Груша В. Information technology of processing data from chlorophyll-fluorometric sensors

The thesis is devoted to the challenging scientific problem to develop the information technology for processing data from chlorophyll-fluorometers based on chlorophyll fluorescence induction (CFI), that includes automation of decision making processes, data acquisition, transmitting and storing information concerning the state of plants, using web-technologies and a neural network approach. The analysis of the chlorophyll fluorometers market shows that manufacturers of chlorophyll fluorometers provide a variety of devices for different purposes such as the use in laboratory or field conditions and even for conducting measurements in water environment. As a result, there suggested the classification of modern chlorophyll fluorometers on the basis of grouping them by functional features, common design decisions and purposes. The analysis of wireless and web technologies shows that they enable developing scalable intelligent service-oriented systems for monitoring plants on the basis of chlorophyll fluorescence induction method. It allows for accumulating the results of monitoring plants from distributed areas, automating process of managerial decision making, predicting plants growth for a long period at the early stages of their growth. The analysis of scientific papers shows that modern researches take into account only parameters of CFI curves. Such an approach causes loses of some intermediate informative values that are important for managerial decisions. The series of experiments for studying chlorophyll fluorescence induction dependencies on different factors were conducted. Dependencies of CFI on environment parameters were studied. It was showed that temperature fluctuations above 5 degrees affect parameters of the CFI curve. Such dependencies of different parts of CFI curves were investigated and determined. The software developed for processing parameters of CFI curves allows for reducing time of data preparation for analysis, calculating main parameters of CFI and appropriate statistical indexes. The results of the calculation are useful for analysis of a plant state under influence of stress factors and under normal environmental condition. The database developed for storing results of researches is intended for the use in systems monitoring plant objects. The developed web-service is aimed at transmitting the measured results via internet by means of http protocol using JSON and XML formats. The research of time dependences of processing files with different amount of files determines that XML format is preferable for transmitting files which contain more than 2000 measurements. The investigation of data normalization methods aimed at improving recognition of plants sprayed by herbicide was conducted. The possibility to reduce dimension of the input vector of a neural network using as an example a plants classification task was investigated. The possibility to determin the water deficit for plants on the base of the measured chlorophyll fluorescence induction curve was studied. It was shown that neural networks based on the data of herbicide treatment experiment provide an early determining the influence of stress factors on the state of a plant before the appearance of the evidence of such influence on a plant. The criterion for automatic determining the sensitivity of CFI parameters to influential factors was proposed. It was suggested to use coefficient of determination of quadratic regression. It allows for automatic searching for such parameters. Efficiency of the criterion was demonstrated on experimental data. A method for data dimensionality reduction of digitalized chlorophyll fluorescence induction curves which were previously discretized using exponential scale was suggested. It consists of restoring lost data using polynomial piecewise approximation of curves and further using the principal component analysis method. A method of forming input data for neural network to determine the water deficit was proposed. It consists in normalizing digitalized values of CFI using z-score and further division by appropriate time ticks when initial values of CFI has been received. It allowed the possibility to improve the quality of neural network learning.