We have presented perClass latest developments at the PhotonicsWest 2022 in San Francisco. It was a great opportunity to reconnect with our partners and customers. The visitors got a chance to see several live demos of perClass Mira and its runtime technology processing data from different spectral cameras. Happy to share more details with you in this post.
It was great to meet again (and finally in person) a number of customers, partners and colleagues in the spectral imaging field!
At perClass stand, we showed our latest development: Support for live acquisition and data processing for snapshot spectral sensors.
Namely, we demonstrated the Cubert Ultris 5 hyperspectral camera with 50 spectral bands in VNIR range.
We have shown live regression analysis estimating powder mixing proportions per object and per pixel:
We have also demoed foreign object detection in food. In both applications, perClass Mira output is a list of object coordinates, bounding boxes and centroids to be provided to an actuator.
In this screenshot, you can see detected objects of interest. The output windows provides their coordinates, centroids, bounding boxes and object decisions. The plot in upper right shows the acquisition speed (of the camera and the total system in green and of the perClass algorithm in red). At the bottom, you can see the spectral signatures for the trained classes in the range between 450 and 850 nm, This shows that Cubert Ultris camera provides finely samples spectrum, not only few bands.
Video of the live processing is available here:
For the first time, we also demoed the Imec SWIR snapshot camera. This unique technology provides 9-band image in the 900-1700nm range making it a perfect fit for plastic or textile recycling, food quality or moisture estimation projects. In our live demo, we showed classification of textile by material robust to color or appearance variations.
This video shows the live demo. Note, that the camera was acquiring roughly at 22 full cubes per second. perClass Mira was capable of processing about 125 full cubes per second on the laptop NVIDIA GPU.
Inno-spec has demonstrated live textile classification on an industrial line-scan RedEye2 system placed over a conveyor belt. perClass Mira provided live acquisition, reflectance correction and classification of different textile materials irrespective of their appearance.
At Headwall Photonics stand, visitors could see live foreign object detection demo. It was based on the MV.X hyperspectral line-scan camera processing the acquired data directly on-board using perClass Mira Runtime. The fact, that the MV.X provides integrated processing means, that the customers do not need a separate industrial computer.
The screen in the center of the photo shows the web interface to the MV.X unit - it is only for the result visualization, not for analysis.
The Headwall demo was illustrating detection of unknown foreign object types. This means that only the product modeling is performed. Everything different from the known product is detected as a foreign object.