Thursday, 5 March 2026
Optics and high precision: on a food production line, green beans fall at high speed. They all look identical. Yet some must be rejected according to precise dimensional criteria. The challenge here is not just to see them, but to measure them in motion, without slowing down production.
Optics and high precision: on a food production line, green beans fall at high speed. They all look identical. Yet some must be rejected according to precise dimensional criteria. The challenge here is not just to see them, but to measure them in motion, without slowing down production. It is in this kind of situation that Psycle's machine vision expertise comes to the fore. And it is precisely on these issues that Martin D., vision and AI engineer, is working; he joined in November 2025 to strengthen the company's optical and algorithmic expertise.
A PhD in optics specializing in photonics, Martin brings a detailed understanding of light phenomena. His background, initially focused on fundamental research, now finds direct application in industry. At Psycle, based in Lacroix-Saint-Ouen, he works at the point where imaging systems, software development and artificial intelligence converge.
A usable image above all
Artificial intelligence cannot make up for a poorly controlled image. It relies on an image built with precision. In the case of our green beans, the work begins by analyzing the real constraints: conveying speed, free fall, variable lighting, integration on a conveyor. Martin then determines the right sensor, calculates the field of view, anticipates the depth of field and simulates the optical parameters to obtain an image reliable enough to be interpreted by the algorithm.
And this methodical approach applies to every sector. In the food industry, the variability of living products complicates analysis, because one product resembles another without ever being strictly identical. Conversely, in the automotive industry, each part must conform to a theoretical model without excessive tolerance. In both of these very different cases, machine vision requires fine-tuning of the optical system and the analysis software.
Fine-tuning to reduce losses
On a biscuit packaging line, bagging defects cause significant losses: badly cut sachets, doubled products and/or trimmed packaging. This leads to production delays, outright losses and other problems that require Psycle's intervention. This was followed by the design of an end-of-line quality control system, combining a vision camera and custom software developed in-house. The algorithm was finely calibrated to automatically detect non-conformities and thus make production more reliable.
“Machine vision does not eliminate jobs, it makes them evolve. Repetitive visual inspection gives way to supervising an intelligent system. Operators run the computer, analyze indicators and step in when there is a deviation. Value therefore shifts toward analysis and understanding of the process.”
Detecting the invisible
More recently, Psycle has worked in the world of premium spirits. The challenge set by SAVERGLASS is to inspect decorated bottles with curved, reflective surfaces. A micro-scratch, a printing defect or a decorative imperfection can compromise visual quality. Martin therefore analyzes the samples in-house, tests different lighting configurations and studies the possibility of using several cameras to cover the entire surface.
The difficulty here lies in the consistency of the imaging system. The image must be precise enough to allow the algorithm to identify defects invisible to the naked eye, while remaining suitable for integration on an industrial line. This study phase draws on both Martin's optical expertise and the team's skills in software development and artificial intelligence.
AI and technological evolution
The core of Psycle's product is software developed specifically for each customer. The artificial intelligence is trained on real data, optimized, then stabilized to guarantee repeatability and compliance with quality standards. Martin contributes to this skills development, particularly on the optical side and the interaction between image and algorithm. His role is as much about producing a usable image as understanding how the computer will interpret it.
On-site integration is then the reality check. Variable lighting conditions, unforeseen mechanical constraints, sustained industrial speeds: every parameter must be validated in real conditions. Psycle provides regular follow-up after commissioning, to secure performance over time.
At the same time, Martin keeps an eye on new technologies and explores them. He tells us in particular about SWIR, based on short-wave infrared, which gives access to additional invisible information.
For Martin, these developments broaden the field of machine vision and open up new prospects for material analysis. The sector still has great discoveries ahead, and Psycle is far from done evolving.