Remote Sensing and Image Processing - Case Study Example

Remote Sensing and Image Processing Introduction This report presents the details of activities performed in class practical for the course of Remote Sensing and Image processing. The activities included estimation of vegetation cover using Normalized Difference Vegetation Index (NDVI) and Ratio Vegetation Index (RVI) as well as supervised and unsupervised classification of different information classes on a provided image. These activities involved the use of images captured through remote sensing satellites. The use of remote sensing allows for detection of objects in real-time while also allowing coverage of a large area within a short time. There are two types of remote sensing which are passive and active. In passive remote sensing, the Sun is used to illuminate the target and the reflected waves are used to carry out the measurement. Active remote sensing, however, offers the ability to carry out measurement at any time of day or season. Furthermore, it also allows for examination of wavelengths that are not sufficiently provided the by sun as well as control over how the target object is illuminated. However, active remote sensing requires a fairly large energy source in order to adequately illuminate the target (Smith 2010). For Task involving estimation of vegetation coverage, the data was collected through the National Oceanic and Atmospheric Administration's (NOAA) Advanced Very High Resolution Radiometer(AVHRR).In order to detect vegetation a band rationing of near Infra-Red (700-1300nm) by Visible Light (400-700nm) is used. This will not only eliminate albedo effects as well as shadows from the processed imaging but would also allow high visibility of vegetation in the image. For Task involving classification of information classes, an image of Churn Farm will be used. This data in this image was collected by a NERC ATM scanner in 1984 which was carried by an airplane. The image has 4 bands. This map outlines sites of known agricultural land-use. Each land cover type has been assigned an integer hence; possible training sites for each type of land cover can be identified. Task 1 - Vegetation Index Methodology The methodology used for this task involves calculation of Ratio Vegetation Index (RVI) as well as Normalized Difference Vegetation Index (NDVI) for Lake Nakuru Thermatic Mapper (TM) image. For TM data, band 3 and band 4 are used where band 3 measures red light while band 4 measures the near Infra-Red light. Lake Nakura is a small salt water lake situated outside the town of Nakuru. The lake is famous for the spectacle when around 1 million pink flamingos flock to it to feed on green algae found in its warm waters. Apart from flamingoes, pelicans as well as cormorants are also found in the lake area in abundance(Smith 2010). Steps Performed 1) Open nakuru.img file as a Raster Layer. In Raster options display the image with Channel 4 on Red and Green gun while blue gun should be on Channel 2. 2) Select Image Interpreter from the Main Manu Bar and then click on Spectral Enhancement option. 3) Use the Indices options from the Spectral Enhancement menu to calculate the indices. Select nukuru.img as Input and nakuru_rvi.img as the Output file. Make sure that the Sensor in the Output Option is set to Landsat TM. Also, use the Select Function box to set the IR/R option which would allow the output image to be a direct ration of Channel4/Channel3. Press OK to start processing. 4)After the processing is done, use the Raster Attribute Editor to edit the Start and End colors of nakuru_rvi.img file. Calculating Normalized Difference Vegetation Index (NDVI) 1) To calculate NDVI, open the image nakuru.img as a Raster Layer. Set Red and Green guns to Channel 4 while Blue gun should be set to Channel 2. 2) Open the Image Interpreter dialog box from the main menu and then select Spectral Enhancement option. 3) Click on Indices option to calculate the indices. Set nakuru.img as Input and nakuru_ndvi.img as the Output file. Make sure that this time the Sensor in Output option set to NDVI. Set the IR/R option form the Select Function box. Click OK to start processing. 4) After the processing is done, use the Raster Attribute Editor to edit the Start and End colors of the nakuru_ndvi.img file. Once the task has been completed compare the nakuru.img, nakuru_rvi and nakuru_ndvi.img file with each other. Advantages of using Ratio Vegetation Index (RVI) and Normalized Difference Vegetation Index (NDVI) The use of both RVI and NDVI for estimation of vegetation cover offers advantages as well as disadvantages. Since RVI is calculated through spectral ratio, it offers advantages such as not being affected by scene illumination. Furthermore, with spectral ratios, one can also discriminate even the minute spectral differences. The disadvantages of RVI include the fact that it accentuates noise which must be removed before rationing. On the other hand, NDVI is preferred because it allows for compensation of changing illumination, slope as well as aspect. It produces more accurate results as it does 2-band calculations. However, NDVI is sensitive to incident solar radiation, the type of sensor used to collect the data, the atmospheric conditions as well as off-nadir viewing. Results Achieved The comparison of the two images reveals that NDVI produces a much brighter and accurate picture of the vegetation in the original image. Figure 1, 2 and 3 show the original image, RVI processed and NDVI processed image respectively. Figure 1 Original Image - Nakuru Lake (Provided) Figure 2 Image Produced by RVI processing Figure 3 Image Produced by NDVI Processing Discussion In this task RVI and NDVI indices were calculate for an image of Nakuru Lake. From the above images, it can be concluded that NDVI image has a much more accurate description of vegetation indices then the RVI image. This is because NDVI uses a much accurate technique (normalized 2-band) for calculation of indices which allows for compensation of variable look angle, illumination and scales between known limits. Task 2 - Unsupervised and Supervised Classification Methodology The methodology for this task involved the use of both unsupervised and supervised methods for image classifications. The unsupervised method used is "K-means", in which initial means are selected for different types of information classes present in the image. These initial values are then used to establish boundaries between parts having different information class. The supervised methods used in this involve "Maximum Likelihood", "Minimum Distance to Means" and "Parallelpiped". In these methods, information class samples are used to train the software which uses this to perform classification on provided images (Smith 2010). Steps Performed Unsupervised Classification 1) Open the provided image file as a Raster Layer and set the Layers to Colors 1(Red), 2(Green) and 3(Blue). This file is shown in Figure 4. Figure 4 Original Image - churn_all.img 2) Open up the Signature Editor from the Classification Menu of the ERDAS software. 3) Use the AOI Tools to collect signature for different information classes. a) Use the Polygon Tool from the AOI Toolbox to create a polygon on a wheat field. b) Add this AOI to the Signature Editor window by clicking on the Add AOI icon. c) Change the Signature Name to Wheat Field_1. d) Change the color of the Signature to Blue. e) Repeat steps a,b,c and d on multiple wheat fields to create variance in the data of the information class. f) Merge all signature of the Wheat information class by selecting the signature merge button. Delete all other classes except this one and rename it to "Wheat". 4) Repeat step 3 for all of the remaining classes in the image. 5) Using the created signatures, perform K-means classification on the image. 6) Use the Histogram tool to evaluate the signatures created using the above method. Supervised Classification The supervised classification methods can be used through the Supervised Classification dialog box of the ERDAS software. For Maximum Likelihood Classifier 1) Open the Supervised classification dialog box and set Input Raster File to churn_all.img, Input Signature File to churn.sig and Output File to churn_max.img. 2) Set Non-Paratmetric Rule to None. 3) Set Parametric Rule to Maximum Likelihood. 4) Click OK to run the classifier 5) After the classification is done, open the churn_max.img file and select Pseudo Color from the Raster Option tab. 6) Compare this file with the churn_ref.img file which shows how the classification of the image should look like. For Minimum Distance to Mean 1) Repeat the steps given for Maximum Likelihood Classifier but set the Output File to churn_mind.img and Parametric Rule to Minimum Distance. 2) Run the classifier and compare with the churn_ref.img file. For Parallelpiped 1) Repeat the steps given for Maximum Likelihood Classifier but set the Output File to churn_piped.img and Parametric Rule to Parallelpiped. Also, set both Overlap Rule and Unclassified Rule to Unclassified. 2) Run the classifier and compare with the churn_ref.img file. Strength and Weaknesses of Unsupervised Classification The use of an unsupervised classification method offers several advantages. These include no requirement of prior information of the region, minimum chances of human error as well as recognition of different classes as distinct units. However, since unsupervised classification methods rely on natural grouping in the data, problems exist in matching these grouping with the field data. This include difference between spectral and information classes, poor control of the analyst over image classes as well as change of spectral properties over time. Strength and Weaknesses of Supervised Classification The use of supervised classification methods offer advantages such as more control for analyst, no need for matching spectral classes with information classes as well processing of areas with known identity only. However, the use of these methods imposes structure on the data. Furthermore, the selected training areas may not truly represent the conditions on whole of the image while it is also not possible to identify those classes not present in the training data. Results Achieved Figure 5 shows the output image of unsupervised classification while figure 5, 6, and 7 show the histograms created for different signatures. Figure 5 Unsupervised Classification Figure 6 Histogram for Lucerne Figure 7 Histogram for Barley Figure 8 Histogram for Wheat Figure 8, 9 and 10 show the results achieved from three types of supervised classification performed in Task 2 which are Maximum Likelihood, Minimum Distance to Mean and Parallelpiped. Discussion In this task classification of various information classes was performed on the image of Churn Farm. Both unsupervised and supervised methods of classifications were used. For unsupervised classification, no prior information of the region being worked on is required. The unsupervised method also allows for distinct recognition of different information classes. However, it offers poor control over the image to the analyst. The supervised methods offer more control over the image to the analyst, however, the use of training data restricts the number of distinct classification that can be made. By comparing the three images produced by different methods of supervised classification, it can be seen that maximum likelihood classifier produces a result that is slightly more identical to the reference image then the other two methods. Conclusion The use of remote sensing allows for accurate observation of objects in real-time while being wireless. It also facilitates the observation of bands not sufficiently provided by the sun. In this report, the details of two tasks were presented. In the first task, the vegetation coverage of Nukura Lake was estimated using vegetation indices. This task was performed using Ratio Vegetation Index (RVI) and Normalized Difference Vegetation Index (NDVI). It was observed from the comparison of the results that NDVI produced a more accurate result then RVI. In the second task, the classification of information classes present in the image of Churn Farm was performed using both unsupervised and supervised methods. The unsupervised method involved creation of signatures from areas belonging to different information classes. The supervised methods involved training of the software using data and then using the software for classification. By comparing the results of the three supervised classification methods used in Task 2, it was observed that the maximum likelihood classifier method produced a comparatively more accurate result then the other two methods. References Smith M. 2010. Lecture Notes. Remote Sensing and Image Processing. Kingston University. Read More