Classification methods for remotely sensed data / Методы классификации данных дистанционного зондирования

Remotely sensed image data is widely used in a range of oceanographic, terrestrial, and atmospheric applications, such as land-cover mapping, environmental modeling and monitoring, and the updating of geographical databases. Hence, the quantized pixel values making up an image may be converted to physical values of radiance and related to some property of the surface being sensed. An example of this approach is the calibration of thermal infrared imagery to produce maps of temperature fields, such as sea surface temperature. In other applications, thematic information is required. A thematic map is one that displays the spatial variation of a specified phenomenon, such as land surface elevation, soil type, geology, or vegetation. It is this second approach that is considered in this book. The term pattern recognition is used to describe the procedures involved in relating vectors of measurements that are spatially referenced to individual pixel locations to the types or categories into which the phenomenon of interest is subdivided. If, for example, the phenomenon of interest is agricultural crops, then the categories are the individual crop types. Each crop type is represented in the thematic image by a numerical label. Vectors of measurements that are spatially referenced to individual pixel locations include image pixel values plus derived values such as texture, coherence, or context, as well as other geographical data that can be related to the pixel location, such as terrain elevation and slope, geology, and soil type.<...>