LINX performs deep learning applications with MVTec HALCON

LINX is the premier distributor for machine vision and automation products in the Japanese market. As part of the LINXDays in Japan, they built a demo setup to showcase HALCON’s deep-learning-based object detection. You can download their example program here.

Components used:

• OS: Windows 10, 64bit
• HALCON version: 18.11
• CPU: Intel Core i7-8700
• GPU: NVIDIA GTX1080Ti

Training:

• Training Data: About 100 objects/images for each class
• Training Time: About 3 hours
• Inference Time (GPU): About 25 ms for each image

Application:

The application detects and recognizes electric components (capacitors, transistors and integrated circuits), a useful preprocessing step for, e.g., automatic optical inspection (AOI) and surface-mount technology (SMT). It is difficult to detect the location of these components by using traditional, rule-based pattern matching technology, due to several types of shapes per category, as well as differing appearances depending on the position on the image. Consequently, using a deep-learning-based detection approach proves to be an efficient solution for this use case.