Industrial machine vision is not new – deep learning for machine vision is.
Machine vision is commonly defined as the use of computer vision in the context of an industrial application, and the first use of machine vision for industrial purposes is often attributed to Electric Sorting Machine Company in the 1930s. They used a type of vacuum tube called a photomultiplier or PMT to sort food. Using this technology, machines could sort red apples from green and later recyclable glass bottles from ones with cracks. Much of the history of machine vision in the industrial sector has involved sorting one thing from another, the good from the bad. As camera technologies have improved, machine vision has been deployed for ever more precise quality control use cases, especially ones that involve parts that would be too small or hazardous for human inspectors. Even so, machine vision was limited to low-mix environments where the machine could be programmed to inspect items for one parameter: blue vs. red, clean vs. dirty, round vs. square. It is the advent of artificial intelligence and more precisely deep learning that has moved industrial computer vision to the next level of value in the manufacturing setting.
DeepAI defines deep learning as “a machine learning technique that constructs artificial neural networks to mimic the structure and function of the human brain. In practice, deep learning, also known as deep structured learning or hierarchical learning, uses a large number of hidden layers -typically more than 6 but often much higher – of nonlinear processing to extract features from data and transform the data into different levels of abstraction (representations).” Using an example from my recent whitepaper, “2020 Executive Guide to Computer Vision in the Enterprise”, we can explain the difference between deep learning and traditional computer vision techniques with the following example of a cat. Traditional computer vision techniques begin with a top-down prescription of the components that constitute the image whereas deep learning-based vision systems use large data sets and countless training cycles to teach the machine, from the bottom-up, how a cat looks. During the training process, the algorithm automatically extracts the relevant features of ‘cats’. This produces a model that can be applied to previously unseen images to produce an accurate classification.
Large scale image sets brought together millions of images with accurately labeled features for deep learning algorithms. In a very short amount of time, the performance of deep learning algorithms surpassed thirty years of work on manual feature detectors. Feeding a sufficient number of well-labeled images to a deep learning-based visual system enables it to understand the exact pixel-level nuances that define the individual components of the larger image. It will automatically learn where the edges are, and how particular combinations of edges, which differ in color and contrast from each other and the background, combine to form certain features.
As the Internet of Things (IoT) continues to grow, the number of available images for training will increase and the opportunities to apply deep learning to machine vision will expand. The technique is applicable across many sectors and use cases. The use cases below are the three that we, at Dynam.AI, see as having the biggest near-term impact for the industrial sector.
When applied to industrial machine vision, deep learning can enable more human-like observations as well as the ability to learn new features. As an example, we can look at an automated cake-baking process. A traditional machine vision quality inspection may be able to judge whether a cake is the right shape or whether it is cooked to a golden brown or burnt. By applying deep learning, machines can identify multiple types of defects and even learn to identify and inspect new items (cookies, for example), quickly. The machine can now identify various imperfections in cakes, as an expert human baker would and then also recognize that instead of cakes, a batch of cookies has come down the line and apply a new set of parameters for cookie inspection. With the use of deep learning, the cost and time required to optimize quality inspection are greatly reduced.
Safety and Security
There are few issues more top-of-mind for managers of manufacturing facilities than employee safety. Plant Engineering lists poor maintenance, permanent hazards, undertrained employees, insufficient first aid, and carelessness as the top factors contributing to employee injuries. Two areas of deep learning that hold promise for reducing work-related injuries are Action and Activity Recognition and Human Pose Estimation. These techniques are being researched for use in both employee training and active safety monitoring. Additionally, robots used in both hospitals and the manufacturing environment need to be able to automatically and accurately identify and react appropriately to human positions and motion. The use of these technologies in robots allow them to operate autonomously in cooperation with human ‘co-workers’ and to warn of inappropriate human activities that may cause injury. [i] Similarly, advances in Object Detection using deep learning apply convolutional neural networks (CNNs)[ii] to recognize various objects that when left out of place or positioned incorrectly, could cause equipment damage or human injury. Finally, face, fingerprint, and retina recognition, while controversial in some settings, have been proven effective at making sure employees do not enter areas of an industrial facility for which they should not have access.
Item Classification and Logistics
The third area we see holding near-term promise for the industrial sector is logistics. Scanners have long been used to track stock and deliveries and optimize shelf space in stores. When deep learning is applied, a camera can not only read a bar code, but also detect if there is any type of label or code on the object, read it, and classify the object based on the information associated with that label. A recent example involves golf equipment industry leader, Titleist. Using deep learning, the Dynam.AI team was able to provide a solution for their professional club fitters that could identify the type of club and then identify the exact model of club head and shaft that a customer is using. This type of classification is both extremely precise and at the same time flexible enough to use across industries from construction[iii] to retail. In addition to managing inventory and predicting demand, deep learning is also being used to determine the most efficient means for raw materials and end-product delivery.
Want to learn more?
While this is not anywhere near an exhaustive list of the potential applications of deep learning as applied to machine vision, this does outline the areas of demand for deep learning that we, as a team of AI scientists and engineers, are beginning to see coming from our customer base. If you are interested in learning more about deep learning for machine vision, I recommend you access our whitepaper using the form below. And if you are already using industrial machine vision and feel you could benefit from the application of deep learning, I encourage you to reach out to us today to talk with our team about scheduling a ‘discovery’, or in-depth analysis of your data and use case, as the first step towards building a successful deep learning program.
Dynam.AI Computer Vision Solutions
Dynam.AI offers end-to-end AI solutions for companies looking to capitalize on the forces sweeping the modern business landscape. Our multidisciplinary team of AI experts, machine learning engineers, and data scientists have produced innovative solutions for the largest companies in the country. We have relevant experience spanning healthcare, financial services, infrastructure, and manufacturing. Our deepest expertise lies in deep learning for computer vision and the integration of innovative new AI technologies into traditional business operations.
Request a free consultation at Dynam.AI today to learn how computer vision and machine learning can transform your enterprise.
Dr. Michael Zeller
Dr. Michael Zeller has over 15 years of experience leading artificial intelligence and machine learning organizations through business expansion and technical success. Before joining Dynam.AI, Dr. Zeller led innovation in artificial intelligence for global software leader Software AG, where his vision was to help organizations deepen and accelerate insights from big data through the power of machine learning. Previously, he was CEO and co-founder of Zementis, a leading provider of software solutions for predictive analytics acquired by Software AG. Dr. Zeller is a member of the Executive Committee of ACM SIGKDD, the premier international organization for data science and also serves on the Board of Directors of Tech San Diego. He is an advisory board member at Analytics Ventures, Dynam.AI’s founding venture studio.