Title: Real-time DetMCD-based classification of NIR-spectra
Authors: Iwein Vranckx - KU Leuven (Belgium) [presenting]
Mia Hubert - KU Leuven (Belgium)
Bart de Ketelaere - Katholieke Universiteit Leuven (Belgium)
Abstract: Modern, state-of-the-art sorting machines process several gigabytes of chemometric measurements in milliseconds, frequently pushing the boundaries of today's available computational power. This is further complicated by the fact that classic machine classifiers are typically not outlier-resistant. Given that they are practically omnipresent in industrial datasets, it is clear that industrial classifiers are often operating at sub-optimal efficiency.The goal is the equipping of food sorting machines with robust, DetMCD-based classifiers. In order for this to work, two major improvements are made: Firstly: DetMCD is pretty fast, but the current implementation still requires too much time for online applications. We propose several algorithmic modifications, among which techniques for parallel computing, such that our industrial time constraints are met. Secondly, the outlier detection capability of DetMCD is directly linked with industrial machine precision and should, therefore, be as high as possible. We improve this by using adaptive C-steps. To illustrate the performance of our improved DetMCD estimator, the algorithm is applied to industrial chemometric datasets of various food related products. We demonstrate the corresponding machine efficiency improvements and highlight the improved classification training times.