What is the scope of this solution? Can be said to be the category of visual system, can also be said to be the category of visual detection equipment. Solution is a complete solution based on algorithm superposition. Our internal view is that AI is not just a technical module, it is a new cognitive framework that is essentially data-driven and standards-driven. First we have to have this cognitive framework, then we go down to our visual solution, what are the core components? What does it mean for these parts?
I extracted the three core parts:
1. Imaging module
Imaging module is the whole machine vision in the imaging of all the devices and programs, what is the basic principle behind it? Is based on traditional algorithms, which are based on quantitative analysis. So we based on the traditional algorithm to do the imaging scheme, its underlying requirements are “quantitative, high contrast”.
Where does this lead? Let’s say we test a surface for many different types of defects. In order to achieve high contrast quantification, maybe I need to do a few shots. Each light may correspond to two kinds of defects, and only later can these defects be fully presented, so the imaging efficiency is very low.
As we enter the age of AI, our requirements for imaging change, as long as it’s visual. The current imaging scheme based on traditional algorithms is still a “photoelectric converter” in essence. It’s just translating a critical signal into an image, and it’s too far away from what we call an eye. Of course, we can not step into the eyes, that at least the stage of the goal we can not reach the level of photography. What’s the benefit of this? On the one hand, it can improve the spatial efficiency of our entire imaging, and more importantly, it is simplified, universal, and low cost. This is an important fundamental change.
2. Algorithm module
Objectively speaking, the current implementation of various projects, the cost is relatively high. The root cause is that most of the AI as an algorithm module, superimposed into the original system, is inefficient. The subsequent algorithm scheme must take AI as the center to get through and optimize the entire computing flow and data flow. This is the optimal way, which can improve the efficiency of training reasoning and reduce the deployment and maintenance cost.
3. Automation module
In the age of traditional algorithms, automation was limited because of the constraints on imaging. AI is actually breaking the shackles of algorithms, and essentially breaking the shackles of our imaging. It can automatically help us take pictures, all kinds of “concave pose” and “pose”. As long as you can picture the defect clearly, you don’t need a bright image. If we do this, we can greatly reduce the automation complexity and improve the universality of automation. And can be relatively simple and efficient to solve the product shape, multi-model small batch imaging problems.
On a large scale, industrial vision solutions evolve in two extreme directions:
Lightweight scenarios: More integrated, extremely easy to use, may require online training.
Complex scenarios: More emphasis on universal solutions, including: general imaging module, general large model, general automation module, reduce the comprehensive cost of the whole link.