At New-Tech, Israel’s largest and most important high-tech and electronics trade fair, we will be represented with our own booth (19C) from 28 to 29 May 2019 among more than 150 participating companies. There we will present our hardware and software portfolio with a focus on deep learning and for high-speed applications such as machine vision, medical technology, PCB inspection and embedded vision with interfaces such as CoaXPress, Camera Link HS and Camera Link. Using a demonstrator, we illustrate a difficult inspection task with deep learning for detecting and classifying six different defects on reflective metallic surfaces. The solution is directly applicable for metalworking and automotive companies.
Five Days for Training and FPGA Implementation
The application implemented on a frame grabber FPGA with Camera Link interface achieves highest bandwidths of 220 MB/s with real-time output of the inspection results at detection rates of 99.4%. No additional CPU or ARM processors are required.
The software basis is a CNN operator within a software application design, which outputs a probability distribution of the defect classes for a acquired image. Using an FPGA programming environment such as VisualApplets, users can use the special CNN operator to integrate the trained net into an application design for its execution (inference).
The exhibition will feature several specialist events and conferences, including numerous lectures on the various focal points of the fair. These events enable visitors to the exhibition to find out about all the innovations and scientific developments in the machine vision industry under one roof.
Analysis result of the defect classification with probability distribution