perClass Mira


Pixel classification

This video shows first steps in a project defining a pixel classifier. Our example is from a food product sorting using NIR spectral camera. We separate product (two types of nuts) from known foreign objects also represented by several types of materials such as nut shells, stoned or plant leaves.

Object segmentation

In this video, we show how to segment objects out from the results of a pixel classifier. We focus on the use-case where each object is composed of a single material type (class). This leads to object detection solution. Removing small objects is demonstrated.

Object classification

In this video, we demonstrate classification of complex objects based on their content. We show how to grade objects using majority voting or custom rules. The example shows sorting of "French fries" removing pieces with rot or greening defects.

Object size distribution

In this video, we demonstrate how to estimate object size distribution and characterize object shape using a hyperspectral data set with different types of seeds.

Spectral index visualization

In this tutorial, we demonstrate how to define common types of spectral indices in interactive manner. We visualize an index related to chlorophyll content in potato plants affected by a virus infection. We limit the index visualization to plant leaves and stem and show how to obtain comparable results on multiple spectral cubes.

Creating custom spectral index

In this tutorial you learn how to interactively define your custom spectral index in perClass Mira. In our example, we create an index to characterize plant health from hyperspectral images. You will also see how to export results from multiple scans into Excel for further analysis.

Regression modeling

In this tutorial, we show how to estimate mixing proportion of powders using regression modeling. We show the full process from object definition, object annotation and regression modeling. The resulting solution can be applied to new scans both at object level and pixel level (distribution of regression output).

Regression modeling importing annotation from Excel

In this tutorial, we demonstrate step-by-step how to build a brix estimator for tomatoes. We show how annotations for multiple scans can be imported from Excel. The tutorial also illustrates how the per-object results can be batch-exported in Excel.

Live data processing

In this tutorial, we show how to apply the model to a live data stream from a Specim hyperspectral sensor. We show how the processing speed can be measured and how the classifier can be sped up by band reduction. In addition to classification, object sorting on the live stream is illustrated providing identical outputs to perClass Mira Runtime embedded in a custom application.

Live data acquisition using Unispectral Monarch II camera

This video shows perClass Mira 4.0 connected to Unispectral Monarch II multi-spectral camera. We walk through the process of acquiring spectral images, correcting data to reflectance using dark and white references, creating scans, building a classifier for almonds and shells, detecting and classifying objects. The solution is then applied to live data stream. The tutorial also illustrates, that only specific spectral bands can be used for the classifier. The Unispectral camera can be configured to acquire only these relevant bands which speeds up acquisition.

Live data processing using HAIP BlackIndustry camera

This tutorial shows perClass Mira performing live acquisition and processing of data from HAIP BlackIndustry camera. The video shows the full process from acquiring training data and references to model building and live object detection on the use-case of almonds and shells.

Plant Phenotyping

This tutorial demonstrates how to use perClass Mira for plant phenotyping application. We work with scans from CropSeeker in-field scanning platform on NPEC phenotyping facility in Wageningen, The Netherlands. In the demo, we show how to segment out flowers in hyperspectral scans and how to extract pigmentation information using batch processing.