WANGs6019501图像处理作业.docx

1、WANGs6019501图像处理作业Question 1: Select the images you want to use from the data set. Perform a fuzzy classification on selected LANDSAT8 OLI images using R. Store screenshots of your classifications (water class only) in an MS Word document. Specify and motivate your choices of the images to include,

2、of the number of classes and of the parameter m. Remember that the aim of the classification is to see changes in the surface area of the lake through time.First, I need to select the appropriate images to do the classification. The principle of my selection is that trying to select the images which

3、 have as less cloud in the lake area as possible and the water area should be as clear as possible. And the date of images should be different. Finally, I select five images to do the following work. They are lc81690602013198.img, lc81690602013182.img, lc81690602013168.img, lc81690602013150.img and

4、lc81690602013246.img, separately.The images I selectedThen I need to do the classification of these images, respectively. The first thing I need to do is determining the parameters which need for the classifications.Nclthe number of classes. For each images I try different Ncl which are 3, 4 and 5,

5、respectively. And then I choose the final Ncl as the one which make the water area most clearly after the classification. Finally, the Ncl is 5 for lc81690602013198.img, lc81690602013182.img, lc81690602013168.img, lc81690602013150.img and 4 for lc81690602013246.img.mthe parameter which represent the

6、 level of fuzziness. The range of parameter m is 1, infinite). The fuzziness will increased as m increase. If m equal to 1, the fuzzy classification will turn to a crisp classification. I try several values of m, and knowing if the m setting to too large, the iteration will not work well, the mean o

7、f classes will hard to separate. And the confidence value will be low. Then I search for some articles about it, it seems that m equal to 2 will be the empirical best value. So finally I select the m equal to 2.After the iteration of classification, I need to identify which class is water. As knowin

8、g the reflection spectrum of the water, the water has the lowest reflectance in band 4-6, So I can identify the water class by code:And the result of fuzzy classifications shows below, in these images the pixel values are the membership values of water class.lc81690602013150.imglc81690602013168.imgl

9、c81690602013182.imglc81690602013198.imglc81690602013246.imgQuestion 2: For each of the membership to water images you obtained, group the pixels belonging to the lake into a lake object. Include screenshots of the lake objects in your Word document. Explain and motivate how you made the objects.As I

10、 need to see changes in the surface area of the lake through time, its better to create the lake objects for the images. Then its convenient to see the changes by objects over time.The first thing to group the lake object is setting the threshold valuethrthe value of threshold. It determines that we

11、ather the pixel could be a part of the lake object or not by comparing its membership with threshold. I saw the histogram of membership of water class and tried several threshold values to see the results of grouping objects. Finally I choose 0.8 as my threshold to group the lake objects. When an in

12、itial pixel is defined, it will grow in 8 directions, any pixel which adjacent to it will be judged by threshold. If the membership value larger than threshold, this adjacent pixel will be included in the lake object and will grow in 8 directions again depend on itself. After the growth, I will obta

13、in many segments of lake objects, and the biggest segment is Lake Naivasha.From previous exercise I know the largest segment is Lake Naivasha, and the second largest segment is Lake Oloiden. In this assignment, I only record the areas of Lake Naivasha to see the changes through time. The result of g

14、rouping lake object shows blow:lc81690602013150lc81690602013168lc81690602013182lc81690602013198lc81690602013246Question 3: Calculate the surface area of the lake for each date. Explain how you calculated the lake surface area and show the result for all LANDSAT8 OLI images in a table and a graph.The

15、 surface area of Lake Naivasha is the sum of the number of pixels which in the largest segment of the lake object multiplied by the pixel size. The surface areas are calculated by R. The results are:The lake surface area each dataOriginal imagesSurface area of Lake Naivasha(m2)lc81690602013150125640

16、900lc81690602013168129148200lc81690602013182134766900lc81690602013198127805400lc81690602013246135562500Question 4: Evaluate your result. Did the surface area change over time? What can you say about the uncertainties of the surface area estimates and the uncertainties in the changes in surface area.

17、The uncertainties arise both in the surface area estimates and the changes in surface area. First, the uncertainties come from the processing. The parameters setting for the classification and grouping object has the influence of uncertainties. Different number of classes or different parameter m, w

18、ill lead to different membership of all classes. And also different threshold value will lead to different results of lake objects. That all will lead to uncertainties on surface area estimates and changes in surface area.Then, the uncertainties come from the environment. From the image I see clouds

19、 shadow and alga in the lake area. That will make problems when I do classification to identify water. And lead to less accuracy on classification and grouping objects. Question 5: If you compare this result with the result of the exercises on fuzzy classification and object monitoring, where you pe

20、rformed similar procedures on a series of Landsat ETM+ images, what can you say about the differences?Comparing these result with the previous exercise which I have done with the landsatETM+ images. As only two images in LandsatETM+ were taken in 2013 and the date are 20130530 and 20130615, covert t

21、hese date to Julian day are 150 and 166.So the corresponding images to previous exercise are lc81690602013150 and lc81690602013168. The fuzzy classification and object monitoring of these images:lc81690602013150 vs l20130530lc81690602013168 vs l20130615Comparing with Landsat ETM+ images imagesArealc

22、81690602013150125640900l20130530125692200lc81690602013168129148200l20130615129022200And if you see on the graph, it will more clearly:From the table and the graph I see the results from these two datasets are quite similar. But there still have some small difference. That maybe because the spectral

23、resolution of these two dataset are different, and also other reasons. For example the time of taking these pictures are not actually the same, and the parameters I setting to do classification and object monitoring are not totally the same. Question 6: If you compare the NDVI plot show with the lak

24、e area plot, can you see a relationship between them? Which relationship do you see? Did you expect to see a relationship between these plots? Why (not)?Histogram of NDVI per image per polygonPlot of mean NDVI of different dateThe change of surface area of the lake and the change of mean NDVI is rel

25、evant.From the above line charts, I see the change of mean NDVI values is almost the inverse change of lake surface area. That mean when the surface area of lake is increased, the mean NDVI value is decreased. It is the expected relationship that I want to see. Because when the lake surface area is increased, the area of water increased, the area of plants decreased, that lead to the decreasing of NDVI.