Data Processing Module
Framework for processing units based on AI algorithms for integration and mining of spectral data.
Spectral Image Correction
Before using the images to train the respective models, they should be corrected for the instrument’s spectral response and the camera’s dark. On many occasions, this process is performed automatically from the data acquisition software. However, in case it is not, the following formula can be used:
Background Removal
Background removal is an essential pre-processing step commonly used in computer vision to improve the performance of machine learning and deep learning-based systems. This technique removes as much noise/ background as possible, making the object of interest the main part of the image. In agriculture, several works and techniques have been presented. Examples are the grabcut algorithm and thresholding of the background based on the HSV color space. Spectral imaging facilitates background removal as we can threshold the image based on specific wavelengths of interest as well as spectral indexes that match our use case.
Partial Least Squares regression (PLS)
Partial least squares regression (PLS regression) is one of the most common statistical methods used to analyze spectral data. It performs multiple linear regression to create a linear model that best describes the relations between two matrices (X and Y). The general model for PLS is:
If the variant/response is categorical, the Partial least squares discriminant analysis (PLS-DA) is used (shown below).
