2D inspection in PEKAT VISION
Importance of automated visual inspection is growing, especially in the current times. Our software has proven to be a good choice for projects from various industrial areas. A lot of this versatility comes from several different modules, which can be used on their own or combined. Here are some of the projects we have solved, showcasing the different modules.
The “Anomaly of Surface” module is trained on good images and looks for any anomalies. One of the projects we solved with this module included finding defects on wheels.
Another module is called “Surface Detector”. It is trained on images with annotated areas and then tries to find surfaces which are similar to the annotated ones. The current version offers three types, each one based on a different neural network. In one of the projects solved with this module we were looking for scratches on cork sheets.
Apart from those two modules, PEKAT VISION also offers the “Detector” module which can be trained to look for objects similar to the ones annotated in training images (optionally with multiple classes). One of the projects solved with the Detector module included finding knots in wood and dividing them into two classes - ingrown knots (considered OK) and encased knots (these were considered as a defect, because they can fall out).
Another module called “Classifier” suits for classifying the whole image, objects found in previous modules (e.g. Detector) or objects on static positions in the image marked by rectangles, into different classes.
Other PEKAT VISION modules include OCR, measuring tool, image preprocessing or adding custom functionality through Python code. This portfolio of modules proved to be sufficient to solve practically any project.
There is also an “output” interface which can be used to trigger an action once an image from the camera is processed (can be set to trigger only for good or bad images). There you can use the command line (e.g. to run a script), send an HTTP request (GET or POST) or establish a connection using Profinet or TCP protocol to your PLC. You can also include the “context” from the application containing information like detected rectangles etc. The rectangles around detected defects can be used to localise the exact position of the defective product and e.g. automatically remove it afterwards.
Evaluation results and statistics of a project can be presented in an automatically generated report in HTML format, which makes it easier to assess how well the software works on a given project (especially on bigger datasets) or to share results with others.