İstanbul Çatalca bölgesinde uzaktan algılama yöntemleri ile metropoliten analizi
İstanbul Çatalca bölgesinde uzaktan algılama yöntemleri ile metropoliten analizi
Dosyalar
Tarih
1993
Yazarlar
Alkan, F. Zehra
Süreli Yayın başlığı
Süreli Yayın ISSN
Cilt Başlığı
Yayınevi
Fen Bilimleri Enstitüsü
Özet
Endüstrileşme ile paralel olarak uzay teknolojisinin hızla gelişmesi, dünya nüfusundaki oranın artması ve buna bağlı olarak tükenen doğal kaynakların incelenmesi, çevre ile ilgili diğer sorunlar, yerleşim ve araziden yararlanma, doğal enerji kaynakları, hammadde ve çevre koşullan gereksinimlerinin karşılanması, deniz ve kara yüzeyleri ile alt yapıların özellik ve konumlarının incelenmesi konutlarının hızlı, ekonomik ve güncel şekilde sağlanması gerektirmektedir.Uzaktan algılama arada mekanik bir temas olmaksızın cisimden yayılan ışımının nitelik ve nicelik yönünden değerlendirilmesi ile cismin özelliklerinin uzaktan ortaya konması ve ölçülmesi şeklinde tanımlanmaktadır. Çok yönlü bilgisayar sistem ve programlarının hızla gelişmesi ile uzaktan algılama pek çok bilim dalı içerisinde uygulama alanı bularak geniş bir şekilde kullanılmaktadır. Uzay teknolojisinin gelişmesine bağlı olarak çeşitli amaçlarla çok sayıda uydu, günümüzde işlevini sürdürmektedir. Bu uydular arasından yer yüzeyinin bilimsel araştırılmasına yönelik olarak algılama yapan LANDSAT (land-satallite) uydularından ikinci varyasyon olarak bilinen^ LANDSAT 4 ve 5 uyduları çok spektrumlu tarayıcı (MSS) ve tem atik haritalama (TM) sistemleri ile yeryüzeyi algılanmaktadır. 1.3.1984 tarihinde yörüngeye oturtulan Landsat-5 uydusu, diğer algılayıcılara göre 30 m çözünürlük ve 7 kanal sayısı üstünlüğü nedeni ile bir çok alanda kullanılmaktadır. Oluşturulan renkli görüntüler, konturollü sınıflandırma yöntemlerinden Maksimum Likelihood ve minimum uzaklık, kontrolsüz sınıflandırma yöntemlerinden Ardışık kümeleme ve ISOOATA kümeleme yöntemlerine göre farklı kanallardan sınıflandırılmıştır. Bölgenin yersel bilgileri ışığında özellikleri değerlendirilerek sınıflandırılmış görüntülerin yorumları yapılmıştır. Bölgenin analizinin uydu verileri ile en iyi şekilde belirlenebileceğini çalışma sonuçları göstermiştir. Böylelikle uygulama bölgesinin yerleşim, sanayi, orman, ekili ve sulara kapalı alanları hakkında yeterli bilgiyi sağlamak üzere çalışma sonuçları hazırlanmıştır. Ayrıca farklı sınıflandırma yöntemleri ile oluşturulan görüntülerin yorumlan kendi aralarında da kıyaslanarak sonuçların benzerliği ortaya konmuştur.
Remote Sensing is the science of detection the nature of objects on the earth surface without touching them. With development of the earth's thecnology fastly.many problems became to be solved Increasing of the human population in connection with environmental pollution.urban planning.reduetion^ natural resources and the others expect urgent and economical solution. The matter like that require interdisciplinary cooperation, modem research strategies and a global perspective. Remote sensing will be an important key tool for solving environmental probems in many kinds of sciencetific branches. Remote sensing methods provide not only an overall view of the ground surface but also enable specific areas to be examined in great details. Acquisition of information about an object are ensured by sensors. Sensor systems used in remote sensing are photographic and non-photographic systems and space programs by aircraft and airborne which enable us to apply the improved capability to study natural resources, economic potentials.environmental engineering applications from data obtained that devices. Application of remote sensing have two steps in general. Acquisition of data using proper sensors and data analysis as computer image processing, classification and interpretation. The source, the atmospheric path, the target and the sensors are four component of remote sensing data-acquisition system. Electromagnetic radiation is the form between the compenents of the remote sensing system. Electromagnetic radiation occurs as a continuum of wavelengths and frequencies from short wavelength, high frequency cosmic waves to long wavelength, low frequency radio waves. In remote sensing, visible and near infrared radiation in the waveband 0.4-0.9 urn infrared radiation in the waveband 5-500 mm are used. The sun is a basic source of electromagnetic radiation. Energy of sun is spectrally distributed through the electromagnetic spectrum which is reflected, transmitted, absorbed and emitted. The reflected energy passes back through the atmospher and again is subjected to spectral and intensity modifications. The acquisition of data from aircraft sensors used in remote sensing applications is affected by the atmosphere between the sensor -vii- and the target The atmospher effects remote sensing data as scatterring and absorption of energy. Scattering occurs when radiation is reflected by particles in the atmosphere and causes some of the energy to be directed outside the field of the sensors. Pure scattering is said to occur in the absence of all absorption. Photographic and non-photographic systems used for data acquisition in remote sensing have capability of sensing visible and near infrared regions of the electromagnetic spectrum. The airborne scanner has improved methods for detecting and recording both the visible, near infrared and longer-wavelength bands, such as thermal infrared and microwave energy. A rotating mirror which sends the radiation on to a detector is used for scanning. Purpose of scanner is to detect measure and-record- electromagnetic ^nergyv Commonly used ^on-photogfaphing systems are satallite-based systems. The development of satallite-based MSS (Multi Spectral Scanner ) systems for commercial use had been started to take information from space in 1960s but MSS imagery become available in 1970s. The majority of sattelite MSS imagery has been obtained from unmanned platforms, in particular from the LANDSAT semes of satellites. The LANDSAT satellite system, ERTS (Earth Resources Technology Satellite) was initiated in 1967 by NASA. LANDSAT-1 was launched in 1972 and equipped with a RBV (Return Beam vldicon) television camera system and four waveband MSS. Since it was launched, remote sensing has proved valuable in various fields such as geology, hydrology, forestry, urban planning, etc. LANDSAT-2 was launched in 1975 and LANDSAT-3 was launched in 1978. LANDSAT 1,2 and 3 had an orbit height of 910km and capacity to image the entire Earth's surface every 18 days. Second generation of LANDSAT satellites are LANDSAT-4 and LANDAST-5 were launched ( 1983 and 1984 ) into repetitive, circular, sun-synchronous, near-polar orbits. The satelite orbit is at an altitude of 705 km. These satallites have either the MSS or Thematic Mapper (TM ) sensors. Transmission of MSS and TM data to ground receiving stations is made possible via the antennas onboard the satellite. The TM sensor is a highly advanced MSS incorporating a number of spectral, radiometric, and geometric design improvements relative to the MSS. The acquisition of data are included in seven bands. Radiometrically, the TM performs its onboard analog to digital signal conversion over a range of 256 digital numbers. Geometrically, TM data resollution is a 30x30 m and have repetitive multispectral covarege and minimal image distortion. LANDSAT data initially used to obtain a synoptic view of a large area of the Earth 's surface for interpretation with aid of aerial photographic interpretation techniques but today, advantage of the digital image processing of remote sensing data, the monitoring of marine -viii- environments, urban and industrial areas, etc. has become one of the most effective and uptodate observastion techniques. in remote sensing, digital images are defined by discrete picture elements called pixels. The average brightness of small area whit in a scene is formed associated pixels. The size of this area affects the scene. The image has a great number of pixels and is formed a matrix. The image is held on magnetic tape and in particular on computer compatible tape ( CCT ). Multispectral images reqire special consideration and are stored in interleaved formats BIP, BIL or BSQ. Image processing and classification stages are the basis of remote sensing. Image processing helps the analyst supervises the process choosing the information categories or classes, selecting training areas the method is called 'Supervised Classification'. In supervised classification, the analyst employs a computer algorithm locating naturally occuring specifications from a heterogeneous sample of pixels. Multispectral classification is the process of sorting pixels into a finite number of individual classes or categories of data, based on their data file values. If a pixel satisfies a certain set of criteria, the pixel is assigned to class that corresponds to that criteria. Depending on the type of information exracted from the orijinal data, classes may be associated with known features on the ground or may simply represent areas that look different to the computer. Pattern recognition is the science and art of finding meaningful patterns in data.which can be extracted through classification. By spetially and spectrally enhancing an image, pattern recognition can be performed with the human eye. With a computer system, statistics are derived from the spectral characteristics of all pixels in an image. Then, the pixels are sorted based on mathematical criteria. The classification process forms into two parts; training and classifying. Training is the process of defining the criteria by which patterns are recognized inthe data. The resulting of training is a set of signatures being statistical criteria for a set of proposed classes. After training pixels of an image are sorted into classes based on the the signatures, by use of a classification decision rule which is a mathematical algorithm uses particular statistics, contained in the signature, to sort the pixels. By identifying patterns, computer system can be trained to identify pixels with similar characteristics. If the classification is accurate, -ix- then each resulting class represents an area of interest within the data that corresponds with the pattern that you originally identified. Beside supervised training, unsupervised training is more computer automated. Some parameters which the computer uses are specified by the analyst to uncover statistical patterns that are inherent in the data. These patterns do not neccessarily correspond to directly meaningful characteristicsof the scene. They are simply groups of pixels with similar spectral characteristics. Unsupervised training is dependent upon the data itself for the definition of classes. This method is usually used when less is known about the data before classification. Analyst is responsible to attach meanings to resulting classes after classification. Unsupervised classification is only useful if the classes can be appropriately interpreted. Usually, classification is performed with a set of target classes in mind. Such a set is called a classification system and the purpose of this system is to provide a framework for organizing and- categorizing the information that can be exracted from the data. Most systems have a hierarchical structer, which can describe a study area in several levels of detail. Maximum Likelihood (Bayesian) algoritm assumes that the histograms of the bands of data have normal distributions and rule is based on the probabilty that a pixel belongs to a particular class. The basic equation supposes that this probabilities are equal for all classes. If a priori knowledge is known that the probabilities are not equal for all classes, weight factors for particular classes can specified. This variation of the Maximum Likelihood decision rule is known as Bayesian decision rule (Hord 1982). Advantages of this rule is to take the most varilables into cosideration consequently it is accurate. In spite of this, Maximum Likelihood is very parametric, meaning that it relies heavily on a normal distribution of data in each input band. The computation of this process takes a long time and his time increases whit the number of input bands. Maximum - likelihood tends to overclassify signatures with relatively large values in the co variance matrix. If there is large dispertion of the pixels in a training sample, then the covariance matrix of that signature will contain large values. The Minimum Distance decision rule calculate the spectral distance between the measurement vector for the candidate pixel and the mean vector for signature. When spectral distance is computed for all possible values of all possible classes, the class of the candidate pixel is assigned to the class for which spectral distance is the lowest The advantages of this rule, since every pixel is spectrally closer to either one sample mean or another, there are no unclassified pixels. Minimum distance is the fastest decision rule to compute. However, Minimum Distance does not consider class variabirty. For example, a class like an urban land cover class is made up of pixels with a high -x- variance, which may tend to be farther from the mean of the signature. In a minimum distance classification, outlying urban pixels may be improperly classified. Inversly, a class with less variance, like water, may tend to overclassify, because the pixels that belong to the class are usually spectrally closer to their mean than those of other classes to their means. Unsupervised training requires minimal initial input It is based on the natural groupings of pixels in image data when they are plotted in spectral space. According to parameters specified, these groups can later be merged, disregarded, manipulated or used as the basis of signature. In the sequential method, pixels are examined one at a time. The spectral distances between each analyzed pixel and the means of previously defined clusters are calculated. Each pixel either contributes to an existing cluster, or begin a new cluster, based on the spectral distances. Clusters are merged if too many are formed. Pixels are analyzed beginning with upper left comer of the image and going left to right, line by line. The measurement vector of the first pixel becomes the the mean vector of the first cluster, with the weight of 1. The weight of each cluster is the number pixels in it. In merging and comparing clusters, the weight is taken in to consideration. Due to sequential methods reguires little set up only a few parameters need to be determined. The advantage of sequential method is much faster than the other clustering methods and the preparation of methods is easier. However it is slightly biased to the top of the data file, since the first pixels that are analyzed tend to weight the means of the clusters more heavily, the method is pra metric that means the algorithm works on the assumption that the data distribution is normal. ISODATA clustering stands for " Iterative Self - Organizing Data Analysis Technique ". This method is iterative in that it repetedly performs an entire classification and recalculates statistics and locates clusters with minimum user input. The ISODATA process begins with a specified number of arbitrary cluster means and processes repetively, so that those arbitrary means will shift to the means of the clusters in data. It is not biased to the top of the data file, as are the one-pass clustering algorithms. The advantages of ISODATA methods are that it is not as parametric as the other clustering algorithm and not geographically biased to the top of the dta file, because it is iterative. On the other hand the disadvanteges of it are that it can processes many times, and is the slowest of the clustering methods. -xi- In this study LANDSAT-5 TM data dated 6.9.1992 were evaluated to Çatalca region in Istanbul Metropolitan ares classifications by means of computer assisted techniques. Firstly, by classification of Maximum-Likelihood methods, trainning samples are processed, evaluated and interpreted in different band combination by means of supervised classification rules. After the same topics are applied by Minimum Distance method with supervised trainning Secondly with unsupervised classification methods Sequential Clustering and ISODATA clustering, process are examined by uncontrolled in unsupervised classification rule. Methods are compared with supervised and unsupervised methods. Because of supervised methods are controlled by the analyst, results are more positive than unsupervised methods. But in some condition unsupervised decision rule is suited. In this study this rule is applicated for comparing with the other results. The results showed that Remote Sensing Methods are very suitable and contain understandable easy applications in mind.
Remote Sensing is the science of detection the nature of objects on the earth surface without touching them. With development of the earth's thecnology fastly.many problems became to be solved Increasing of the human population in connection with environmental pollution.urban planning.reduetion^ natural resources and the others expect urgent and economical solution. The matter like that require interdisciplinary cooperation, modem research strategies and a global perspective. Remote sensing will be an important key tool for solving environmental probems in many kinds of sciencetific branches. Remote sensing methods provide not only an overall view of the ground surface but also enable specific areas to be examined in great details. Acquisition of information about an object are ensured by sensors. Sensor systems used in remote sensing are photographic and non-photographic systems and space programs by aircraft and airborne which enable us to apply the improved capability to study natural resources, economic potentials.environmental engineering applications from data obtained that devices. Application of remote sensing have two steps in general. Acquisition of data using proper sensors and data analysis as computer image processing, classification and interpretation. The source, the atmospheric path, the target and the sensors are four component of remote sensing data-acquisition system. Electromagnetic radiation is the form between the compenents of the remote sensing system. Electromagnetic radiation occurs as a continuum of wavelengths and frequencies from short wavelength, high frequency cosmic waves to long wavelength, low frequency radio waves. In remote sensing, visible and near infrared radiation in the waveband 0.4-0.9 urn infrared radiation in the waveband 5-500 mm are used. The sun is a basic source of electromagnetic radiation. Energy of sun is spectrally distributed through the electromagnetic spectrum which is reflected, transmitted, absorbed and emitted. The reflected energy passes back through the atmospher and again is subjected to spectral and intensity modifications. The acquisition of data from aircraft sensors used in remote sensing applications is affected by the atmosphere between the sensor -vii- and the target The atmospher effects remote sensing data as scatterring and absorption of energy. Scattering occurs when radiation is reflected by particles in the atmosphere and causes some of the energy to be directed outside the field of the sensors. Pure scattering is said to occur in the absence of all absorption. Photographic and non-photographic systems used for data acquisition in remote sensing have capability of sensing visible and near infrared regions of the electromagnetic spectrum. The airborne scanner has improved methods for detecting and recording both the visible, near infrared and longer-wavelength bands, such as thermal infrared and microwave energy. A rotating mirror which sends the radiation on to a detector is used for scanning. Purpose of scanner is to detect measure and-record- electromagnetic ^nergyv Commonly used ^on-photogfaphing systems are satallite-based systems. The development of satallite-based MSS (Multi Spectral Scanner ) systems for commercial use had been started to take information from space in 1960s but MSS imagery become available in 1970s. The majority of sattelite MSS imagery has been obtained from unmanned platforms, in particular from the LANDSAT semes of satellites. The LANDSAT satellite system, ERTS (Earth Resources Technology Satellite) was initiated in 1967 by NASA. LANDSAT-1 was launched in 1972 and equipped with a RBV (Return Beam vldicon) television camera system and four waveband MSS. Since it was launched, remote sensing has proved valuable in various fields such as geology, hydrology, forestry, urban planning, etc. LANDSAT-2 was launched in 1975 and LANDSAT-3 was launched in 1978. LANDSAT 1,2 and 3 had an orbit height of 910km and capacity to image the entire Earth's surface every 18 days. Second generation of LANDSAT satellites are LANDSAT-4 and LANDAST-5 were launched ( 1983 and 1984 ) into repetitive, circular, sun-synchronous, near-polar orbits. The satelite orbit is at an altitude of 705 km. These satallites have either the MSS or Thematic Mapper (TM ) sensors. Transmission of MSS and TM data to ground receiving stations is made possible via the antennas onboard the satellite. The TM sensor is a highly advanced MSS incorporating a number of spectral, radiometric, and geometric design improvements relative to the MSS. The acquisition of data are included in seven bands. Radiometrically, the TM performs its onboard analog to digital signal conversion over a range of 256 digital numbers. Geometrically, TM data resollution is a 30x30 m and have repetitive multispectral covarege and minimal image distortion. LANDSAT data initially used to obtain a synoptic view of a large area of the Earth 's surface for interpretation with aid of aerial photographic interpretation techniques but today, advantage of the digital image processing of remote sensing data, the monitoring of marine -viii- environments, urban and industrial areas, etc. has become one of the most effective and uptodate observastion techniques. in remote sensing, digital images are defined by discrete picture elements called pixels. The average brightness of small area whit in a scene is formed associated pixels. The size of this area affects the scene. The image has a great number of pixels and is formed a matrix. The image is held on magnetic tape and in particular on computer compatible tape ( CCT ). Multispectral images reqire special consideration and are stored in interleaved formats BIP, BIL or BSQ. Image processing and classification stages are the basis of remote sensing. Image processing helps the analyst supervises the process choosing the information categories or classes, selecting training areas the method is called 'Supervised Classification'. In supervised classification, the analyst employs a computer algorithm locating naturally occuring specifications from a heterogeneous sample of pixels. Multispectral classification is the process of sorting pixels into a finite number of individual classes or categories of data, based on their data file values. If a pixel satisfies a certain set of criteria, the pixel is assigned to class that corresponds to that criteria. Depending on the type of information exracted from the orijinal data, classes may be associated with known features on the ground or may simply represent areas that look different to the computer. Pattern recognition is the science and art of finding meaningful patterns in data.which can be extracted through classification. By spetially and spectrally enhancing an image, pattern recognition can be performed with the human eye. With a computer system, statistics are derived from the spectral characteristics of all pixels in an image. Then, the pixels are sorted based on mathematical criteria. The classification process forms into two parts; training and classifying. Training is the process of defining the criteria by which patterns are recognized inthe data. The resulting of training is a set of signatures being statistical criteria for a set of proposed classes. After training pixels of an image are sorted into classes based on the the signatures, by use of a classification decision rule which is a mathematical algorithm uses particular statistics, contained in the signature, to sort the pixels. By identifying patterns, computer system can be trained to identify pixels with similar characteristics. If the classification is accurate, -ix- then each resulting class represents an area of interest within the data that corresponds with the pattern that you originally identified. Beside supervised training, unsupervised training is more computer automated. Some parameters which the computer uses are specified by the analyst to uncover statistical patterns that are inherent in the data. These patterns do not neccessarily correspond to directly meaningful characteristicsof the scene. They are simply groups of pixels with similar spectral characteristics. Unsupervised training is dependent upon the data itself for the definition of classes. This method is usually used when less is known about the data before classification. Analyst is responsible to attach meanings to resulting classes after classification. Unsupervised classification is only useful if the classes can be appropriately interpreted. Usually, classification is performed with a set of target classes in mind. Such a set is called a classification system and the purpose of this system is to provide a framework for organizing and- categorizing the information that can be exracted from the data. Most systems have a hierarchical structer, which can describe a study area in several levels of detail. Maximum Likelihood (Bayesian) algoritm assumes that the histograms of the bands of data have normal distributions and rule is based on the probabilty that a pixel belongs to a particular class. The basic equation supposes that this probabilities are equal for all classes. If a priori knowledge is known that the probabilities are not equal for all classes, weight factors for particular classes can specified. This variation of the Maximum Likelihood decision rule is known as Bayesian decision rule (Hord 1982). Advantages of this rule is to take the most varilables into cosideration consequently it is accurate. In spite of this, Maximum Likelihood is very parametric, meaning that it relies heavily on a normal distribution of data in each input band. The computation of this process takes a long time and his time increases whit the number of input bands. Maximum - likelihood tends to overclassify signatures with relatively large values in the co variance matrix. If there is large dispertion of the pixels in a training sample, then the covariance matrix of that signature will contain large values. The Minimum Distance decision rule calculate the spectral distance between the measurement vector for the candidate pixel and the mean vector for signature. When spectral distance is computed for all possible values of all possible classes, the class of the candidate pixel is assigned to the class for which spectral distance is the lowest The advantages of this rule, since every pixel is spectrally closer to either one sample mean or another, there are no unclassified pixels. Minimum distance is the fastest decision rule to compute. However, Minimum Distance does not consider class variabirty. For example, a class like an urban land cover class is made up of pixels with a high -x- variance, which may tend to be farther from the mean of the signature. In a minimum distance classification, outlying urban pixels may be improperly classified. Inversly, a class with less variance, like water, may tend to overclassify, because the pixels that belong to the class are usually spectrally closer to their mean than those of other classes to their means. Unsupervised training requires minimal initial input It is based on the natural groupings of pixels in image data when they are plotted in spectral space. According to parameters specified, these groups can later be merged, disregarded, manipulated or used as the basis of signature. In the sequential method, pixels are examined one at a time. The spectral distances between each analyzed pixel and the means of previously defined clusters are calculated. Each pixel either contributes to an existing cluster, or begin a new cluster, based on the spectral distances. Clusters are merged if too many are formed. Pixels are analyzed beginning with upper left comer of the image and going left to right, line by line. The measurement vector of the first pixel becomes the the mean vector of the first cluster, with the weight of 1. The weight of each cluster is the number pixels in it. In merging and comparing clusters, the weight is taken in to consideration. Due to sequential methods reguires little set up only a few parameters need to be determined. The advantage of sequential method is much faster than the other clustering methods and the preparation of methods is easier. However it is slightly biased to the top of the data file, since the first pixels that are analyzed tend to weight the means of the clusters more heavily, the method is pra metric that means the algorithm works on the assumption that the data distribution is normal. ISODATA clustering stands for " Iterative Self - Organizing Data Analysis Technique ". This method is iterative in that it repetedly performs an entire classification and recalculates statistics and locates clusters with minimum user input. The ISODATA process begins with a specified number of arbitrary cluster means and processes repetively, so that those arbitrary means will shift to the means of the clusters in data. It is not biased to the top of the data file, as are the one-pass clustering algorithms. The advantages of ISODATA methods are that it is not as parametric as the other clustering algorithm and not geographically biased to the top of the dta file, because it is iterative. On the other hand the disadvanteges of it are that it can processes many times, and is the slowest of the clustering methods. -xi- In this study LANDSAT-5 TM data dated 6.9.1992 were evaluated to Çatalca region in Istanbul Metropolitan ares classifications by means of computer assisted techniques. Firstly, by classification of Maximum-Likelihood methods, trainning samples are processed, evaluated and interpreted in different band combination by means of supervised classification rules. After the same topics are applied by Minimum Distance method with supervised trainning Secondly with unsupervised classification methods Sequential Clustering and ISODATA clustering, process are examined by uncontrolled in unsupervised classification rule. Methods are compared with supervised and unsupervised methods. Because of supervised methods are controlled by the analyst, results are more positive than unsupervised methods. But in some condition unsupervised decision rule is suited. In this study this rule is applicated for comparing with the other results. The results showed that Remote Sensing Methods are very suitable and contain understandable easy applications in mind.
Açıklama
Tez (Yüksek Lisans) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 1993
Anahtar kelimeler
Uzaktan algılama,
İstanbul-Çatalca,
Şehircilik,
Remote sensing,
İstanbul-Çatalca,
Urbanization