perClass Mira 5.0 has introduced dual scanning mode: Simultaneous scanning of the same sample by two spectral cameras. 


The use-case for dual scanning is to simplify decision making which camera type or spectral range provides better interpretation results. For example, is the NIR range (900-1700nm) necessary to classifiy certain defect or is the VNIR range (400-1000nm) sufficient? Without dual scanning, one would have to scan the samples twice. This is both time consuming and error prone. With dual scanning, we only scan our samples once but end up with two data sets, one for the NIR and one for the VNIR camera. We carry on the data analsys and model building separately for each data set. Comparing results, obtained on these two data sets is easy as the same physical object is represented by the two scans with the same names. In this way, we may quickly and reliably quantify benefits of each camera type.


Conceptually, an instance of perClass Mira only controls one camera. We call this first instance a master. We may spawn another instance, called a worker. The worker instance is connected to another camera. Both, the master and the worker, can be configured as needed in the usual way. An example dual scanning system is shown in the following photo. The master instance (on the left screen), is connected to the MV.C NIR camera and to perClass Mira Stage. The worker instance, on the right screen, is connected to the tiny MV.C VNIR camera attached to the side of MV.C NIR.



During the scanning session, we only operate the master instance. Actions, we invoke in the master, are propagated to the worker. For example, when we start scanning in the master, the stage starts moving and the MV.C NIR camera initiates the acquisition. Simultaneously, the worker instance is notified and starts acquiring data from the MV.C VNIR camera, as well. When we define a scan name in the master's recording panel, it gets copied over the the worker. The master and the worker store data to different directories. Each physical object is, therefore, represented by a scan of the same name in these two directory locations.


Note, that dual scanning does not provide co-registered data. Although each of the two data sets views the same objects but these views are not pixel aligned. As described above, the primary goal of dual scanning is data acqusiition for model selection and training, not for production deployments with multiple sensors.


Next, we walk through the full example of using dual scanning.