Swarm Intelligence for Mixed Pixel Resolution

Transcription

1 Swarm Intelligence for Mixed Pixel Resolution V.K.PANCHAL *, NITISH GUPTA + * Defence Terrain Research Laboratory, Defence Research and Development Organisation + Guru Gobind Singh Indraprastha University New Delhi INDIA *vkpans@gmail.com, + ntshgpt@gmail.com Abstract: - Mixed pixels are usually the biggest reason for lowered success in classification accuracy. Aiming at the characteristics of remote sensing image classification, the mixed pixel problem is one of the main factors that affect the improvement of classification precision in remote sensing classification. How to decompose the mixed pixels precisely and effectively for multispectral/hyper spectral remote sensing images is a critical issue for the quantitative remote sensing research. As Remote sensing data is widely used for the classification of types of land cover such as vegetation, water body etc thus Conflicts are one of the most characteristic attributes in satellite remote sensing multilayer imagery. Conflict occurs in tagging class label to mixed pixels that encompass spectral response of different land cover on the ground element. In this paper we attempted to present a new approach for resolving the mixed pixels using Biogeography based optimization. The paper deals with the idea of tagging the mixed pixel to a particular class by finding the best suitable class for it using the concept of immigration and emigration. Key-Words: - Geographical Information systems, Decision Support, Process Modeling, Artificial Intelligence, Mixed Pixel, Biogeography based optimization, Remote Sensing. 1 Introduction Remote sensing [1] data can be put to use in classifying the features in an image into distinct categories. The categorized images can then be used in different ways a farmer may use thematic maps to monitor the health of his crops without going out to the field. A geologist may use the images to study the types of minerals or rock structure found in a certain area. A biologist may want to study the variety of plants in a certain location. Remote sensing with multi-spectral satellite imagery works on the concept that different features of the land cover reflect electromagnetic radiation over a range of wavelengths in their own characteristic way according to their chemical composition and physical state. The problem arises when the terrain features do not correspond to the pure object or feature and may not have a prior spectral signature library. Pixels which are in the interface region of two classes e.g. vegetation & Water body, cannot be clubbed together to a single category, say vegetation, because these pixels do not follow any one particular class s spectral signature. These types of pixels are generally known as mixed pixels. Assigning a class tag to the mixed pixels is cause of conflict [2] in the expert s mind and it is the focus point of the study. Mixed pixel resolution of remote sensing images is one of valid assistance means for improving quality of feature extraction from the images. A new algorithm implement mixed pixel resolution of remote sensing images and takes full advantage of neighbourhood information, makes the resolution result more human, and presents better robustness to environment. The paper is organized into four sections. The section following the introduction illustrates the problem formulation which deals with the concepts of Biogeography based optimization and mixed pixels origin and then methodology. The third section describes problem solution in which the new algorithm for mixed pixel resolution is being proposed. The fourth section show experiment and summarizes the important findings for this paper, is to observe this mixed pixel scenario and attempt to resolve the mixed pixel and tag them with a specific class. 2 Problem Formulation 2.1 Biogeography Based Optimization Biogeography is the study of the distribution of biodiversity spatially and temporally. Over areal ecological changes, it is also tied to the concepts of ISBN:

2 species. Immigration is the introduction of new people into a habitat or population. It is a biological concept and is important in population ecology,a habitat (which is Latin for "it inhabits") is an ecological or environmental area that is inhabited by a particular species of animal, plant or other type of organism. It is the natural environment in which an organism lives, or the physical environment that surrounds (influences and is utilized by) a species population. Biogeography unfolds the geographical distribution of biological organisms. The mindset of the engineer is that we can learn from nature. This motivates the application of biogeography to optimization problems. The science of biogeography can be traced to the Work of nineteenth century naturalists such as Alfred Wallace [3] and Charles Darwin [4]. In the early 1960s, Robert MacArthur and Edward Wilson began working together on mathematical models of biogeography, their work culminating with the classic 1967 publication The Theory of Island Biogeography [5]. Their interest was primarily focused on the distribution of species among neighbouring islands. The application of biogeography to engineering is similar to what has occurred in the past few decades with genetic algorithms (GAs), neural networks, fuzzy logic, particle swarm optimization (PSO), and other areas of computer intelligence. The term island here is used descriptively rather than literally. That is, an island is any habitat that is geographically isolated from other habitats. We therefore use the more generic term habitat in this paper (rather than island ) [6]. Geographical areas that are well suited as residences for biological species are said to have a high habitat suitability index (HSI) [7]. Features that correlate with HSI include such factors as rainfall, diversity of vegetation, diversity of topographic features, land area, and temperature. The variables that characterize habitability are called suitability index Variables (SIVs). SIVs can be considered the independent variables of the habitat, and HSI can be considered the dependent variable. Habitats with a high HSI have many species that emigrate to nearby habitats, simply by virtue of the large number of species that they host. Habitats with a high HSI have a low species immigration rate because they are already nearly saturated with species. Habitats with a low HSI have a high species immigration rate because of their sparse populations. Biogeography is nature s way of distributing species, and is analogous to general problem solutions. Suppose that we are presented with a problem and some candidate solutions. A good solution is analogous to an island with a high HSI, and a poor solution represents an island with a low HSI. High HSI solutions resist change more than low HSI solutions. We call this new approach to problem solving biogeography-based optimization (BBO)[8]. 2.2 Mixed Pixel Each pixel [9] of remote sensing image contains the information from multifarious ground objects due to the difference from the resolution of remote sensing image, called Mixed pixel. The basic building block of the study is the concept of mixed pixel, and what way the concept representing vital knowledge is elaborated using rough sets. A fundamental assumption that is commonly made in remote sensing is that each pixel in the image represents the area on the Earth s surface that contains a single class. This is often not the case, with the mixed pixels containing presence of more than one class Origin of the mixed pixels may be attributed to the following: 1. Mixed caused by the presence of small, subpixel targets within the area it represents (Fig 1.i). 2. Mixing as a result of the pixel straddling the boundary of discrete thematic classes (Fig 1.ii). 3. Mixing due to gradual transition observed between continuous thematic classes (Fig 1.iii). 4. Mixing problem due to the contribution of a target (black spot) outside the area represented by a pure but influenced by its point spread function.(fig 1.iv) Also the pixel appears superficially be pure. The portion of mixed pixels in an image is often large in remote sensing studies. The exact proportion of mixed pixels in an image is an interactive function of properties of sensor (e.g. the class composition, spatial arrangement) and so the mixed pixel problem is a contextual issue as well. To tag the mixed pixel is really a matter of concern and requires strong analysis of terrain features before tagging it to a particular class. ISBN:

3 Figure 1. Some common origins of mixed pixels (adapted from Foody) Figure 2. False colour satellite Image of Alwar area 2.3 Methodology The image is classified into pure pixels corresponding to different land features like water, vegetation, rocky, barren and urban. Hence beforehand we had the data set of mixed pixels and pure pixels.our objective is to resolve the mixed pixel and tag it to the specific class to which it revolves on the basis of the seven band digital data (DN values). We have taken a multi-spectral multiresolution & multi-sensor image of Alwar area in Rajasthan. The area is selected since it contains a good variety of land use features like urban, water body, rocky, vegetation & barren areas (Fig.2). The multi-spectral geo-referenced image-set consists of satellite images of dimension 472 X 546 pixels. The 4-spectral bands are in the visible bands namely: red, green, near- infrared (NIR) and middle infra-red (MIR) from LISS-III sensor of Indian Remote Sensing sat satellite Resourcesat-1.Also,two SAR images namely: low incidence S1 beam (RS1) and other is High incidence S7 beam ,(RS2) of the same area taken from Canadian Radarsat-1 satellite. The seventh band is digital elevation model (DEM) of the area. The ground is resolution of the image from LISS-III and Radarsat- 1 image is 23.5m and 10m respectively. The DEM dataset is also generated from SAR interferometry using RS1 and RS2 and have 25-meter resolution. We are having spectral signatures from seven bands namely Red, Green, NIR, MIR, RS1, RS2 and DEM of the mixed classes (water-vegetation) (Table.1) and training data set of pure pixels (water) (Table.2) and training data set of pure pixels(vegetation) (Table 3) provided by the expert. The table has, therefore 7 attributes i.e. Red, Green, Near-Infrared, Middle Infrared, Radarsat-1, Radarsat-2, DEM. Table 1. Data set of water-vegetation mixed pixel. Table 2.Data set of pure water pixels. Table 3.Data set of pure Vegetation pixels. 3 Problem Solution 3.1 Swarm Based Mixed Pixel Resolution Algorithm Definition Definition 1: A habitat is a vector of integers that represents a feasible solution to some problem (Here habitat is the data set of pure pixels being classified). Definition 2: A suitability index variable (SIV) is an integer that is allowed in a habitat. is the set of all integers that are allowed in a habitat. (SIV s are the seven band DN values i.e. Red, Green, Near- ISBN:

4 Infrared, Middle Infrared,Radarsat-1, Radarsat-2, DEM). Definition 3: A habitat suitability index HSI: is a measure of the goodness of the solution that is represented by the habitat.(his is the mean of standard deviation of each band represented as column in table). Definition 4: An ecosystem is a group of habitats. The size of an ecosystem is constant ( A set mixed pixels with each mixed pixel being considered as habitat). Definition 5: Immigration rate is a monotonically Non increasing function of HSI. is proportional to the likelihood that SIVs from neighbouring habitats will migrate into habitat. Definition 6: Emigration rate is a monotonically Non-decreasing function of HSI. is proportional to the likelihood that SIVs from habitat will migrate into neighbouring habitats Algorithm Frame work of swarm intelligence based mixed pixel resolution algorithm. Input- Dataset of Pure and mixed pixels of land features. Output- All mixed pixels are classified. [Begin] Initialization Condition= no of different sets of mixed pixel /* Data set of Water-vegetation, urban-rocky, urban- Vegetation are taken*/ Reading training data set of all pure pixels. Reading training data set of all mixed pixels. While (condition! = 0) /* taking one unique set of mixed pixel for each Iteration for condition*/ Loop= no of pixels in mixed training data set taken. Original_HSI_1= mean (standard deviation of each Band DN values of pure pixel data set of class_1 of which a mixed pixel corresponds) Original_HSI_2= mean (standard deviation of each band DN values of pure pixel data set of class_2 of which a mixed pixel corresponds) for(i=0; i< Loop; i++) Add pixel [Loop] from mixed pixel to tables of both the pure pixel of which the Mixed pixel corresponds./* Emigration*/ Calculate New_HSI_1, New_HSI_2 /* after Addition*/ Deviation_1=Original_HSI_1- New_HSI_1; Deviation_2=Original_HSI_2- New_HSI_2; If (Deviation_1<Deviation_2) Classify Pixel [Loop] as Class 1; /*Immigration*/ Else Classify Pixel [loop] as class_2; /*Immigration*/ [End] 4 Conclusion 4.1 Experiment Alwar district is situated in the northeast of Rajasthan between 27o4' and 28o4' north Latitudes and 76o7' and 77o13' east Longitude. Its greatest length from south to north is about 137 K.M. and greatest breadth from east to west about 110 K.M. The image of the district with dimensions 472 X 546 is taken. On the basis of spectral behaviour observed by the sensors of satellite, The image is being classified into set of pure pixels of various land features and different pairs of mixed pixels depending upon the land featured encountered in image. Now the swarm based mixed pixel resolution algorithm is applied on training data set of pure pixels and mixed pixels. On the basis of DN values of bands of mixed pixels, these are tagged under unique class which shows the less deviation after adding that mixed pixel in it. The Figure 3 and Figure 4 shows the deviation of water-vegetation mixed pixels (9 pixels of distinguished values from data set of mixed pixel are taken) from pure water and pure vegetation respectively. Hence which ever shows the less deviation from the pure pixel class, that pixel is classified and iteration is carried out till all the ISBN:

5 mixed pixels are being tagged (Table 4). Similarly the case of urban-rocky mixed pixels is taken and graph is plotted and all the mixed pixel cases are solved using this algorithm. Judging from the experimental results, the mixed pixel is resolved and hence the whole image is classified totally. Due to constraint of space, only the watervegetation mixed pixels resolution is shown (9 pixels). (assumed reflectance values being considered on the basis of common platform of each land feature like vegetation is being observed as on the platform of photo-synthesis). 1. The mixed pixels are resolved using the BBO technique considering all the permissible values of seven bands, which are used for observing the land feature. 2. The new algorithm is based on the local search of individuals in swarm so the method utilizes adequately neighbouring information of mixed pixels in range. 3. Spectral bands have definite contribution towards the mixed pixel resolution. Figure 3. Mixed Pixel in Pure water and their respective deviation Figure 4. Mixed Pixel in pure vegetation and their respective deviation Table 4. Mixed pixels are being classified References: [1] Ralph W.Kiefer, Thomes M. Lillesand, Principles of Remote Sensing,2006. [2] V.K.Panchal, Sonakshi Gupta, Nitish Gupta, Mandira Monga Eliciting conflicts in expert s decision for land use classification, at ICEEA(IEEE), [3] A. Wallace, The Geographical Distribution of Animals (Two Volumes).Boston, MA: Adamant Media Corporation, [4] C. Darwin, The Origin of Species. New York: Gramercy, [5] R. MacArthur and E. Wilson, The Theory of Biogeography. Princeton, NJ: Princeton Univ. Press, [6] I. Hanski and M. Gilpin, Meta population Biology. New York: Academic,1997. [7] T. Wesche, G. Goertler, and W. Hubert, Modified habitat suitability index model for brown trout in southeastern Wyoming, North Amer.J. Fisheries Manage., vol. 7, pp , [8] Dan Simon, Biogeography based optimization. : IEEE transactions on evolutionary computation, vol. 12, no. 6, December 2008 [9] P. Fisher, The Pixel: a Snare or a Delusion, International Journal of Remote Sensing, Vol.18: pp , Conclusion Swarm based mixed pixel resolution algorithm is put forward based on the mechanism of Biogeography based optimization in view that every pixel of particular land feature comes under the a range of DN (reflected) values or they are related on the basis of closeness and similarity between them ISBN: