Articles and Reports

The introduction of integrated solutions optimizes the system and facilitates the implementation of deep learning systems AI

The implementation of comprehensive solutions reduces the complexity of system construction, making it easier for an industry to introduce AI deep learning systems. By investing time and effort in their familiar domain know-how, businesses can create more industrial value.

After being a topic of discussion and research for more than fifty years, Artificial Intelligence (AI) has finally stepped out from the laboratory following the development of neural network algorithms of deep learning techniques. Nowadays, AI is no longer a technology exclusive to R&D; in addition to being used in supercomputers that have beaten the world’s best chess players, driverless vehicles currently being road-tested, and robots that have been developed to have a sense of personality and identity, AI is also becoming widely accepted as an application tool with a high level of practicability. For example, deep learning technology is employed in a wide range of specialized applications such as quality screening for fruit and vegetables, retail traffic counting and analytics, medical imaging, and intelligent transportation.

A general concept of AI deep learning is to:

  1. Collect a mass amount of learning materials
  2. Acquire a knowledge model developed from an AI training system
  3. Introduce a model that has been designed for application in a specific field

These seemingly simple steps actually require a considerable amount of hardware and software expertise and relevant field expertise. As industries begin to utilize AI deep learning to improve their quality and efficiency, many relevant technologies will start to emerge in the market. Such a prosperous yet competitive business market is not only a strength but also a weakness for those seeking to implement AI deep learning. The strength for users lies in the variety of resources that developers may choose from, while major weaknesses include the time-consuming and painstaking task of integrating different types of hardware and software.

Simplifying the Development of AI Deep Learning Systems

While actively investing in AI deep learning, Advantech has found that aside from the arduous task of collecting learning materials, the most common problem faced by industries introducing AI deep learning is the tedious work involved in system construction. For example, what kind of hardware platform is needed in order to have sufficient functionality for complex computations? What type of hardware specifications are needed to meet the strict requirements of the application environment (e.g., public transportation, factories, clean rooms, and medical institutions)? How do AI systems connect to upper management software or cloud platforms? Are there any knowledge models available to shorten the overall training time? Is a new system to be developed or is an existing system to be upgraded to a deep learning system?

To solve the aforementioned problems, Advantech provides a complete deep learning solution with integrated hardware/software. The solution includes a training platform for developing deep learning models, a knowledge model for an inference platform to make real-time inferences, an SDK applicable to the development of deep learning systems, a ready-to-use knowledge model developed from the training process, and system planning and technical consulting services provided by professional teams. This simplifies system construction, making it easier for developers to build an AI deep learning system. This allows developers to concentrate on their domain know-how to generate applications that are more innovative and practical.

Precise Traffic Flow Monitoring and Efficient Law Enforcement

Advantech’s deep learning solutions have been successfully implemented in manufacturing, retail, transportation and many other industries. Applications in intelligent transportation, for example, include the statistical analysis of traffic flow, MRT passenger detection, car and license plate identification, parking space detection, enforcement of bus parking violations, large vehicle control and railway intrusion detection.

Among these, the solution for the statistical analysis of road traffic involves installation of the SKY-6100 server-grade training and inference platform at a traffic control center as well as MIC-7500 high-functionality inference platform on the roadside. These two platforms work together to identify vehicles in separate lanes by car type (e.g., bicycle, motorcycle, car, truck, bus, etc.), and this data is uploaded to a cloud platform. Additionally, the APIs in Advantech’s SDK also allows for seamless connectivity between data and system integrator applications for generating traffic management reports. The dashboard also provides a real-time display for information that is critical to smart transportation system control. In contrast to previous systems, where induction coils were installed on the roadside to detect and count the number of vehicles passing by within a given period, the introduction of deep learning system eliminated the need for induction coil installation while yielding statistical data that is more complete and precise.

In the case of enforcing bus stop parking violations, a compact MIC-7200 inference platform equipped with a built-in knowledge model is installed on site to receive captured images. When a vehicle is detected parking in the bus stop and found not to be a bus (as identified by the inference platform), an on-site digital display board and broadcaster will warn the owner of the illegally parked vehicle. Additionally, relevant data will be uploaded to the license plate identification system and the police station cloud platform in just over three minutes after an infringement occurs. This aids police with enforcing relevant traffic laws and regulations. Through such technological law enforcement tools, police stations with reduced human resources can remotely monitor bus stops and issue fines for parking infringements without needing to be on site; furthermore, violators will find it difficult to flee the scene because of traffic camera surveillance.

Lowering Technology Barriers to Promote Innovative Deep Learning

AI is a tool that has the potential to solve many problems affecting humans. Deep learning with self-training ability significantly improves the clarity of images, video, and text, making it a versatile tool for any application. However, system developers who might be competent at data collation and analysis in some fields might not necessarily understand what kind of computing environment is required for smooth deep learning.

Advantech has extensive experience in hardware and software integration within vertical industries. The company also owns its own production lines and manufacturers a wide range of products. This means that Advantech can provide complete solutions suitable for deep learning while also beings able to introduce valuable resources from many third-party partners, which simplifies system construction and lowers technical barriers. Ultimately, developers with completing their projects with the shortest possible time.

Advantech believes that through such a comprehensive resource integration service, smart and innovative applications extended from AI deep learning technology will soon start to thrive and prosper without boundaries.

Written by Xiao-Jing Yu. Interview with Advantech Smart System Business Group Director Hou-Yi Liu and Zhi-Wei

PreviousThe Next Three Years: A Critical Period for IoT Development
NextWhat is Time Sensitive Networking?