Substantiation of Methods and Development of an Algorithm for Neural Network Estimation of Object Classification Parameters in Aerial Imagery

The final decision on the nomenclature of the input parameters of the neural network is made as a result of rather lengthy comparative experiments. To reduce the number of these experiments, it is advisable to determine the importance of each of the possible parameters of computer vision. Since modern formalized methods for assessing the importance of the classification parameters of objects in aerial imagery do not meet the accuracy requirements, it was decided to use expert estimation to form the convolution parameters. It is proposed to use the method of paired comparisons, which is explained by its proven efficiency in cases of a large number of test objects, with which the parameters of object classification in aerial images are associated. In this case, the input data of the model is a vector, the elements of which will be matrices of expert assessments of the significance of computer vision parameters. As a result, a model of the processes of integration of object classification parameters on aerial imagery used to recognize unidentified objects and previously classified objects has been developed. The developed model is used to determine the parameters of computer vision, which can be used in neural network tools for optimizing scanners for processing airborne data based on the automation of an expert assessment of current parameters.