Hyperspectral imaging
Whereas RGB cameras work with three color channels (red, green and blue) to capture superficial color differences, hyperspectral technology offers much deeper analysis. This advanced technology captures tens to hundreds of narrow wavelength bands ranging from visible to infrared light.
Hyperspectral technology not only recognizes color, but also analyzes the chemical and structural properties of materials, something that is not possible with a standard RGB camera, making contamination detection more robust.
When inspecting potatoes, wavelengths between 700 and 1000 nm are essential due to their unique chemical composition (water content, starch and sugars). The Specim FX10 Hyperspectral camera, with a range of 400 to 1000 nm, is perfectly tailored to these requirements. This camera provides an affordable and efficient solution for detecting contaminants such as plastics, vegetation and other unwanted materials.
While the Specim FX17, with a range of 900 to 1700 nm, is suitable for more advanced applications, the FX10 is the ideal and cost-effective choice for standard contamination detection between potatoes.
Pre-processing and classification
Principal Component Analysis (PCA).
Each type of material has unique characteristics that become visible at specific wavelengths. Principal Component Analysis (PCA) is a powerful pre-processing technique on hyperspectral images that effectively analyzes these features. PCA highlights the most relevant spectral information, creating greater contrasts between materials, such as potatoes, vegetation and other contaminants making grouping easier.
Applying Principal Component Analysis (PCA) as a pre-processing step significantly improves the speed and robustness of contamination detection. PCA reduces the dataset to the most relevant features, making analyses faster and more efficient. This optimization allows the technology to be integrated in-line.
K-Nearest Neighbor algorithm (KNN).
By using PCA as a pre-processing step for contamination detection, the data is easier and more efficient to work with and lays the foundation for using smart algorithms such as K-Nearest Neighbor (KNN).
The KNN algorithm analyzes the pixels in the image and determines whether a pixel is part of a potato or a contaminant. This is done by comparing the chemical properties of a pixel with reference data. Pixels with the same properties are grouped together, making even small differences between potatoes and unwanted materials such as plastic vegetation and other contaminants easier.
Pre-processing and classification software developed by Vision Partners reliably identifies contaminants between potatoes in an industrial manner.
Vision Partners has demonstrated the industrial implementation of SWIR hyperspectral technology in a product line
