Digital Image Processing Techniques - Report Example

1) Introduction Remote sensing techniques can be used reliably to ify vegetation and to differentiate between vegetation of different kinds. This paper will attempt to analyse the relationship between image characteristics and vegetation cover as well as bio mass. Different techniques such as RVI and NDVI will be utilised. The various sensors used will be Landsat TM, Lansat MSS and NOAA AVHRR. The Landsat TM is used for thematic imaging only and can reveal only a limited amount of information such as the availability or non avialblity of vegetation. On the other hand the Landsat MSS (multi spectral surveillance) can be used to distinguish between various kinds of vegetation too. The NOAA AVHRR sensor is used to estimate the thermal emission or cooling of the surface. The presence of vegetation modifies thermal emission rates by lowering them while the absence of vegetation speeds them up. A complete treatment of the ideas presented above is outlined below to delineate the relative strengths and weaknesses of each system. 2) Part One Vegetation Analysis of Lake Nakuru This section deals with the vegetation analysis of Lake Nakuru using simple and advanced models to distinguish the patterns of vegetation. Methodology The first analysis will utilise RVI (Ratio Vegetation Index) to attempt to explain how vegetation is interpreted from an image.The RVI is a ratio between the NIR (Near Infra Red) and R (Red) from each pixel in an image. Vegetation in general tends to reflect NIR as much as possible because NIR does not contribute significantly to plant nourishment and is speculated to cause plants to overheat. (Tucker, 1979) Red is reflected far less than NIR. Water, soil and manmade features have a far more static response to both NIR and R throughout the year. (Banman, 2001) A walk through of the methodology is presented below to enhance understanding. Analytical Steps The image supplied is for Lake Nakuru and is sized at 500 rows and 640 columns and possesses 4 bands. The red, green and blue bands have been set at 4, 4 and 2 respectively in fig (1). Spectral enhancement has then been utilised to analyse the image. The ratio of NIR to R is a ratio of channel 4 to channel 3. The output sensor has been selected as Landsat TM in fig (2). Step 1 (Figure 1) Step 2 (Figure 2) The output from this process is shown below in fig (3) in comparison to the actual image fig (4). (Figure 3) (Figure 4) The image presented above is then re-coloured using a pseudo colour system with brown and green as limits. Step 3 (Figure 5) Step 4 (Figure 6) This produces the image presented below (Figure 7). (Figure 7) A complete appraisal of factors is not possible in (fig 7) so another method which is NDVI which appraises vegetation cover better. Summarised form of Steps Performed 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. Select Image Interpreter from the Main Manu Bar and then click on Spectral Enhancement option. 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. After the processing is done, use the Raster Attribute Editor to edit the Start and End colors of nakuru_rvi.img file. NDVI All other settings utilised are the same as the ones used for the RVI analysis. However, the sensor used here is the Landsat MSS because it is far more fitting. Step 5 (Figure 8) The output is shown below. (Figure 9) (Figure 10) The RVI analysed image is shown on the right in the figure (9) while the NDVI analysed image is shown on the right in the figure (10). Next the image data for Tunisia will be analysed to determine if desertification is occurring. The analysis is NDVI while the sensor employed is the NOAA AVHRR that can compensate for time effects far better than other types of sensors. (Holben, 1986) Step 6 (Figure 11) The various locations that were utilised to analyse the desertification situation were navigated using pixel references. The output is shown below for the various locations. Figure (12) Figure (13) (Figure 14) (Figure 15) (Figure 16) (Figure 17) (Figure 18) In figure 11 to 17 the desertification situation was navigated using pixel references. The analysis is NDVI while the sensor employed is the NOAA AVHRR. Weakness and Strength Weakness and Strength is describes for part 1 in this section. A simple comparison of the images presented above(fig 4) reveals that the first above tends to coalesce human structures and certain rocks with vegetation. The resulting image cannot be used to classify vegetation with reliability. Instead the second image (fig 7) produced can delineate vegetation much better as can be seen. The second image has excluded vegetation near the centre especially and around it where human made structures exist. This image may be used to delineate a few factors that control vegetation distribution but expecting a complete appraisal of factors is not possible. The NDVI method has been used next to analyse the image because it tends to produce an empirical scale between +1 and -1 which appraises vegetation cover better. NDVI is the ratio of (NIR – R) to (NIR + R). It can be clearly seen that the NDVI image fig (10) is far more detailed in terms of description of vegetation as compare to RVI image fig (9). The RVI image is less descriptive and tends to combine the various bands of vegetation while the NDVI image tends to differentiate the various bands of vegetation. Results It can be clearly seen that the cursor is moving south progressively. The graphs and the ensuing regression are shown below based on the Excel data provided. (Figure 19) (Figure 20) (Figure 21) (Figure 22) (Figure 23) (Figure 24) Discussion These graphs clearly indicate that relationship between NDVI and the green biomass as well as percentage vegetation is strong especially for the month of February where R2 lies above 0.75. However, the correlation begins to recede in June and falls much lower in September where R2 lies just below 0.45 for vegetation cover versus NDVI. Literature review indicates that NDVI and vegetation cover are strongly related (Kennedy, 1989a) (Kennedy, 1989b). However, this contention assumes another dimension when moisture and rainfall are included. (Wellens, 1997) The graphs above clearly indicate that as moisture levels change due to rainfall so does the correlation of NDVI and vegetation cover as well as green biomass. The absence of moisture tends to weaken the relationship as witnessed during the dryer months. 3) Part Two Unsupervised and Supervised Classification Methodology The Churn Farm is being analysed next in order to delineate how remote monitoring could be used to discover relevant classes without any prior information. Analytical Steps The layer colours were set to 4, 2 and 2 respectively for red, green and blue. The ensuing output is provided below which delineates between vegetative and non-vegetative land easily but little else could be differentiated. Step 1 (Figure 25) The output from this process is shown below in fig (26) in comparison to the actual image . (Figure 26) The ISODATA algorithm is being utilised to optimise the solution using 10 iterations. This will ensure that a reliable enough result is produced without compromising on computing power too much. The output colour scheme has been set at 4, 3, 2 respectively for red, green and blue. The convergence threshold has been set at 0.95 in an attempt to obtain reliable results within feasible computing limits. Step 2 (Figure 27) (Figure 28) The relevant output has been displayed above in (fig 28) and it can be clearly seen that the vegetative and non vegetative features can be distinguished as before. If the previous output is compared with this output using swiping, then clear differentiation within vegetation can be easily realised. Next the Churn Farm will be analysed through signature techniques. For this purpose the image file was opened similarly and the raster options were set at 4,3 and 2 respectively for red, green and blue. The relevant output is also displayed with the settings tab. Step 3 (Figure 29) (Figure 30) In this step next signatory polygons were added to the image in order to make it ready for a signatory analysis. The output is displayed below for reference. (Figure 31) A wheat signature was added subsequently as shown below. Step 4 (Figure 32) .Step 5 (Figure 33) Other relevant signatures for other crops were also added as shown above in figure 33. The demarcated output based on polygon signatures is shown below for reference. (Figure 34) The Churn Farm image will now be explored using various approaches to signatures such as maximum likelihood, minimum distance and parallelepiped. The outputs from each of these will be compared to determine the relative strength and weaknesses of each system. Step 6 (Figure 35) Both clear and non clear displays were generated through the settings tab below. Step 7 (Figure 36) The resulting non clear and clear outputs are displayed below for a maximum likelihood treatment. (Figure 37) (Figure 38) Similar to the above, this treatment is valid for a minimum distance signatory treatment comparison. Step 8 (Figure 39) (Figure 40) Step 9 (Figure 41) (Figure 42) The last signatory technique in use is the parallelepiped which tends to produce the output shown above. Strengths and weaknesses If the output image (figure 28) is compared with the original classification of the Churn Farm, certain things come to light. For one thing the entire wheat spectrum has not been classified properly as the centre of the image indicates. Moreover, peas (top left corner) are much the same as grass with only minor differences. However, urban build up is clearly recognisable. Differentiating trees is much harder because the thicker groves show distinct patterns while the thinner groves show a predominant grass effect. This tells us that spectral response can be variable and related but it can be distinguished with enough clarity to furnish a good differentiating analysis. Supervised Classification Extra analyst control, no harmonization of spectral and information classes and known identity area processing are some of the benefits that can be achieved by making use of supervised classifications. Besides these advantages, structure imposition of data, incomplete representation of conditions of entire image by selected training areas and no chance of recognition of classes not included in the training data are the negative sides of supervised classification. Unsupervised Classification There are numerous advantages of unsupervised classifications such as advance information of the area is not needed, identification of dissimilar classes as separate divisions and least possibility of human error. On the other hand there are few disadvantages of unsupervised classification such as the complete dependence on natural grouping in the records, matching problems with the field data like dissimilarity between information and spectral classes, weak command of the analyst over image classes and variation of spectral characteristics with the passage of time. Discussion All classification schemes fall al little short in delineating the true vegetative mixture but perhaps the best output is provided by maximum likelihood. This may occur as the maximum likelihood algorithm incorporates a large array of factors that are based on probability than distance to evaluate a signature class. In comparison, the minimum distance technique uses geometrical calculations in preference to probability and other techniques. This causes certain disfigured geometrical interpretations to come to light such as the overlapping instances in the middle of the image. On the other hand the parallelepiped configuration tends to produce a highly overlapped image as the compensation is being done through parallel piped figures. The resulting overlap disfigures the image to a very large extent so this technique is not favourable at all. A few signatures were similar during the signature collection phase. This construction would encourage the overlapping of treatments. Again, the maximum likelihood treatment would be least affected and the parallelepiped treatment would be most affected by overlaps. However, the amount of overlap increases as similar signatures are utilised. The assumptions made were largely valid for the Churn Farm situation. However, certain geometrical features on ground such as sharp cuts and corners could not be properly classified. Moreover, the presence of grass under trees in thinner groves in forcing the trees to be classified as grass. Moreover, the urban classification is also not distinct totally and tends to overlap in certain areas especially near the middle of the picture. 4. Conclusion Digital Image Processing (DIP) techniques can be successfully used to delineate vegetation cover as well as green biomass distribution. The examples presented above clearly show that vegetative cover versus non vegetative cover can easily be recognised using a NDVI classification. NDVI has better results in more moist months than in hotter and drier months. Moreover, signature classification can be utilised reliably to demarcate and discern between various classes on an image containing vegetation. The most reliable of these signature classifications is maximum likelihood. Thus the reliability of image processing techniques is sufficiently established. . Bibliography Banman, C., 2001. Remote Sensing of Vegetation and Soil. [Online] Available at: HYPERLINK "http://www.emporia.edu/earthsci/student/banman3/remoteip.htm" http://www.emporia.edu/earthsci/student/banman3/remoteip.htm [Accessed 10 August 2011]. Holben, B.N., 1986. Characteristics of Maximum-Value Composite Images from Temporal AVHRR Data. International Journal of Remote Sensing, 7(11), pp.1417-34. Kennedy, P.J., 1989a. Monitoring the vegetation of Tunisian grazing lands using the normalised difference vegetation index. Ambio, 18, pp.119-23. Kennedy, P.J., 1989b. Monitoring the phenology of Tunisian grazing lands. International Journal of Remote Sensing, 10, pp.835-45. Tucker, C.J., 1979. Red and Photographic Infrared Linear Combinations for Monitoring Vegetation. Remote Sensing of Environment, 8(2), pp.127-50. Wellens, J., 1997. Rangeland vegetation dynamics and moisture availability in Tunisia: an investigation using satellite and meterological data. Journal of Biogeography, 24, pp.845-55. Read More