Edge vision AI analytics

Background

Vision AI model training has matured, with medical image analysis in some cases slightly exceeding the judgment of general practitioners. YOLO has advanced to version 8 with substantial performance gains. However, limitations in deployment hardware and regulatory frameworks have prevented widespread practical adoption of vision AI inference.

Healthcare remains relatively conservative with long development cycles, but modular hardware and software approaches are emerging. New participants bring demand for both new technologies and reusable functional modules.

Edge AI use case: medical vision AI inference

Although AI models are largely mature, much training and inference still takes place in the cloud, without full edge deployment. Limited 5G availability and costs still make large-scale image and video uploads uneconomical. The industry therefore continues exploring edge computing and hybrid cloud-edge deployment for practical AI applications.

Long healthcare development cycles have not prevented several companies from planning edge AI inference early. Advantech’s extensive edge experience attracted a major Chinese medical equipment manufacturer to jointly develop a modular edge AI inference unit and advance its compliance process. The main challenges were hardware/software stability and deployment of a general-purpose AI environment. After repeated evaluation, the customer chose Advantech’s industrial-grade ROM-5722 computer-on-module as the main controller for an endoscopy system.

Solution

Quality hardware

Advantech ROM-5722 is a SMARC computer-on-module based on NXP i.MX8M Plus. Four Cortex-A53 cores at 1.8 GHz provide stable, efficient computing. It offers extensive interfaces, including 2 LAN ports, multiple USB, UART, I2C, GPIO, PCIe3.0, and CAN-FD. Operation across -40–85°C and a stable BSP customized by Advantech support long-term hardware/software reliability.

• Long-term hardware stability
• Complex peripheral expansion
• Long lifecycle supply

Hardware and software integration

Advantech’s low-level hardware/software engineering team averages over 10 years of industry experience across Yocto, Ubuntu, Debian, UOS, Kylin, Tongxin, Android, UNX, and other systems. Years of research produced Advantech Industrial Linux (AIL), also known as JelinaOS. Advantech then combined its cloud-edge and hardware/software integration strengths into Jelina, a comprehensive edge-to-cloud IoT management platform and edge computing software solution.
Jelina provides long-term system-level edge stability, secure remote global edge management software, and a comprehensive service framework for software and hardware platform upgrades. The team supports customer applications from UBOOT, EC/MCU and kernel optimization and trimming, filesystem customization and component adjustment, and application framework preparation through selected module development and full application migration. Advantech’s software team provides support spanning embedded development through edge-to-cloud integration.
Advantech’s general IoT experience, smart power plant solution design, and specialized energy storage projects have helped connect segments of the power industry. Numerous edge gateways extend China’s power network, helping collect more realistic operational data and fit more accurate models, laying the groundwork for automated dynamic power network adjustment.

Advantech Embedded+ software services

ROM-5722 specifications

• High-performance NXP Arm® Cortex®-A53 i.MX8M Plus Quad/Dual processor, up to 1.8 GHz
• 1 x Arm Cortex-M7 core
• Onboard 6 GB LPDDR4 memory, 4000 MT/s
• 1 x 4-lane MIPI-CSI, 1 x dual-channel LVDS or 1 x display interface
• 4 x USB2.0, 1 x USB 2.0 OTG, 4 x UART, 5 x I2C, 14 x GPIO, 1 x PCIe3.0
• 2 x Gigabit LAN, 2 x CAN-FD
• NPU AI performance of 2.3 TOPS
• Multiple operating systems, including Linux and Android

For more product and application information, call Advantech’s embedded solutions hotline at 400-001-9088.