Information-extreme machine training of on-board recognition system with optimization of RGB-component digital images

Abstract

The research increases the recognition reliability of ground natural and infrastructural objects by use of an autonomous onboard unmanned aerial vehicle (UAV). An information-extreme machine learning method of an autonomous onboard recognition system with the optimization of RGB components of a digital image of ground objects is proposed. The method is developed within the framework of the functional approach to modeling cognitive processes of natural intelligence at the formation and acceptance of classification decisions. This approach, in contrast to the known methods of data mining, including neuro-like structures, provides the recognition system with the properties of adaptability to arbitrary initial conditions of image formation and flexibility in retraining the system. The idea of the proposed method is to maximize the information capacity of the recognition system in the machine learning process. As a criterion for optimizing machine learning parameters, a modified Kullback information measure was used, this informational criterion is the functionality of exact characteristics. As optimization parameters, the geometric parameters of hyperspherical containers of recognition classes and control tolerances for recognition signs were considered, which played the role of input data quantization levels when transforming the input Euclidean training matrix into a working binary training matrix using admissible transformations of a working training matrix the offered machine learning method allows to adapt the input mathematical description of recognition system to the maximum full probability of the correct classification decision acceptance. To increase the depth of information-extreme machine learning, optimization was conducted according to the information criterion of the weight coefficients of the RGB components of the brightness spectrum of ground object images. The results of physical modeling on the example the recognition of terrestrial natural and infrastructural objects confirm the increase in functional efficiency of information-extreme machine learning of on-board system at optimum in information understanding weight coefficients of RGB-components of terrestrial objects image brightness.