CHAPTER II LITERATURE REVIEW. ALOS (Advanced Land Observation Satellite) was successfully launched on January 24,

Transcription

1 5 CHAPTER II LITERATURE REVIEW 2.1 ALOS Image (Advanced Land Observation Satellite) ALOS (Advanced Land Observation Satellite) was successfully launched on January 24, 2006 from the Tanegashima Space Centre. The ALOS (renamed "Daichi") has three remote sensing instruments: the Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM), the Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2), and the Phased Array type L-band Synthetic Aperture Radar (PALSAR) which in turn show in Figure 1. Structures and components ALOS show in the Figure 1. PRISM using for digital elevation mapping (DEMs) and AVNIR-2 is for precise land coverage observation. And PALSAR for day-and-night and allweather land observation and enables precise land coverage observation and can collect enough data by itself for mapping on a scale of 1:25,000, without relying on points of reference on the ground. Some of its objectives are cartography, disaster monitoring, natural resource surveys and technology development. To Major Characteristics ALOS imagery show in Table 1 (Anonim, 2001). Figure 1: Components and Sensor of ALOS (Anonim, 2001) Table. 1 ALOS Satellite Sensor 5 Characteristics

2 6 Resolution Launch Vehicle Launch Site Satellite Weight Power Designed Life Orbit 2.5m panchromatic 10m multispectral H-IIA Rocket Tanegashima Space Centre Approximately 4,000kg (at Lift-off) Approximately 7,000W (End of Life) 3 to 5 years Sun Synchronous Sub-Recurrent Orbit Recurrent Period: 46 days Sub cycle: 2 days Altitude: Approximately 692km (above the equator) Inclination: Approximately 98.2 degrees Characteristics of ALOS AV IR-2 Images (Advanced Visible and ear Infrared Radiometer Type 2) AVNIR-2 is a successor to AVNIR that was on board the Advanced Earth Observing Satellite (ADEOS), which was launched in August Its instantaneous field-of-view (IFOV) is the main improvement over AVNIR. AVNIR-2 also provides 10 m spatial resolution images, an improvement over the 16m resolution of AVNIR in the multi-spectral region. Improved CCD detectors (AVNIR has 5,000 pixels per CCD; AVNIR-2 7,000 pixels per CCD) and electronics enable this higher resolution, for the length of the track ALOS AVNIR-2 satellite show in Figure 2. A cross-track pointing functions for prompt observation of disaster areas is another improvement. The pointing angle of AVNIR-2 is +44 and - 44 degree (ALOS, 2007). Figure 2. Long tracks of ALOS (AVNIR 2) (Schneider Mathias, et al., 2006).

3 7 The instrument has a pointing function (±44 ), which makes it possible to change target areas in the cross-track direction for prompt observation and then for the Main Specifications of the AVNIR-2 show in Table 2. Table 2. AVNIR-2 Main Specifications o. of bands 4 Wavelength Band1: µm (blue) Band2: µm (green) Band3: µm (red) Band4: µm (NIR) Ground resolution 10 m (nadir-looking) Swath 70 km (nadir) Signal-to-noise ratio >200 Spatial frequency transfer function Bands 1-3: >0.25 Band 4: >0.20 o. of detectors 7,000/band Pointing angle ±44 (cross-track direction) Bit length 8 bits AVNIR-2 data are available in Level 1A, Level 1B1 and Level 1B2, which are generated by applying radiometric and geometric corrections to the acquired data. For the Definitions of processing levels of AVNIR-2 products show in Table 3. Level 1A 1B1 Table 3. Definitions of Processing Levels of AVNIR-2 Products Definitions This is AVNIR-2 raw data, generated by subset of the observed data into scenes and applying decompression and line generation. Radiometric and geometric information necessary for Level 1B processing and onward is applied. This product is generated by applying radiometric correction to the Level 1A data and adding absolute correction coefficients. Geometric and other necessary information for Level 1B2 processing and onward

4 8 1B2 is applied. This product is generated by applying geometric correction to the Level 1B1 data. The following correction options are available. R:Geo-referenced G:Registered to maps. Geo-coded. In the ALOS Data Processing Subsystem, the scene of a Raw product (geometrically uncorrected) and a Geo-reference product (map-projected based on the flight direction) are defined by determining image position and image range use input data according to the RSP (Reference System for Planning). And the scene of a Geo-coded product (projected based on the direction on the map) is defined by rotating the same range of the Geo-reference image to mapnorth (NEC/THOSHIBA, 2009). This section describes the definitions of the scene related information for AVNIR-2 products. 1. Uncorrected image AVNIR-2 has four bands, and each band has 7100-pixel CCD. The odd number pixels and the even number pixels are arrayed by stagger alignment, where both pixels are approximately 5 pixels away from each other. The odd number pixels come first and are located below (satellite direction). Since there is a registration error between bands, the imaging position of each band is not the same exactly (NEC/THOSHIBA, 2009). 2. Level 1B2 Geo-referenced image Level 1B2 Geo-reference image is framed based on the centreline of an uncorrected image and is a map-projected image with 70 km x 7000 lines (when pixel spacing is 10 m). Column direction of the Geo-reference image is framed to fit inside the effective area of the uncorrected image in the Geo-referenced image (variable length). Band three is used for framing.

5 9 For the ascending image, image direction is flipped to make nearly north of the image upward. (Satellite direction will be upward.) (NEC/THOSHIBA, 2009). 3. Level 1B2 Geo-coded image Level 1B2 Geo-coded image is made by framing to make map-north upward. In this case, framing is done by making the four corner points of the Geo-reference image touch the Geocoded image sides (NEC/THOSHIBA, 2009) ALOS PRISM (Panchromatic Remote-Sensing Instrument For Stereo Mapping) PRISM (Panchromatic Remote Sensing Instruments for Stereo Mapping) consists of forward-looking, nadir-looking and backward-looking radiometers. PRISM observes the ground from 3 directions within an orbit using the 3 radiometers. Each radiometer has 6 or 8 CCDs on its focal plane. Its data are used for creating high-precision Digital Elevation Models (DEM) for the length of the track PRISM satellite show in Figure 3. The instrument has three optical systems for acquiring topographic data with altitude information. It acquires data in three directions forward, nadir and backward along the satellite track at the same time. Usually, 4 CCDs are used for a radiometer. Pixel size is designed to be 2.5 m. One of the most important objectives of PRISM is medium-scale mapping and DEM production without ground control points. For Main the Specifications of the PRISM show in Table 4 (Kamiya, 2007).

6 10 Figure 3. Long tracks of ALOS (PRISM) (Schneider, et al., 2006). Table 4. PRISM Main Specifications o. of bands 1 (panchromatic) Wavelength 0.52~0.77µm o. of optics 3 (nadir-, forward- and backward views) Stereoscopic B/H ratio 1.0 (between forward and backward views) Ground resolution 2.5 m (nadir view) Swath 70 km (nadir view only) / 35 km (triplet mode ) Signal-to-noise ratio >70 Spatial frequency >0.2 transfer function o. of detectors 28,000/band (Swath 70km) 14,000/band (Swath 35km) Pointing angle ±1.5 (triplet mode, cross-track direction) Bit length 8 bits Each optic system consists of three off axis mirrors for push-broom scanning. The swath for nadir-view observation is 70 km, while that of forward and backward views is 35 km. This enables frequent acquisitions of highly-accurate topographic data. PRISM data are available in Level 1A, Level 1B1 and Level 1B2, which are generated by applying radiometric and geometric corrections to the acquired data. For the Definitions of processing levels of AVNIR-2 products show in Table 5.

7 11 Level 1A 1B1 1B2 Table 5. Definitions of Processing Levels of PRISM Products Definition This is PRISM raw data, generated by subset of the observed data into scenes and applying decompression and line generation. Radiometric and geometric information necessary for Level 1B processing and onward is added. This product is generated by applying radiometric correction to the Level 1A data and adding absolute correction coefficients. Geometric and other necessary information for Level 1B2 processing and onward is added. This product is generated by applying geometric correction to the Level 1B1 data. The following correction options are available. R:Geo-referenced G:Registered to maps. Geo-coded. PRISM consists of three independent radiometers for nadir (N), backward (B) and forward (F) view. Each radiometer is composed of 6 (N) 8 (F, B) CCD-arrays containing 4992 or 4928 pixels for nadir or forward/backward views respectively. There is a nominal overlap of 32 pixels between two neigh boring CCD-arrays. Usually, an image is acquired using a subset of 4 consecutive CCD-arrays. The pixels, which are not used on the right and left CCD-array respectively, are regarded as so called dummy pixels and not used for the processing (Schneider, et al., 2006). 2.2 Geometric Correction Geometric correction is undertaken to avoid geometric distortions from a distorted image, and is achieved by establishing the relationship between the image coordinate system and the geographic coordinate system using calibration data of the sensor, measured data of position and attitude, ground control points, atmospheric condition etc. The most widely used geometric

8 12 correction technique relies on the use of Ground Control Points (GCPS) located on the image and the corresponding map in order to empirically determine a mathematical coordinate transformation to correct the geometry. The quality of the correction will depend on the number and quality of the control points which is often related to the quality of the source (Kardoulas, 1996). Internal geometric errors are introduced by the remote sensing system itself or in combination with Earth rotation or curvature characteristics. These distortions are often systematic (predictable) and may be identified and corrected using pre-launch or in-flight platform ephemeris (i.e., information about the geometric characteristics of sensor and the Earth at data acquisition). Geometric distortions in imagery that can sometimes be corrected through analysis of sensor characteristics and ephemeris data include: 1) Skew caused by Earth rotation effects, 2) Scanning system induced variation in ground resolution cell size, relief displacement, and tangential scale distortion (Fahim, 2009). External geometric errors are usually introduced by phenomena that vary in nature through space and time. The most important external variables that can cause geometric error in remote sensor data are random movements by the aircraft (or spacecraft) at the exact time of data collection, which usually involve (Kardoulas, 1996): 1) Altitude changes, and/or 2) Attitude changes (roll, pitch, and yaw). In order for a GIS software package to correctly display a raster or vector image the image must contain geospatial information. The geospatial information relates the image pixel

9 13 coordinate values to real world coordinate values (e.g., the British National Grid). The geospatial information enables the software to display the images with the correct scaling and orientation when viewed in conjunction with other GIS data, such as a DEM (Kardoulas, 1996). There are different levels of geometric correction of remotely sensed imagery: The first level is Registration. In this stage the alignment of one image to another image of the same area. The next step is Rectification. This means alignment of image to a map so that the image is plan metric, just like the map. Also know as geo-referencing. The next level is Geo-coding. A special case of rectification that includes scaling to a uniform standard pixel GIS. The use of standard pixel sizes and coordinates permits convenient layering of images from different sensors and maps into a GIS. The final level is ortho rectification. This level focus correction of the image, pixel by pixel for topographic distortion. The result is that every pixel appears to be viewing the earth from directly above, i.e. the image is in a strict orthographic projection (Fahim, 2009). The model required for geometric correction is then built from the positional differences between reference points and their location in the imagery. This approach does not require knowing either the source or the magnitude of the errors. It is, therefore, the more frequently applied approach (Santhosh and Renuka, 2011). Geometric correction accuracy can be known from the price Root Mean Square Error (RMSE). RMSE value should be less than equal to 1. RMSE values closer to zero the better the geometric correction. The concept of RMSE is a magnitude of difference or deviation between the coordinate transformation results with certain models of the coordinates of the control points in the field. The magnitude of this deviation should be at a certain limit (tolerance) (Antoneta and Suprayogi, 2013).

10 14 Tolerance RMSE values results count on the map image geometric correction is generally determined using the assumptions of: 0.5 XRS. For PRISM image with RS = 2 meters, the allowable tolerance is: 1 meters (Antoneta and Suprayogi, 2013). The concept of RMSE is used on the coordinate transformation has been performed, then image result of the geometric correction will be tested against several ground control points that are too reference to a projection system specific with area that has a coverage the same with image corrected (Antoneta and Suprayogi, 2013). 2.3 Fusion Techniques and Methods of Image Sharpening Multisensory In the context of fusing images, there are three abstraction levels at which the combination mechanism can take place: pixel (signal), feature, and symbol. Fusion levels can be distinguished by several attributes which are the output format, the representation of sensory information, the required degree of registration accuracy, and the methodologies used for fusion (Basaeed et al., 2010). The input of pixel level fusion is the set of images. The fusion process is a local operation where each pixel in fused image is determined by considering a single pixel value or a small region of pixels in input images. The process can be applied in spatial or transform domains. The output is another image with on increased information content. This is to say that the fused image would allow new information to be inferred that would not be possible considering each input separately (Basaeed et al., 2010). In general, the image fusion techniques can be divided into two classes: colour related techniques, and statistical or numerical methods. The first group comprises of the colour composition in the red, green, blue RGB colour space as well as more sophisticated

11 15 transformations (for example IHS). Statistical approaches use channel statistics including correlation (principal components analysis (PCA), regression), and filters (high pass), while numerical methods follow arithmetic operations such as image addition, division, and subtraction. A sophisticated and very successful numerical approach uses wavelet transform in a multi resolution environment (Pohl and van Genderen in Svab and Ostir, 2006). Pan sharp is one of method used to combine between satellite data monochrome/panchromatic (black and white) with multispectral satellite image data (colored) automatically. Method of incorporation of the data automatically between panchromatic and multispectral satellite imagery data was developed by Zang of the Department of Geodesy and Geometrics University of New Brunswick (Abdi, 2011). In the proposed method, pan-sharpening is accomplished by replacing brightness information estimated from the original lower resolution multispectral (MS) bands with values derived from the higher spatial resolution panchromatic band. This process is performed while preserving the spectral information for subsequent image classification and analysis (Hanaizumi, 2008). Use Image Sharpening tools to automatically merge a low-resolution colour image with a high-resolution greyscale image (with resampling to the high-resolution pixel size). ENVI has two image sharpening techniques, using an HSV transform, and using a colour normalization (Brovey) transform. The images must either be geo-referenced or have the same image dimensions. The RGB input bands for the sharpening should be stretched byte data or selected from an open colour display (Laben et al., 2005). 1) Fusion Techniques and Methods Transformation of HSV (Hue Saturation Value)

12 16 Use HSV sharpening to transform an RGB image to HSV colour space, replace the value band with the high-resolution image, automatically resample the hue and saturation bands to the high-resolution pixel size using a nearest neighbour, bilinear, or cubic convolution technique, and finally transform the image back to RGB colour space. The output RGB images will have the pixel size of the input high-resolution data (Laben, et al., 2005). HSV method done with transforming an image in colour space red-green-blue (RGB) become image in colour space HSV (Hue Saturation Value) by: replacing the channel value (V) with high spatial resolution image, the automatic channel-sampling channel hue (H) and Saturation (S) be the size of the image elements of the high spatial resolution by using a nearest neighbour technique, bilinear, or cubic convolution. Finally, transforming back colour image RGB to colour space. Output image RGB will have size image elements same with input highresolution image data (Laben et al., 2005). For the visual system the geometric interpretation can be show on Figure 4. R IHS G IHS B IHS = +( ) +( ) +( ) + = + + Where ä IHS =Pan Iand the fused image [,, ] T is obtained from the resized original image [R, G, B] T simply by addition operations. From the visual system, one can conclude that the intensity change has little effect on the spectral information and is easy to deal with. For the fusion of the high-resolution and multispectral remote sensing images, the goal is ensuring the spectral information and adding the detail information of high spatial resolution, therefore, the fusion is even more adequate for treatment in IHS space (Al-Wassai et al., 2011).

13 17 (d) (e) (f) Figure 4. Models of IHS Color Spaces a) the Color Cube Model (b) The Color Cylinder Model (c) The Hexcone Color Model (d) The Bi-Conic Color (Al-Wassai, et al., 2011). Some are also named HSV (hue, saturation, value) or HLS (hue, luminance, saturation). (Figure 4.) illustrates the geometric interpretation. While the complexity of the models varies, they produce similar values for hue and saturation. However, the algorithms differ in the method used in calculating the intensity component of the transformation. The hexcone transformation of IHS is referred to as HSV model which derives its name from the parameters, hue, saturation, and value, the term value instead of intensity in this system (Al-Wassai, et al., 2011). Most literature recognizes IHS as a third-order method because it employs a 3 3 matrix as its transform kernel in the RGB IHS conversion model. Many published studies show that various IHS transformations, which have some important differences in the values of the matrix refer to the formula that used and R = Red, G = Green, B = Blue I = Intensity, H = Hue, S = Saturation, V1 V2 = Cartesian components of hue and saturation (Al-Wassai, et al., 2011).

14 Manual On-Screen Digitizing Method In remote sensing land cover mapping study, accuracy assessment is important to evaluate remote sensing final product. The purpose of assessment is important to gain a warranty of classification quality and user confidence on the product (Marangoz, 2012). Evolving computer technology enabled digitizing interactively which was made in the former times on digitizing tables. The features on graphical map are traced on the screen via proper software. The result data is a compound of many operator defined layers. The topology is created and edited by the operator himself (Marangoz, 2012). 2.5 Accuracy of Image Interpretation In thematic mapping from remotely sensed data, the term accuracy is used typically to express the degree of correctness of a map or classification. A thematic map derived with a classification may be considered accurate if it provides an unbiased representation of the land cover of the region it portrays. In essence, therefore, classification accuracy is typically taken to mean the degree to which the derived image classification agrees with reality or conforms to the truth, A classification error is, thus, some discrepancy between the situation depicted on the thematic map and reality (Campbell, 1996 on the Foody, 2002). Accuracy assessment involved the derivation of accuracy metrics that were based on a comparison of the class labels in the thematic map and ground data for a set of specific locations. These site-specific approaches include measures such as the percentage of cases correctly allocated referred to as overall accuracy. A key characteristic feature of this stage is that measures of accuracy that use the information content of the confusion matrix more fully than the basic percentage of correctly allocated cases, such as the kappa coefficient of agreement, are frequently derived to express classification accuracy (Foody, 2002). The confusion matrix is

15 19 currently at the core of the accuracy assessment literature. As a simple cross-tabulation of the mapped class label against that observed in the ground or reference data for a sample of cases at specified locations, it provides an obvious foundation for accuracy assessment (Canters, 1997 ; Foody, 2002). Indeed, the confusion matrix provides the basis on which to both describe classification accuracy and characterize errors, which may help refine the classification or estimates derived from it (Foody, 2002). 1) Kappa Coefficient The kappa analysis is a discrete multivariate technique used in accuracy assessment to statistically determine if one error matrix is significantly different from another. The result of performing a kappa analysis is a KHAT statistic (actually K, an estimate of kappa), which is another measure of agreement of accuracy (Congalton, 2009). Remotely sensed classification and the reference data, positive KHAT values are expected. Characterized the possible ranges for KHAT in to three grouping: a value greater than 0.80 (i.e., >80%) represent strong agreement; a value between 0.40 and 0.80 (i,e., 40-80%) represent moderate agreement; and a value below 0.40 (I,e,. <40%) represent poor agreement (Congalton, 2009). Variance of the KHAT statistic and the Z statistic used to determine if the classification is significantly better than a random result. At the 95% confidence level, the critical value would be therefore, if the absolute value of the test Z statistic is greater than 1.96., the result is significant and you would conclude that the classification is better than random (Congalton, 2009).

16 20 2) Confusion Matrix (Pixel) The confusion matrix is calculated by comparing the location and class of each ground truth pixel with the corresponding location and class in the classification image. Each column of the confusion matrix represents a ground truth class and the values in the column correspond to the classification image s labelling of the ground truth pixels (Anonim, 2015). 3) Confusion Matrix (Percent) The Ground Truth (Percent) table shows the class distribution in percent for each ground truth class. The values are calculated by dividing the pixel counts in each ground truth column by the total number of pixels in a given ground truth class (Anonim, 2015). 4) Commission Errors of commission represent pixels that belong to another class that are labelled as belonging to the class of interest. The errors of commission are shown in the rows of the confusion matrix (Anonim, 2015). 5) Omission Errors of omission represent pixels that belong to the ground truth class but the classification technique has failed to classify them into the proper class. The errors of omission are shown in the columns of the confusion matrix (Anonim, 2015). 6) Producer Accuracy The producer accuracy is a measure indicating the probability that the classifier has labelled an image pixel into Class A given that the ground truth is Class A (Anonim, 2015). 7) User Accuracy User accuracy is a measure indicating the probability that a pixel is Class A given that the classifier has labelled the pixel into Class A (Anonim,2015).

17 Land Use Most major metropolitan areas face the growing problems of urban sprawl, loss of natural vegetation and open space, and a general decline in the extent and connectivity of wetlands and wildlife habitat. The public identifies with these problems when they see residential and commercial development replacing undeveloped land around them. Urban growth rates show no signs of slowing, especially when viewed at the global scale, since these problems can be generally attributed to increasing population. Cities have change from small, isolated population centers to large, interconnected economic, physical, and environmental features (Briassoulis, 2008). Urban growth and the concentration of people in urban areas are creating societal problems world-wide. One hundred years ago, approximately 15 percent of the world's population was living in urban areas. Today, the percentage is nearly 50 percent. In the last 200 years, world population has increased six times, stressing ecological and social systems. Over that same time period, the urban population has increased 100 times, concentrating more people on less land even as the total land devoted to urbanization expands. Yet the temporal and spatial dimensions of the Land uses that shape urbanization are little known (Anonim, 1999). Land use involves the management and modification of natural environment or wilderness into built environment such as fields, pastures, and settlements. It has also been defined as "the arrangements, activities and inputs people undertake in a certain land cover type to produce, change or maintain it" (Zhi-Yong Yin, et al., 2006). Zhi-Yong Yin, et al. (2006) reported that, land use' is a key term in the language of city planning. Commonly, political jurisdictions will undertake land use planning and regulate the use of land in an attempt to avoid land use conflicts. Land use plans are implemented through land

18 22 division and use ordinances and regulations, such as zoning regulations. Management consulting firms and Non-governmental organizations will frequently seek to influence these regulations before they are codified. In the analysis of Land use, it is first necessary to conceptualize the meaning of change to detect it in real world situations. At a very elementary level, land use and land cover change means (quantitative) changes in the area extent (increases or decreases) of a given type of land use respectively. It is important to note that, even at this level, the detection and measurement of change depends on the spatial scale; the higher the spatial level of detail, the larger the changes in the area extent of land use which can be detected and recorded (Briassoulis, 2008). Land use and land management practices have a major impact on natural resources including water, soil, nutrients, plants and animals. Land use information can be used to develop solutions for natural resource management issues such as salinity and water quality. The major effect of land use on land cover since 1750 has been deforestation of temperate regions. More recent significant effects of land use include urban sprawl, soil erosion, soil degradation, Stalinization, and desertification. Land-use changes, together with use of fossil fuels, are the major anthropogenic sources of carbon dioxide, a dominant greenhouse gas (Briassoulis, 2008).