At ANUGA FoodTech show in Cologne, Germany, we demonstrated live integration of perClass Mira with several spectral cameras. We have showed live tomato brix estimation, foreign object detection and also running the full GUI environment on NVIDIA Jetson
At ANUGA FoodTech show in Cologne, Germany, we presented perClass Mira software solution and integration with different spectral cameras to the vertical food technology market.
In this demo, we run live estimation of tomato brix using Imec SWIR camera capturing data in 900-1700nm region. perClass Mira was performing pixel and object classification followed by quality estimation step where the sugar content was calculated on object level. One more novel aspect of this demo is that we run the full process not from a usual PC but from NVIDIA Jetson Xavier AGX embedded computer. While the embedded processing based on perClass Mira Runtime technology has been available for more than two years, this demo shows the full perClass Mira GUI running the acquisition, processing and data visualization on the embedded system.
In the demo at the perClass stand the Cubert Ultris S5 hyperspectral video sensor was used to detect trained and untrained foreign objects in a food product. We used a complex product - a combination of different bean varieties, Various real foreign objects from food production were placed in the scene such as hand glove or head cover pieces, plastic machine fittings, wood or metal objects. While some common foreign object materials were trained, others were not - perClass Mira models were rejecting these as too different from trained concepts. We prefer to refer to this Cubert sensor as hyperspectral because it provides 50 channel spectrum for each image pixel. This rich information offers an additional opportunity for spectral data processing such as spectra filtering, derivatives and similar algorithms unavailable to multi-spectral sensors with few bands.
The foreign object demo at Cubert stand was detecting contamination on pizza. Due to long exposure to halogen lights, you could be guided to the Cubert stand by delicious pizza smell from far away!
We have also presented video of the cashew nut sorting machine built by Visratek. The sorting machine uses Headwall MV.X hyperspectral line-scan and perClass Mira to detect and remove foreign objects. Apart of standard FOs such as plastic, wood, metal or stone, it is capable of separating other types of nuts. This task would not be feasible using standard machine vision due to high similarity of color and shape. Spectral imaging, on the other hand, provides an informative data representation allowing for discrimination of objects by material. The sorting algorithms, trained in perClass Mira, can be run directly on the processing unit embedded in the MV.X camera. This greatly simplifies application deployment.
In another demo, we showed live foreign object detection using Ximea camera based on Imec Mosaic RedNIR sensor. This tiny 16-band camera can achieve very nice performance in separating nuts and shells. Please not, that the live stream visualization shows per-object decisions (not per-pixel decisions). This is a new functionality enabled in perClass Mira 4.0.
In this demo at Inno-spec stand, live detection of fungal infection on grains was performed using NIR-range (900-1700nm) RedEye camera.
