Remote Sensing and GIS Application for Earth Observation on the base of the Neural Networks in Aerospace Image Classification

Up-to-date curetted task of the present geoinformation system is the processing of the remote sensing data. Image analyses in point of mathematical view bases on the theory of the image recognition where it is necessary identification of the input data to the appropriate classis of the objects. Beside with methods of fuzzy logic one of the advance methods of solution the foregoing problem is wide application of Neural Networks in the area. These methods have been taken an adequate instrument which can be described regulation of classification without application of the high accuracy mathematical value as it accepted term for understanding like "small", "significant" etc. There are two approaches of image recognition based on the spectral and spatial characteristics of the investigated class of objects. It is expedient the combination of the both approaches in order to increase an accuracy of recognition and classification of scanned area with complex relief (highlands, settlements, mixed forest, etc.). Each selected individual elements of landscape shall be found out based on the image fragment analyses, their forms, colors, correlation, and heterogeneity. Neural Network allows assessing interrelation of each selected individual elements of landscape. One of the advantages of the Neural Network is that all elements have an ability to operate in parallel that is essentially increasing the efficiency of problem solution, especially in the area of the image processing. Within the framework of this studies have been offered the approach of development of the Artificial Neural Networks with Back-propagation Error where there is a solution of the problem of creation decoding indicators based on superposition of spectral and spatial characteristics of image of Landsat TM satellite. During the study of classification of the object elements considered learning access is known as the indicator of identification of relation of the classes. It is required development of the Neural Network which has to be determining the relation of appropriate classes to the object. The number of parameters on spatial characteristic increase due to involving parameters calculated by horizontal, vertical and diagonal pixels taking part in process. As a result it can be find out an increase of a volume of computation process. The main aim of those research activities is development of methodology of high-performance classification of aerospace images using the Neural Networks for modeling of spectral and spatial images of the investigated class of objects. Integration in GIS Neural Networks is the suitable instrument for effective encouragement of decision-making. As an input and output data the Neural Network uses of GIS spatial data.