Feature Methods for Processing Remote Sensing Data (On the Example of Ili Alatau Park Natural)

Abstract

Monitoring of land of SPNTs (specially protected natural territories) is an indispensable and important task in the growing human impacts and environmental pollution and the need to develop systems of
specially protected natural territories on the lands of a large city and competent management increases. SPNT or Alatau reserve park monitoring is using natural resources and minimizing the impact of anthropogenic factors on the environment. This requires the monitoring of the composition of SPNTs land using effective aerospace methods. Improving the techniques of automatic decoding the satellite imagery and developing technology of aerospace monitoring SPNTs of Ili Alatau reserved Park are relevant research questions. In the paper the method of estimating the overall accuracy of automatic decoding of satellite imagery and its increase by an average of 10.5% was improved, substantiated and implemented. The dependence of the overall accuracy of the classification on the number of parts, which are divided into the original picture was determined. Dependence shows that classification accuracy increases according to the separation of the image, but if it is disaggregated more than 4 parts. Further division does not significantly improve the accuracy. In the article the solutions of the urgent problems of classification of agricultural land of SPNTs or Ili Alatau, improvement of methods of automatic decoding of satellite images be means of maximum likelihood method, which allows to increase the overall accuracy of the classification by an average of 10.5% are presented. A comparative analysis of methods of supervised classification and cluster analysis of satellite imagery is made on the basis of various algorithms for automatic decoding: the minimum distance, the parallelepiped, the maximum likelihood. It was established that in order to achieve the most accurate classification it is necessary to choose the maximum likelihood algorithm (T = 71.5%). Other methods give results with less classification accuracy.