AI kit is believed to be first to use sparse modelling
Embedded computer company, congatec has partnered with Japanese articifical intelligence (AI) company, Hacarus, to offer what they claim is the the world’s first embedded computing kit for AI that uses sparse modelling technology.
Sparse modelling needs little training data to make highly accurate predictions. For vision-based inspection systems in particular, this means the reject rate is naturally lower when manufacturing quality is high. With sparse modelling it is possible to create a new inspection model starting with 50 or even fewer images. This is significantly less than the 1,000 or more images required for traditional AI, reports congatec. The Sparse Modeling Kit, available from Hacarus, can be used as an add-on to existing inspection systems or as a standalone kit.
Although primary customers are vision system providers and system integrators, it can also be used by machine and system builders who want to use vision-based AI but have been reluctant to do because the wide variety of individual customer installations requires algorithms to be adapted, which adds cost.
“With sparse modelling, developers are able to build next-generation inspection systems that can be trained for individual requirements and can therefore function anywhere,” said Christian Eder, director of marketing at congatec. “There is no longer a need for optimal conditions, such as constant lighting,” he continued. OEMs also gain greater flexibility to adapt to changing production processes, which is essential for the move to industrial IoT/Industry 4.0 controlled batch size production, he added
“Humans can recognise friends and family on the basis of key characteristics – such as eyes or ears. Sparse modelling integrates a comparable logic into intelligent image processing systems,” said Takashi Someda, CTO of Hacarus.
This approach means the entire volume of big data does not have to be processed, as is the case with conventional AI. Algorithms based on sparse modelling reduce the data to unique characteristics, for a smaller AI footprint, which is suitable for fanless low-power systems that are in continuous 24/7 use.
The starter kit based on congatec hardware and Hacarus software can instantly be deployed and tested in any GigE and USB 3.x environment. Designed on the basis of palm sized computer-on-modules (CoMs), the system measures only 173 x 88 x 21.7mm (6.81 x 3.46 x 0.85-inches). It includes the latest Intel Atom and Celeron processors (Codename Apollo Lake) that are all available for series production today. There are two GbE application-ready standard interfaces for GigE Vison, one USB3.0/2.0, four USB2.0 and one UART (RS-232) port. Extensions are possible with two Mini-PCIe with USIM socket, one mSATA socket and 16-bit programmable general purpose I/O. The wide range DC voltage input is 9.0 to 32V.