|
ИСТИНА |
Войти в систему Регистрация |
Интеллектуальная Система Тематического Исследования НАукометрических данных |
||
COGNITIVE TECHNOLOGIES IN OPTICAL REMOTE SENSING DATA PROCESSINGстатья
- Авторы: Kozoderov V.V., Dmitriev E.V., Sokolov A.A.
- Журнал: Climate&Nature
- Номер: 1(2)
- Год издания: 2015
- Первая страница: 5
- Последняя страница: 45
- Аннотация: Optical remote sensing has become a valuable tool in pattern recognition and scene analysis techniques. This research and development discipline deals with an automation of decision making procedures to separate different classes of land surface objects on the hyperspectral images in visible and near infrared region. Cognitive technologies are designed to process these images using optimization procedures of spectral and textural features extraction for the related objects. Sensors are described for the imaging spectrometers based on the Charge Coupled Device (CCD) technology that enables to obtain hyperspectral cubes (two horizontal coordinates and wavelength). Computer vision procedures are outlined for optical remote sensing data processing. Some classifiers are considered for this purpose (artificial neural networks, linear and non-linear discriminant analysis, support vector machine method, Bayesian strategy in spatial and spectral optimization domains). The special emphasis is given to improvements of the machine-learning algorithms of pattern recognition for forests of different species and ages on the hyperspectral images. The applications are concerned the accuracy enhancement by selecting pixels relating to tree‟s sunlit tops, completely shaded background, partly illuminated by the Sun and partly shaded elements of the forest canopy for particular classes. The recognition errors of different classes of forests have shown to be similar to errors in the routine ground-based forest inventory techniques. This fact gives an opportunity to use the automation procedures for the assessment of the current forest ecosystem state while processing remote sensing hyperspectral images. Direct and inverse problems of atmospheric optics are solved to retrieve the biological productivity parameters for the recognized classes of forests based on the proposed techniques of hyperspectral imagery processing and modeling results of the projective cover and density of the forest canopy retrieval. Applying the cognitive technologies, we have detected object classes on a particular image, forest classes of different species and ages within their contours on the image, parameters of biological productivity for these forest classes.
- Добавил в систему: Козодеров Владимир Васильевич