Artificial vision is everywhere in our lives. You can find it in automated passport scans at airports, security systems for your smartphone, and even in parking lots to read your license plate. Computers are constantly watching us, and this is not a bad thing. This sort of “Big Brother” system allows computerized systems to take decisions quickly and efficiently, saving a lot of human time and effort in complex tasks.
Computer vision is also relevant to a manufacturing production line for the same reasons. Imagine a factory that produces wheels for cars. The quality assessment of the wheels could be done manually by an employee, but this also can be done by computerized artificial vision systems. Such systems are able to detect the tiniest of fractures and irregularities in the wheel. Computer vision is a key element of production lines in factories today.
Although artificial vision provides some advantages to improving production, the current technology is not sophisticated enough for advanced fabrication lines. One challenge is low-quality images due to older camera hardware, poor lighting, vibrations, and factors present in most manufacturing facilities. In addition, an artificial vision system typically needs a very high level of accuracy to detect defects, as well as the ability to explain the results. Imagine going to your boss saying that you don’t know why a machine rejected a part!
Multiverse Computing, a provider of value-based quantum computing solutions, has run a pilot project together with manufacturing company Ikerlan to assess the performance of quantum computer vision in detecting defects of pieces in factory production lines.
Quantum computers are the holy grail in algorithmic science: They hold the promise of faster, more efficient, and more accurate calculations compared with today’s conventional computers because they manipulate information according to the laws of quantum physics. Quantum computing algorithms can detect anomalies in subtle patterns of data that a conventional computer would never be able to find. Quantum computers are also able to extract more information with less data, as proven in an article recently published in Nature Communications.
The Multiverse Computing and Ikerlan teams have demonstrated that today’s quantum computers, despite their hardware limitations, perform better than classical computer vision systems in finding defects in manufacturing images. The results, published in a research paper on arXiv, show that quantum computing vision outperforms classical counterparts in precision, even while maintaining the training and inference times of the algorithms. From a broader perspective, this is one of the first landmark results showing that quantum computers can offer real business value for the industry today for certain specific applications.
Proving the power of quantum computing
Multiverse and Ikerlan used a dataset of 2,727 X-ray images of automotive parts with and without casting defects to train a quantum artificial vision system to see whether an image has a defect or not. This is an incredibly hard task for computer vision systems due to the low quality of some images and the intricacy of defects, but it is the necessary first step to pinpoint flaws like a tiny fracture in a car wheel that could eventually lead to an accident. The study concluded that some of today’s commercial quantum computers can run quantum classification algorithms that systematically detect such defects with higher accuracy (between 10% and 20% more) than traditional explainable machine-learning methods, and with similar inference time. Multiverse algorithms were also explainable, in the sense that a human could understand the reasons behind the decisions made by the algorithm. This is a must in many industrial applications.
In addition to outperforming classical computers, these quantum algorithms from Multiverse learned to spot defects faster than other advanced algorithms currently being tested.
When compared with the best deep-learning algorithms by Amazon Solutions Lab, researchers found that the Multiverse/Ikerlan quantum algorithm obtained essentially the same accuracy but was 24× faster to train, with the time required going from almost two-and-a-half hours to less than six minutes. This speedup is key in environments where the algorithm must adapt to different circumstances on the fly. So overall, not only are the quantum vision systems explainable but more accurate than their classical counterparts and significantly faster than deep learning.
The results from the Multiverse and Ikerlan collaboration are the first example of a quantum computer vision system applied to a manufacturing line. These findings prove that quantum computing has quality control value in factories today. This is only the tip of the iceberg: Next time you go through an automated passport control, there might be some quantum physics behind the analysis!
Please find the original article by EE Times Europe here