DEEP LIGHTING, the multi-LED dome that corrects specular reflections
DEEP LIGHTING is a multi-LED, multidirectional dome that, through a set of algorithms developed by the Deep Capture team, corrects issues with specular reflections (bright spots) on parts with highly reflective properties (lacquered, chrome, steel, stainless steel, etc.).
All light source positions are calibrated relative to one another. Defects are displayed using the developed algorithms, and the images are cleaned of any remaining unnecessary information.
The photos are captured using a high-performance image sensor based on CMOS technology, operating at a framerate of 60 FPS. This allows for the detection of small surface defects (less than 0.1 mm) that are sometimes invisible to the naked eye.
How does the detection process work?
The use of Deep Lighting technology operates through supervised learning on our detection software, DEEP CAPTURE.
Detection
Operating through supervised learning on our DEEP CAPTURE detection software, the model relies on a dataset of several thousand photos that comprehensively represent all defects.
- Our online labeling tool, DEEP LABEL, allows our teams to take control remotely and enables collaborative work on expanding the dataset.
- Segmentation tools allow the definition of defect acceptance levels based on their size and quantity on the pen.
Deep learning industrial vision
What is it? Why is it more efficient?
Deep learning (or deep learning) is a set of techniques based on artificial neural networks. This underlying paradigm is revolutionizing the way applications are developed, particularly in the field of industrial vision. We’ve moved from developing deterministic models, validated by industry data and experts, to developing non-deterministic models that learn to solve problems based on a large amount of industry data and successive trials. The advantage of using these techniques is the model’s ability to learn to recognize invisible or unconscious features that a software developer could not easily incorporate into a traditional industrial vision model.
The very construction of these neural networks allows the model to adapt and recognize objects it has not been specifically trained on. This feature makes deep learning an essential tool in industrial vision for detecting various defects in a controlled production environment.
