamless Integration: Compatible with various hardware setups, making it adaptable to different operational requirements.

Versatile Hardware Compatibility

One of the standout features of this demo is its ability to run on various hardware configurations, such as:

  • NVIDIA IGX Orin
  • NVIDIA AGX Orin
  • Standard Computers: Even on lower-end computers based on an ASUS motherboard with an Intel i7 processor.

This flexibility ensures that high-performance machine learning applications are accessible without the need for specialized or expensive hardware.

Why SWIR Imaging Matters

The inclusion of SWIR cameras (HE-5300-SI) adds significant advantages in defect detection:

  • Enhanced Material Inspection: SWIR imaging can penetrate certain materials, revealing hidden defects such as cracks, voids, or contaminants.
  • Non-Destructive Testing: Allows for inspection without damaging the product.
  • Broad Application Range: Useful in industries like pharmaceuticals, electronics, food inspection, and more.

Real-World Applications

This demo is not just a theoretical showcase; it has practical implications across various industries:

  • Pharmaceuticals: Detect defects in blister packs, ensuring product safety and quality.
  • Manufacturing: Identify imperfections on assembly lines, reducing waste and improving efficiency.
  • Food Industry: Spot contaminants or foreign objects in food products.
  • Electronics: Inspect PCBs and components for defects not visible to the naked eye.

Technical Advantages

  • High-Speed Imaging: Capture images at high frame rates without compromising quality.
  • Real-Time Analysis: Immediate defect detection allows for on-the-fly adjustments.
  • Scalability: Suitable for small setups or large industrial applications.
  • Cost-Effective: Runs efficiently on various hardware, reducing the need for costly equipment upgrades.

See It Live at VISION 2024

We invite you to witness this groundbreaking demo in person at VISION 2024. Experience how the integration of advanced imaging technology and machine learning can revolutionize defect detection and quality control in your operations.

Join us at VISION 2024 in Stuttgart, Hall 8, Booth 8D60.

Don't Miss Out!

Stay tuned for more sneak peeks as we approach the event. We're excited to share more of our innovative solutions that are set to redefine standards in industrial imaging and machine learning applications.

We can't wait to see you there!

#EmergentVision #VISION2024 #MachineLearning #DefectDetection #SWIRImaging #Sony #HighSpeedImaging #eCapturePro #IndustrialAutomation #GPUDirect #ZeroCopy #Innovation

Feel free to share this post or tag colleagues who might be interested.

More information:


Things change quickly in our developments. Eros was planned to be 5GigE in release but provisioned for 10GigE. Turns out that at 29mm x 29mm Emergent will present at the Vision Show (VISION) the smallest and lowest power Eros 10GigE cameras in both RJ45/10GBaseT/POE and SFP+ forms. Power consumption of the 10GBaseT model with POE is about 4.8W and without POE 4W. For the SFP+ single mode fiber model, the power drops to an astonishing 3W. Will also be lowest cost! All this while running the Sony IMX540 25Mpix sensor at 30fps. This beats ALL providers of 5G and 10G cameras with a substantial margin. Couple this with our demo of our 100G Zenith HZ-65000G running at 71fps and utilizing GigEVision 3.0 RDMA RoCEv2. Fastest demo using GigEVision 3.0 proposal. Couple this with multiple GPU direct demos including a multi-camera Machine Learning blister pack inspection. Couple this with a 36 x 10G camera , single server, real time 3D reconstruction demo - what fun! Couple this with the fastest line scan demos out there with 100G Pinnacle LZT-9KG5 TDI at 600KHz and LZ-16KG5 16K at 400KHz. Come visit Emergent with Thomas Detjen, Christian Loeb, and myself John Ilett at the nexus of the universe. Messe Stuttgart – Hall 8 Booth 8D60

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了