Intelligent Processing of Data From Chlorophyll Fluorometric Sensors

Abstract

Introduction. Chlorophyll fluorescence induction (CFI) is a monitoring method of plant objects. CFI is a radiation of chlorophyll in red spectrum during a chlorophyll lighting of alive plant in blue spectrum. Chlorophyll fluorometers – the special devices that are used for measurement of CFI. Series of such devices were developed in V.M. Glushkov Institute of Cybernetics of the NAS of Ukraine. In particular, fluorometer «Floratest» and a network of wireless sensors were developed for CFI measurement. An accumulation of massive amount of measurements resulted into possibility to use intellectual methods like neural networks. The purpose of the paper is to research the possibilities of machine learning methods (neural networks, support vector machine (SVM), XGBoost algorithm) for analysis of CFI curves that were measured by means of sensors developed in V.M. Glushkov Institute of Cybernetics of the NAS of Ukraine. Results. Neural networks, SVM, XGboost ensure early detection of influence of stress factors on state of plants before appearance of external symptoms on plants that was showed on basis of data received during experiments with treatment of plants by herbicide. Analogically there was showed the possibility of using the machine learning methods for determination of soil humidity. The better methods for given tasks were determined. The study of possibilities to enhance the results of mentioned methods by means of normalization was conducted. The best results were demonstrated by z-score normalization and by minimax normalization to the range [-1;1]. Conclusions. The application of different machine learning algorithm for processing CFI curves demonstrated that SVM and XGBoost better suit for task of classification plants treated by means of herbicide. Neural network demonstrated worst results. The application mentioned methods for task of determination of artificial watering necessity demonstrated that neural network shows better result, SVM shows worse result and XGBoost shows worst result. Keywords: Chlorophyll fluorescence induction, neural network, support vector machine, algorithm XGBoost.