Assoc. Prof. Dr. Songyot Nakariyakul
We study the use of hyperspectral imaging for non-invasive food product inspection. We use machine learning to locate anomalies in various agricultural products such as detecting skin tumors and feces on chicken carcasses, classifying internally damaged almond nuts, and so on. In order to to make real-time detection possible, we propose selecting only a few important wavebands for classification. Our proposed product inspection system is fast and cost-effective.
Feature selection refers to algorithms that select a subset of d features from an initial larger set of D features, where a criterion function is used to measure the effectiveness of the selected feature set. Feature selection is important because it not only reduces the cost of collecting features but also improves the performance of a classifier. We propose new effective feature selection algorithms for pattern recognition applications. These include the adaptive branch and bound (ABB) algorithm, the improved forward floating selection (IFFS) algorithm, and so on.
We are currently developing computational methods to perform sequence analysis in proteins. For example, through the application of our proposed computational methods, we could predict the thermostability of a given protein from its primary sequence, which plays a crucial role in protein engineering and biotechnological research. More Recently, we attempt to predict interaction specificity of PDZ domains and their peptide ligands from their primary sequence.