3_Classification - Classification Classification...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Classification Classification Multispectral classification may be performed using aavariety of methods, Multispectral classification may be performed using variety of methods, including: including: algorithms based on parametric and nonparametric statistics that use algorithms based on parametric and nonparametric statistics that use ratio- and interval-scaled data and nonmetric methods that can also ratio- and interval-scaled data and nonmetric methods that can also incorporate nominal scale data; incorporate nominal scale data; the use of supervised or unsupervised classification logic;, the use of supervised or unsupervised classification logic;, the use of hard or soft ((fuzzy)set classification logic to create hard or the use of hard or soft fuzzy) set classification logic to create hard or fuzzy thematic output products; fuzzy thematic output products; the use of per-pixel or object-oriented classification logic, and the use of per-pixel or object-oriented classification logic, and hybrid approaches. hybrid approaches. Classification Classification Parametric methods such as maximum likelihood classification and Parametric methods such as maximum likelihood classification and unsupervised clustering assume normally distributed remote sensor data and unsupervised clustering assume normally distributed remote sensor data and knowledge about the forms of the underlying class density functions. knowledge about the forms of the underlying class density functions. Nonparametric methods such as nearest-neighbor classifiers, fuzzy Nonparametric methods such as nearest-neighbor classifiers, fuzzy classifiers, and neural networks may be applied to remote sensor data that classifiers, and neural networks may be applied to remote sensor data that are not normally distributed and without the assumption that the forms of are not normally distributed and without the assumption that the forms of the underlying densities are known. the underlying densities are known. Nonmetric methods such as rule-based decision tree classifiers can operate Nonmetric methods such as rule-based decision tree classifiers can operate on both real-valued data (e.g., reflectance values from 00to 100%) and on both real-valued data (e.g., reflectance values from to 100%) and nominal scaled data (e.g., class 11= forest; class 22= agriculture). nominal scaled data (e.g., class = forest; class = agriculture). Supervised Classification Supervised Classification In aasupervised classification, the identity and location of some of the landIn supervised classification, the identity and location of some of the landcover types (e.g., urban, agriculture, or wetland) are known aappriorithrough cover types (e.g., urban, agriculture, or wetland) are known riori through aa combination of fieldwork, interpretation of aerial photography, map combination of fieldwork, interpretation of aerial photography, map analysis, and personal experience. The analyst attempts to locate specific analysis, and personal experience. The analyst attempts to locate specific sites in the remotely sensed data that represent homogeneous examples of sites in the remotely sensed data that represent homogeneous examples of these known land-cover types. These areas are commonly referred to as these known land-cover types. These areas are commonly referred to as training sites because the spectral characteristics of these known areas are training sites because the spectral characteristics of these known areas are used to train the classification algorithm for eventual land-cover mapping of used to train the classification algorithm for eventual land-cover mapping of the remainder of the image. Multivariate statistical parameters (means, the remainder of the image. Multivariate statistical parameters (means, standard deviations, covariance matrices, correlation matrices, etc.) are standard deviations, covariance matrices, correlation matrices, etc.) are calculated for each training site. Every pixel both within and ooutside the calculated for each training site. Every pixel both within and utside the training sites is then evaluated and assigned to the class of which it has the training sites is then evaluated and assigned to the class of which it has the highest likelihood of being aamember. highest likelihood of being member. Unsupervised Classification Unsupervised Classification In an unsupervised classification, the identities of land-cover In an unsupervised classification, the identities of land-cover types to be specified as classes within aascene are not generally types to be specified as classes within scene are not generally known a priori because ground reference information is known a priori because ground reference information is lacking or surface features within the scene are not well lacking or surface features within the scene are not well defined. The computer is required to group pixels with similar defined. The computer is required to group pixels with similar spectral characteristics into unique clusters according to some spectral characteristics into unique clusters according to some statistically determined criteria. The analyst then re-labels and statistically determined criteria. The analyst then re-labels and combines the spectral clusters into information classes. combines the spectral clusters into information classes. Hard vs. Fuzzy Classification Hard vs. Fuzzy Classification Supervised and unsupervised classification algorithms typically use hard Supervised and unsupervised classification algorithms typically use hard classification logic to produce aa classification map that consists of hard, classification logic to produce classification map that consists of hard, discrete categories (e.g., forest, agriculture). discrete categories (e.g., forest, agriculture). Conversely, it is also possible to use fuzzy set classification logic, which Conversely, it is also possible to use fuzzy set classification logic, which takes into account the heterogeneous and imprecise nature of the real world. takes into account the heterogeneous and imprecise nature of the real world. Per-pixel vs. Object-oriented Classification Per-pixel vs. Object-oriented Classification In the past, most digital image classification was based on processing the In the past, most digital image classification was based on processing the entire scene pixel by pixel. This is commonly referred to as per-pixel entire scene pixel by pixel. This is commonly referred to as per-pixel classification. classification. Object-oriented classification techniques allow the analyst to decompose Object-oriented classification techniques allow the analyst to decompose the scene into many relatively homogenous image objects (referred to as the scene into many relatively homogenous image objects (referred to as patches or segments) using aamulti-resolution image segmentation process. patches or segments) using multi-resolution image segmentation process. The various statistical characteristics of these homogeneous image objects The various statistical characteristics of these homogeneous image objects in the scene are then subjected to traditional statistical or fuzzy logic in the scene are then subjected to traditional statistical or fuzzy logic classification. Object-oriented classification based on image segmentation classification. Object-oriented classification based on image segmentation is often used for the analysis of high-spatial-resolution imagery (e.g., 1 11 is often used for the analysis of high-spatial-resolution imagery (e.g., 1 m Space Imaging IKONOS and 0.61 0.61 m Digital Globe QuickBird). m Space Imaging IKONOS and 0.61 0.61 m Digital Globe QuickBird). Be Careful Be Careful No pattern classification method is inherently superior to any No pattern classification method is inherently superior to any other. The nature of the classification problem, the biophysical other. The nature of the classification problem, the biophysical characteristics of the study area, the distribution of the characteristics of the study area, the distribution of the remotely sensed data (e.g., normally distributed), and a priori remotely sensed data (e.g., normally distributed), and a priori knowledge determine which classification algorithm will yield knowledge determine which classification algorithm will yield useful results. Duda et al. (2001) provide sound advice: "We useful results. Duda et al. (2001) provide sound advice: "We should have a healthy skepticism regarding studies that should have a healthy skepticism regarding studies that purport to demonstrate the overall superiority of a particular purport to demonstrate the overall superiority of a particular learning or recognition algorithm." learning or recognition algorithm." Jensen, 2005 Jensen, 2005 Land-use and Land-cover Classification Schemes Land-use and Land-cover Classification Schemes Land cover refers to the type of material present on the landscape (e.g., Land cover refers to the type of material present on the landscape (e.g., water, sand, crops, forest, wetland, human-made materials such as asphalt). water, sand, crops, forest, wetland, human-made materials such as asphalt). Land use refers to what people do on the land surface (e.g., agriculture, Land use refers to what people do on the land surface (e.g., agriculture, commerce, settlement). commerce, settlement). The pace, magnitude, and scale of human alterations of the Earth's land The pace, magnitude, and scale of human alterations of the Earth's land surface are unprecedented in human history. Therefore, land-cover and surface are unprecedented in human history. Therefore, land-cover and land-use data are central to such United Nations' Agenda 21 issues as land-use data are central to such United Nations' Agenda 21 issues as combating deforestation, managing sustainable settlement growth, and combating deforestation, managing sustainable settlement growth, and protecting the quality and supply of water resources. protecting the quality and supply of water resources. Land-use and Land-cover Classification Schemes Land-use and Land-cover Classification Schemes All classes of interest must be selected and defined carefully All classes of interest must be selected and defined carefully to classify remotely sensed data successfully into land-use to classify remotely sensed data successfully into land-use and/or land-cover information. This requires the use of a and/or land-cover information. This requires the use of a classification scheme containing taxonomically correct classification scheme containing taxonomically correct definitions of classes of information that are organized definitions of classes of information that are organized according to logical criteria. If aahard classification is to be according to logical criteria. If hard classification is to be performed, then the classes in the classification system should performed, then the classes in the classification system should normally be: normally be: mutually exclusive, mutually exclusive, exhaustive, and exhaustive, and hierarchical. hierarchical. Land-use and Land-cover Classification Schemes Land-use and Land-cover Classification Schemes * Mutually exclusive means that there is no taxonomic overlap * Mutually exclusive means that there is no taxonomic overlap (or fuzziness) of any classes (i.e., deciduous forest and (or fuzziness) of any classes (i.e., deciduous forest and evergreen forest are distinct classes). evergreen forest are distinct classes). * Exhaustive means that all land-cover classes present in the * Exhaustive means that all land-cover classes present in the landscape are accounted for and none have been omitted. landscape are accounted for and none have been omitted. * Hierarchical means that sublevel classes (e.g., single-family * Hierarchical means that sublevel classes (e.g., single-family residential, multiple-family residential) may be hierarchically residential, multiple-family residential) may be hierarchically combined into aa higher- level category (e.g., residential) that combined into higher- level category (e.g., residential) that makes sense. This allows simplified thematic maps to be makes sense. This allows simplified thematic maps to be produced when required. produced when required. Land-use and Land-cover Classification Schemes Land-use and Land-cover Classification Schemes It is also important for the analyst to realize that there is aa It is also important for the analyst to realize that there is fundamental difference between information classes and fundamental difference between information classes and spectral classes. spectral classes. * Information classes are those that human beings define. * Information classes are those that human beings define. * Spectral classes are those that are inherent in the remote * Spectral classes are those that are inherent in the remote sensor data and must be identified and then labeled by the sensor data and must be identified and then labeled by the analyst. analyst. Land-use and Land-cover Classification Schemes Land-use and Land-cover Classification Schemes Certain hard classification schemes can readily incorporate land-use and/or landCertain hard classification schemes can readily incorporate land-use and/or landcover data obtained by interpreting remotely sensed data, including the: cover data obtained by interpreting remotely sensed data, including the: American Planning Association Land-Based Classification System which is American Planning Association Land-Based Classification System which is oriented toward detailed land-use classification; oriented toward detailed land-use classification; United States Geological Survey Land-Use/Land-Cover Classification System for United States Geological Survey Land-Use/Land-Cover Classification System for Use with Remote Sensor Data and its adaptation for the U.S. National Land Cover Use with Remote Sensor Data and its adaptation for the U.S. National Land Cover Dataset and the NOAA Coastal Change Analysis Program (C-CAP); Dataset and the NOAA Coastal Change Analysis Program (C-CAP); U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands and Deepwater Habitats of the United States; and Deepwater Habitats of the United States; U.S. National Vegetation and Classification System; U.S. National Vegetation and Classification System; International Geosphere-Biosphere Program IGBP Land Cover Classification International Geosphere-Biosphere Program IGBP Land Cover Classification System modified for the creation of MODIS land-cover products System modified for the creation of MODIS land-cover products The Land-Based The Land-Based Classification System Classification System (LBCS) contains detailed (LBCS) contains detailed definitions of urban/ definitions of urban/ suburban land use. The suburban land use. The system incorporates system incorporates information derived in situ information derived in situ and using remote sensing and using remote sensing techniques. This is an techniques. This is an oblique aerial photograph of oblique aerial photograph of aamall in Ontario, CA. mall in Ontario, CA. Hypothetical activity and Hypothetical activity and structure codes associated structure codes associated with this large parcel are with this large parcel are identified. Site development identified. Site development and ownership information and ownership information attribute tables are not attribute tables are not shown (courtesy American shown (courtesy American Planning Association). Planning Association). U.S. Geological Survey's Land-Use/Land-Cover U.S. Geological Survey's Land-Use/Land-Cover Classification System for Use with Remote Sensor Data Classification System for Use with Remote Sensor Data The U.S. Geological Survey's Land-Use/Land-Cover Classification The U.S. Geological Survey's Land-Use/Land-Cover Classification System for Use with Remote Sensor Data is aaresource-oriented landSystem for Use with Remote Sensor Data is resource-oriented landcover classification system in contrast with people or activity landcover classification system in contrast with people or activity landuse classification systems such as the APA's Land-Based use classification systems such as the APA's Land-Based Classification System. The USGS rationale is that "although there is Classification System. The USGS rationale is that "although there is an obvious need for an urban-oriented land-use classification an obvious need for an urban-oriented land-use classification system, there is also aa need for aa resource-oriented classification system, there is also need for resource-oriented classification system whose primary emphasis would be the remaining 95 percent system whose primary emphasis would be the remaining 95 percent of the United States land area." The USGS system addresses this of the United States land area." The USGS system addresses this need with 88of the 99Level IIcategories that treat land area that is not need with of the Level categories that treat land area that is not in urban or built-up categories. The system is designed to be driven in urban or built-up categories. The system is designed to be driven primarily by the interpretation of remote sensor data obtained aat primarily by the interpretation of remote sensor data obtained t various scales and resolutions and not data collected in situ. various scales and resolutions and not data collected in situ. Four Levels of the U.S. Geological Survey Four Levels of the U.S. Geological Survey Land-Use/Land-Cover Classification Land-Use/Land-Cover Classification System for Use with Remote Sensor Data System for Use with Remote Sensor Data and the type of remotely sensed data and the type of remotely sensed data typically used to provide the information. typically used to provide the information. Four Levels of the U.S. Geological Survey Four Levels of the U.S. Geological Survey Land-Use/Land-Cover Classification Land-Use/Land-Cover Classification System for Use with Remote Sensor Data System for Use with Remote Sensor Data and the type of remotely sensed data and the type of remotely sensed data typically used to provide the information. typically used to provide the information. U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands and Deepwater Habitats of the United States and Deepwater Habitats of the United States U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands and U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands and Deepwater Habitats of the United States Deepwater Habitats of the United States The U.S. Department of the Interior Fish & Wildlife Service is rresponsiblefor mapping The U.S. Department of the Interior Fish & Wildlife Service is esponsible for mapping and inventorying wetland in the United States. Therefore, they ddevelopedaawetland and inventorying wetland in the United States. Therefore, they eveloped wetland classification system that incorporates information extracted from remote sensor data classification system that incorporates information extracted from remote sensor data and in situ measurement (Cowardin et al., 1979). and in situ measurement (Cowardin et al., 1979). The Cowardin system describes ecological taxa, arranges them in aasystem useful to The Cowardin system describes ecological taxa, arranges them in system useful to resource managers, and provides uniformity of concepts and terms. Wetlands are resource managers, and provides uniformity of concepts and terms. Wetlands are classified based on plant characteristics, soils, and frequency of flooding. Ecologically classified based on plant characteristics, soils, and frequency of flooding. Ecologically related areas of deep water, traditionally not considered wetlands, are included in the related areas of deep water, traditionally not considered wetlands, are included in the classification as deep-water habitats. Five systems form the highest level of the classification as deep-water habitats. Five systems form the highest level of the classification hierarchy: marine, estuarine, riverine, lacustrine, and palustrine. Marine classification hierarchy: marine, estuarine, riverine, lacustrine, and palustrine. Marine and estuarine systems each have two subsystems: subtidal and intertidal. The riverine and estuarine systems each have two subsystems: subtidal and intertidal. The riverine system has four subsystems: tidal, lower perennial, upper perennial, and intermittent. system has four subsystems: tidal, lower perennial, upper perennial, and intermittent. The lacustrine has two, littoral and limnetic, and the palustrine has no subsystem. The lacustrine has two, littoral and limnetic, and the palustrine has no subsystem. Within the subsystems, classes are based on substrate material aandflooding regime or Within the subsystems, classes are based on substrate material nd flooding regime or on vegetative life form. The same classes may appear under one oormore of the systems on vegetative life form. The same classes may appear under one r more of the systems or subsystems. Jensen, 2005 or subsystems. Jensen, 2005 U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands and U.S. Department of the Interior Fish & Wildlife Service Classification of Wetlands and Deepwater Habitats of the United States Deepwater Habitats of the United States The Cowardin system was adopted as the National Vegetation Classification Standard The Cowardin system was adopted as the National Vegetation Classification Standard for wetlands mapping and inventory by the Wetlands Subcommittee of the Federal for wetlands mapping and inventory by the Wetlands Subcommittee of the Federal Geographic Data Committee (FGDC, 1996). The Cowardin wetland classification Geographic Data Committee (FGDC, 1996). The Cowardin wetland classification system is the most practical scheme to use if you are going to eextractwetland system is the most practical scheme to use if you are going to xtract wetland information from remotely sensed data and share the information with others interested information from remotely sensed data and share the information with others interested in wetland-related problems. in wetland-related problems. U.S. National Vegetation Classification System U.S. National Vegetation Classification System The Vegetation Subcommittee of the Federal Geographic The Vegetation Subcommittee of the Federal Geographic Data Committee has endorsed the National Vegetation Data Committee has endorsed the National Vegetation Classification System (NVCS) which produces uniform Classification System (NVCS) which produces uniform vegetation resource data at the national level. The NVCS vegetation resource data at the national level. The NVCS uses aasystematic approach to classifying aacontinuum of uses systematic approach to classifying continuum of natural, existing vegetation. The combined physiognomicnatural, existing vegetation. The combined physiognomicfloristic hierarchy uses both qualitative and quantitative floristic hierarchy uses both qualitative and quantitative data appropriate for conservation and mapping at various data appropriate for conservation and mapping at various scales. Physiognomic characteristics include the more scales. Physiognomic characteristics include the more general and less precise levels of taxonomy, whereas the general and less precise levels of taxonomy, whereas the floristic characteristics are found in the more specific floristic characteristics are found in the more specific levels of taxonomy. levels of taxonomy. U.S. National Vegetation Classification System U.S. National Vegetation Classification System The Vegetation Subcommittee of the Federal The Vegetation Subcommittee of the Federal Geographic Data Committee has endorsed Geographic Data Committee has endorsed the National Vegetation Classification the National Vegetation Classification System (NVCS) which produces uniform System (NVCS) which produces uniform vegetation resource data at the national level. vegetation resource data at the national level. The NVCS uses aasystematic approach to The NVCS uses systematic approach to classifying aacontinuum of natural, existing classifying continuum of natural, existing vegetation. The combined physiognomicvegetation. The combined physiognomicfloristic hierarchy uses both qualitative and floristic hierarchy uses both qualitative and quantitative data appropriate for quantitative data appropriate for conservation and mapping at various scales. conservation and mapping at various scales. Physiognomic characteristics include the Physiognomic characteristics include the more general and less precise levels of more general and less precise levels of taxonomy, whereas the floristic taxonomy, whereas the floristic characteristics are found in the more specific characteristics are found in the more specific levels of taxonomy. levels of taxonomy. International Geosphere-Biosphere Program IGBP Land-Cover Classification International Geosphere-Biosphere Program IGBP Land-Cover Classification System Modified for the Creation of MODIS Land-Cover Products System Modified for the Creation of MODIS Land-Cover Products If aascientist is interested in inventorying land cover at the rregional,national, and global scale, If scientist is interested in inventorying land cover at the egional, national, and global scale, then the modified International Geosphere-Biosphere Program Land-Cover Classification then the modified International Geosphere-Biosphere Program Land-Cover Classification System may be appropriate. For example, the Moderate Resolution Imaging System may be appropriate. For example, the Moderate Resolution Imaging Spectroradiometer (MODIS) of NASA's Earth Observing System (EOS) is providing global Spectroradiometer (MODIS) of NASA's Earth Observing System (EOS) is providing global land-surface information at spatial resolutions of 250 to 1,000 m. There are approximately 44 land-surface information at spatial resolutions of 250 to 1,000 m. There are approximately 44 standard MODIS-derived data products that scientists are using to study global change. The standard MODIS-derived data products that scientists are using to study global change. The MODIS Land Science Team is producing aaglobal land-cover change product at 11-km(0.6 MODIS Land Science Team is producing global land-cover change product at -km (0.6 mile) resolution to depict broad-scale land-cover changes. mile) resolution to depict broad-scale land-cover changes. The land-cover type and land-cover change parameters are produced at 11-kmresolution on aa The land-cover type and land-cover change parameters are produced at -km resolution on quarterly basis. The land-cover parameter identifies 17 categories of land-cover following the quarterly basis. The land-cover parameter identifies 17 categories of land-cover following the IGBP global vegetation database which defines nine classes of natural vegetation, three IGBP global vegetation database which defines nine classes of natural vegetation, three classes of developed lands, two classes of mosaic lands, and three classes of nonvegetated classes of developed lands, two classes of mosaic lands, and three classes of nonvegetated lands (snow/ice, bare soil/rocks, water). The first global land-cover map based on MODIS lands (snow/ice, bare soil/rocks, water). The first global land-cover map based on MODIS data was distributed in August, 2002. data was distributed in August, 2002. Observations about Classification Schemes Observations about Classification Schemes Geographical information (including remote sensor data) is often imprecise. For Geographical information (including remote sensor data) is often imprecise. For example, there is usually aa gradual transition at the interface of forests and example, there is usually gradual transition at the interface of forests and rangeland, yet many of the aforementioned classification schemes insist on aa hard rangeland, yet many of the aforementioned classification schemes insist on hard boundary between the classes at this transition zone. The schemes should contain boundary between the classes at this transition zone. The schemes should contain fuzzy definitions because the thematic information they contain is fuzzy. Fuzzy fuzzy definitions because the thematic information they contain is fuzzy. Fuzzy classification schemes are not currently standardized. They are typically developed classification schemes are not currently standardized. They are typically developed by individual researchers for site-specific projects. The fuzzy classification systems by individual researchers for site-specific projects. The fuzzy classification systems developed may not be transferable to other environments. Therefore, we tend to see developed may not be transferable to other environments. Therefore, we tend to see the use of existing hard classification schemes, which are rigid, based on a priori the use of existing hard classification schemes, which are rigid, based on a priori knowledge, and generally difficult to use. They continue to be widely employed knowledge, and generally difficult to use. They continue to be widely employed because they are scientifically based and different individuals using the same because they are scientifically based and different individuals using the same classification system can compare results classification system can compare results Observations about Classification Schemes Observations about Classification Schemes If aareputable classification system already exists, it is If reputable classification system already exists, it is foolish to develop an entirely new system that will probably foolish to develop an entirely new system that will probably be used only by ourselves. It is better to adopt or modify be used only by ourselves. It is better to adopt or modify existing nationally or internationally recognized existing nationally or internationally recognized classification systems. This allows us to interpret the classification systems. This allows us to interpret the significance of our classification results in light of other significance of our classification results in light of other studies and makes it easier to share data. studies and makes it easier to share data. Observations about Classification Schemes Observations about Classification Schemes There is aa relationship between the level of detail in aa There is relationship between the level of detail in classification scheme and the spatial resolution of remote classification scheme and the spatial resolution of remote sensor systems used to provide information. Welch (1982) sensor systems used to provide information. Welch (1982) summarizes this relationship for mapping urban/suburban summarizes this relationship for mapping urban/suburban land use and land cover This suggests that the level of detail land use and land cover This suggests that the level of detail in the desired classification system dictates the spatial in the desired classification system dictates the spatial resolution of the remote sensor data that should be used. Of resolution of the remote sensor data that should be used. Of course, the spectral resolution of the remote sensing system course, the spectral resolution of the remote sensing system is also an important consideration, especially when is also an important consideration, especially when inventorying vegetation, water, ice, snow, soil, and rock. inventorying vegetation, water, ice, snow, soil, and rock. Nominal spatial resolution Nominal spatial resolution requirements as aafunction of requirements as function of the mapping requirements for the mapping requirements for Levels IIto IV land-cover Levels to IV land-cover classes in the United States classes in the United States (based on Anderson et al., (based on Anderson et al., 1976). Note the dramatic 1976). Note the dramatic increase in spatial resolution increase in spatial resolution required to map Level II classes. required to map Level II classes. Relationship between the level of Relationship between the level of detail required and the spatial detail required and the spatial resolution of representative resolution of representative remote sensing systems for remote sensing systems for vegetation inventories. vegetation inventories. ...
View Full Document

This note was uploaded on 09/08/2010 for the course GEOG 182 at San Jose State University .

Ask a homework question - tutors are online