Top 10 Deep Learning application types in industrial vision systems

deep learning applications

Top 10 Deep Learning application types in industrial vision systems

Deep Learning technology in Machine Vision

Deep Learning is a new reliable solution for machine vision problems that could not have been solved before. Last time I wrote about untypical vision solutions, so this time I will focus on the common industrial deep learning applications. In this article I will show you solution types for automotive, PCB and food industry, but, in my opinion, deep learning tools can be used in almost every manufacture.

Deep Learning vision application examples

1. Anomaly detection in food inspection

Food industry is very specific and problematic for traditional computer vision algorithms, because each product has some acceptable differences. That’s why  Deep Learning is very useful in food industry. Algorithms like anomaly detection are able to learn how good product looks and find any anomalies.

deep learning food industry

Deep Learnnig  application for detecting anomalies on cookies. Source

2. OCR on metal parts

The most challenging OCR (optical character recognition) applications like laser etched codes on electronic components, low contrast characters, embossed characters on injection molded products and dot peens codes on metal parts are often not feasible using traditional machine vision methods. Deep Learning algorithms are really efective in those examples. They are able to extract text from background even in very difficult and changing conditions.

Deep learning OCR

Deep Learning OCR on metal parts. Source

3. Assembly verification in automotive

Vision assembly verification in automotive industry could be a topic for a separate article. Detecting and localizing parts sometimes can be challenging, but in most cases feasible with traditional machine vision. So why Deep Learning is so popular for those applications? Usually there are many different types of the same product, in different color variants, with many variants of buttons, etc. As a result, creating these applications using Deep Learning is simply much faster.

Vision door assembly verification

Door assembly verification using Deep Learning methods.

4. Anomaly detection in textile inspection

The main problem in vision textile inspection is the pattern and fabric types complexity. There are several types of algorithms, which can be used to detect anomalies in textiles, but basically only Deep Learning gives us confidence that every type of defect will be found. Anomaly detection algorithms are very often used for this type of application because defect-free images are enough to train the Deep Learning model.

Deep Learning in textile inspection

Deep Learning application for detecting anomalies in textiles. Image source.

5. Detecting scratches on metal parts

In automotive industry many components have complex shape and, what’s more, they are made of metal. This combination often makes inspection using traditional machine vision algorithms impossible. Deep Learning supervised tools / feature detecction tools are ideal for those applications. The best algorithms can detect even several pixel defects with almost 100% efficiency.

Deep Learning in automotive

Production line in automotive. Image source.

6. Anomaly detection on plastic objects

As I mentioned before, Deep Learning algorithms are perfect for inspecting objects, which geometry is variable. Those objects are difficult to determine by parametric models. Plastic objects made in injection moulding process are such irregular objects.

deep learning application for injection moulding

Anomaly detection of plastic objects. Source

7. Part classification in automotive industry

Reason for using Deep Learning for part classification is the same as in case of assembly verification in automotive. In other words, Deep Learning is used to reduce the time needed to prepare the algorithm for the vision system. Another advantage is that when classifying many types of objects, Deep Learning performs much better than traditional algorithms.
rimd classification deep learning applications

Deep Learning application for classification objects from automotive. Source

8. Deep Learning applications for counting objects in food industry

Objects in food industry are highly irregular, and some of them just can’t be counted using traditional computer vision algorithms. The use of deep learning is the only sensible machine vision solution in this case. Counting tagliatelle can be a great example here.
food industry deep learning

Deep Learning algorithm concept for pasta counting.

9. Parts localization and segmentation on the PCB

One of the most challenging tasks in computer vision is to segment small components, which occur in several dozen variants. Deploying such an application without deep learning algorithm could take months or even years. So it’s hardly surprising that the location and segmentation of components on PCBs is one of the most popular Deep Learning applications in the industry.
PCB deep learning

Parts localization and segmentation on the PCB using Deep Learning algorithms. Source

10. Packing completeness in food industry

Deep learning application chocolate praline

Deep Learning based application for checking completeness of box of chocolates. Source

After checking quality and quantity, it is time to check packing completness. In food industry it is not only important to check the number of packaged products, but also their arrangement. Who would like to buy sushi box with mixed and damaged pieces? Not me.
deep learning application sushi

Deep Learning based application for checking completeness of sushi boxes. Source

Deep Learning applications in industrial vision systems

The development of artificial intelligence technology, especially Deep Learning, allows the creation of many previously unattainable solutions. As you can see, Deep Learning applications are becoming increasingly more popular in machine vision. In my opinion, they will never replace traditional computer vision algorithms, but certainly, more and more vision systems will be based on Deep Learning technology.
2 komentarze

Post a Comment