Advantech launches AIR Edge AI inference system

1 min read

Advantech has launched the AIR series of AI inference systems, addressing various AI applications including retail facial recognition, AOI/vision inspection, robotic and intelligent transportation.

Advantech is also providing an Edge AI Suite, a software toolkit with easy-to-use graphic user interface (GUI) and pre-trained AI models to help customers enable AI inference on edge devices.

The AIR series includes AIR-100, AIR-101 and AIR-200 which integrates Intel Movidius Myriad X VPU, and AIR-300 which supports a PCIe x16 high power graphic card.

Powered by Intel Atom x7-E3950 CPU and one Intel Movidius Myriad X VPU, AIR-100 is able to support multi-4K displays and is suitable for interactive kiosks and other retail applications with facial recognition functions.

AIR-101 supports DIN-Rail design and 12V~28V wide-range power input with Intel Atom E3940 Quad Core processors and two Intel Movidius Myriad X VPUs and is intended for AGV and factory automation applications.

Designed for higher computing power needs, the AIR-200 is an Intel Core i5 platform with two Intel Movidius Myriad X VPUs and supports up to 1080p video encoding, decoding and multi-channel processing and is suitable for traffic monitoring, license plate recognition and vehicle classification applications.

The powerful AIR-300 supports Intel Xeon E3/ 7th & 6th generation Core i series processor and a PCIe x16 high performance graphic card to enable fast inference and on-premises training for robotics and optical inspection applications.

The Edge AI Suite software toolkit integrates the Intel OpenVINO toolkit R3.1 to enable accelerated deep learning inference on edge devices and real-time monitoring of device status on the GUI dashboard. The latest Advantech Edge AI Suite v1.2 includes an optimised pre-trained Yolov3 object detection model popular in surveillance applications which delivers 2.8x better performance on VPU than running with OpenVINO only.

Customers can use Edge AI Suite v1.2 to execute the Yolov3 model with VPU+GPU heterogeneous acceleration to get 5x better AI inference performance than using a single device.