A computational intelligence approach for earth resource satellite image processing, classification and decision-making
- Paper number
IAC-08.B1.4.8
- Author
Dr. Ernesto Araujo, Instituto Nacional de Pesquisas Espaciais (INPE), Brazil
- Coauthor
Mr. Marcelo Henrique Essado de Morais, Instituto Nacional de Pesquisas Espaciais (INPE), Brazil
- Year
2008
- Abstract
An intelligent system for processing images obtained by earth resource satellites is presented in this paper. This hybrid computational intelligent approach merges neural and fuzzy approaches in a neuro-fuzzy decision system for visual perception and pattern recognition. This intelligent hybrid system for visual-driven decision making employs, first, a neural system for image classification in charge of extracting information from image through a mapping between visual input and output datum. The resulting classification will, latter, be furnished as input data to the fuzzy decision support system for dealing with uncertainty and imprecision present in the available information. Advantages of both techniques are exploited in a complementary manner. The advantage of employing the fuzzy system for making decisions is related to its desirable attribute to cope with uncertainties. Thus, the hybrid system is able to deal with imprecision or noise in the information generated within image processing. This paper aims to demonstrate the technical viability of a computational intelligence model based on a hybrid neuro-fuzzy components when dealing with images sized 650 by 839 and 24 bits per pixel following the RGB color system. Moreover, earth resource satellites usually supply more than 200 thousands images and deals with more than 10 thousands types of fauna and flowers. Finding out a mechanism for exploring new techniques of image treatment in order to guarantee mutual low cost and high reliability and quality is necessary. The neuro-fuzzy system suggested in this paper attempts to insert new mechanisms in the geographic image processing and treatment in the space sector. Another reason for choosing this multidisciplinary approach was the opportunity of exploiting the integration of distinct areas since the system get together concepts and perceptions from the scientific field, public policies, geographic and cartographic information, only to mention few, in order to supply dissemination and processing for different applications and communication according to user necessities. Important mechanism for protect the environment, this system is an alternative to support specialists in the task of image treatment of this nature and magnitude helping them in dealing with the images skillfully and efficiently. As an example, this method was applied in automatic evaluation of images from the northeast Rio-Grandense plateau as well as a large part of the National Forest of Sao Francisco de Paula, considered the older preservation unit of the Rio Grande do Sul state, Brazil. The neural network in charge of classifying regions and classes previously defined presented appropriate consonance and correlation when compared to the original image and previous works. After the neural network analysis its result is furnished as input to a fuzzy system which, in turn, will decide based on a known class previously recognized and a set of rules. The main focus is, then, to characterize what sort of activity should be carried out in that region due to earth resource image classification. They range from exploration of forest products and sub-products to environmental education, preservation and maintenance of the world natural patrimony. The decision-making process is the critical part of the system since it requires specialized knowledge for satisfying the multiple requirements. It is worth mentioning that, such a process attempts to establish a strait relationship between the human vision potential and the inference capacity of a specialist. Based on the expertise obtained from specialists, this approach demonstrated the importance of their knowledge and perception in the decision-making model. In this application, results demonstrated that the proposed approach is useful in applications where images are used to supply information to specialists who will choose suitable actions. Results indicate that the proposed hybrid neural-fuzzy system achieves its goal. Moreover, it get together different perspectives for a common solution characterized by remote sensing, environment engineering and digital image computing that guarantee an efficient response for this area of study.
- Abstract document
- Manuscript document
IAC-08.B1.4.8.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.