Machine vision software
Deep learning recognizes anomalies
MVTec has released the latest version 19.11 of its standard software Halcon. The release is designed to simplify inspection tasks and speed up reading processes
For example, HALCON's anomaly detection enables deep-learning-based inspection tasks to be carried out more efficiently. This is because only a handful of images of defect-free objects are required to train the deep learning network. The process is then independently able to reliably localize deviations in other images, i.e. defects. For this defect detection, it is therefore no longer necessary to label training images of defective objects in advance.
Thanks to the new generic Box Finder, boxes and crates of any size can also be reliably recognized and localized within 3D point clouds and their dimensions determined. This means that a model no longer needs to be trained for each box size, which significantly increases the efficiency of many applications, especially in the logistics and pharmaceutical industries.
Read datacodes up to three times faster
Another innovation: ECC-200 datacodes are now read much faster in multicore systems. The reading processes are up to three times faster, which also increases usability on embedded systems. Data codes that are difficult to read can now also be recognized even more robustly.
Also new in Halcon 19.11 is the ability to import deep learning models in ONNX (Open Neural Network Exchange) format. This allows developers to use a wider range of deep learning networks within Halcon. Halcon has also been expanded to include a new model for line-scan cameras with telecentric lenses. as









