Partnering with an up-and-coming computer vision software company, Premio leverages its most advanced rugged computers at the edge to train neural networks in image detection
The Challenge
- Manual processes applied at a fish processing plant were inefficient and problematic
- The judgment of manual labor was highly subjective, leading to inconsistencies in quality assessments
- Staff was unable to maintain pace with the volume of fish entering the plant, causing a backlog and potentially spoiled goods
- Counting objects quickly by hand is primed for mistakes
- A powerful hardware solution was required to deploy the software algorithm to automate these processes
The Solution
- Premio's advanced, rugged edge computer provided the right combination of capabilities and cost
- The system's graphics engine enables deep learning and inference at the network edge
- Multifaceted visual data drives expert assessment of both quality and quantity
- Automating manual tasks would ensure consistency, speed and accuracy
The Benefit
- Taking a collaborative approach produced a right-fit solution extensible to other markets
- Neural network trains and retrains the system with new data for greater efficiency and non-stop improvement
- Robust system design leverages extended roadmaps of industrial components to assure long system life
- Environmental demands and variations can be easily accommodated with a rugged system built to withstand the elements
- Lower human involvement streamlines processes and ensures efficiency
- Staff reassigned to more rewarding tasks, tapping skills to promote company and professional growth as well as increased job satisfaction
The Company
The customer is a computer vision company based in Chile, developing artificial vision algorithms to enable highly precise object detection. The company's technology allows organizations to apply these algorithms to real-time processes and glean compelling data that significantly improves strategic decision making. The software is particularly well-suited to detailed quality assessment and highly-accurate quantity estimation. Empowering the deployment of artificial intelligence, inference analysis and computer vision systems, the company leads the charge in the automation of challenging tasks that have historically required a great deal of manpower, time, expertise and expense.
The Challenge
Premio’s customer's core strength is in designing the software algorithm for intelligent applications. The company was working with a food processing plant to revamp its quantity and quality control efforts. Leveraging inference, which is essentially creating new logic from old logic, the software could more efficiently count and evaluate the quality of fish fillets. These tasks had historically been performed manually, resulting in miscounts and quality control issues.
"Salmon is abundant here in Chile and is big business for the country. Our customer, a major fish distributor, was in search of an efficient alternative to the manual labor it had used to quantify and qualify salmon processing," said the project manager. "Appearance, including features such as size, color, clarity and other attributes, drives pricing decisions and distribution plans. But human labor could only do so much and was insufficient to effectively manage the incoming 100,000 fish per day. When the firm came to us, we knew a deep neural network powered by our software could be just the solution they were fishing for."
But, Premio’s customer knew that for its software to run at high speeds and deliver a high-performance neural network, it needed to partner with a hardware expert whose products were proven rugged and capable of running the machine learning algorithm in an industrial environment. To accommodate the quick pace and massive scale of the task at hand, the customer also saw tremendous value in advancing the software's real-time capabilities via edge computing – an alternative to the cloud, in which computing is performed at or near the data source. Edge computing would favorably impact latency, privacy and security, and bandwidth, helping the plant meet and potentially exceed its production goals.
The Solution
The customer approached several hardware vendors for a suitable solution but
discovered capabilities and costs varied widely. The company recognized it would have to dive deep to find a hardware partner that could design and deliver ideal industrial PC performance. No solution that offered the integrated GPU and graphics card necessary to harness the software's power was readily available. The customer ultimately connected with Premio, a global solutions provider specializing in ruggedized hardware that could be optimized for inference analysis and would operate efficiently and safely in severe conditions.
To meet the customer's ambitious goals and underlying criteria, Premio recommended its VCO-6020-1050TI, an advanced rugged edge computer with powerful supporting GPUs. This robust computer leverages a powerful graphics engine to enable deep learning and inference analysis at the network edge. It also supports real-time identification and decision making based on input data faster and more reliably than humans. Impressed with Premio's expertise in AI and machine learning, and the fact that its robust computers met all their unique performance and environmental requirements, the customer selected the hardware vendor and began collaborating. From the outset and throughout the engagement, Premio's responsive sales and support team demonstrated a strong commitment to listening to the unique IIoT computing challenges of the end user and devising the most appropriate path forward.
With a deep understanding of how object detection would impact the outcome, Premio designed the solution for optimal training of the neural network – where all the visual data continually feeds into the system. This is where comprehensive training and retraining occurs, refining and honing data to achieve the desired result. By continually adjusting the neural network to teach it to recognize the object, a fish, the system parses out its attributes. Once the most efficient algorithm is achieved, it is deployed on a computer dispatched to the rugged edge. With fish coming straight off the fishing vessel, it was imperative that the system tolerate shock, vibration, temperature variants and exposure to ocean water – making environmental performance a critical catalyst to deployment of a rugged edge computer.
The Benefit
Premio's VCO-6020-1050TI combines hardware with the customer's 
software algorithm to process complex visual tasks more efficiently than the human eye. These systems include programmed sensors, software algorithms, and both CPU and GPU computing power capable of analyzing images at high-speed intervals. Faster computational power in x86 processors and powerful real-time graphics accelerators allow for this industrial PC to provide results faster and more accurately than plant personnel. The results are significant improvements in productivity and efficiency gains.
Premio's industrial GPU computer with the customer’s software delivers AI performance in the shape of training neural networks to generate decision-making algorithms. It applies those algorithms to make fast, accurate decisions in challenging industrial environments. With personnel at the plant now able to make more informed decisions using real-time information, deviations in food quality can be identified more quickly and accurately than human monitoring alone, for greater safety and efficiency.
"I was impressed with how Premio forged professional relationships, not only with us but with the end client as well. Our Premio liaison was accessible and responsive, making communication a breeze," said the program manager. "With an excellent grasp on the client's goals and a deep understanding of how our algorithm works, Premio dove in with a robust industrial computer design that delivered results. Today our plant is operating more efficiently, and management is asking what more they can do with our technologies in the future."
With the right hardware in place, Premio’s customer is poised to expand its footprint and transform automation of industrial processes, not just within food processing and manufacturing but in any arena where quality evaluation and quantity computations impact the bottom line.
As more companies embrace machine learning to automate mundane but high-value tasks, they are looking for solutions – both hardware and software – to help them achieve their goals for more streamlined operations. Any business whose products or services require quality or quantity detection can benefit significantly from such technology. Harnessing the power of object detection software at the edge demands a robust industrial computer that brings those algorithms to life and positions the company for greater success.
