International Journal of Innovations in Scientific and Engineering Research (IJISER)
ISSN: 2347-971X (online) ISSN: 2347-9728(print)
AN INTELLIGENT SYSTEM FOR FORESTFIREDETECTION USING ANISOTROPICDIFFUSION AND FUZZY C-MEANS ALGORITHM 1
K.Nanthini, 2D.Prabhakaran, 3C.Ramkumar Research Scholar, Research and Development, Coimbatore, India 2,3 Assistant Professor, Department of CSE, Sasurie College of Engineering,Tirupur, India 1 nanthinidec30@gmail.com, 2prabha619@gmail.com, 3c.ramkumar84@gmail.com 1
Abstract: The extreme mounting in the volume of spatial data puts behind its growth to the technologies that assist in spatial data acquisition. Spatial data mining has arised as voluminous research area. Fire plays an important role in a majority of the forest environments. Forest fires possess serious threats that result in deterioration of economy and environment apart from causing danger to human lives. This paper provides measure to detect forest fires from the whole forest spatial data. The proposed approaches make use of spatial data mining, image processing techniques for the effective detection of forest fires. Keywords: Color space, anisotropic diffusion, Fuzzy C-Means, Gaussian Kernel Density. forest and fires have been witnessed in the past. [10, 12] 1. INTRODUCTION Fire is a dynamic combustion process, instability and They have played a remarkable role in determining continuous development, so fire flame always keeps landscape structure, pattern and eventually the species flickering and beating, and its visual characteristics are composition of ecosystems. Intermittently forest firehas mainly embodied in the edge of the flame image. This compelled the evacuation of vulnerable communities phenomenon such as Edge jitter, generating sharp besides heavy loss amounting to millions of dollars. corners and the irregular beating of the numbers of the 21.23% of the Indian land has been classified into forest sharp corners, will both lead to the overall flame height, area as per forest survey report. This paper intends to area and other corresponding changes, whose change detect forest fires from the forest spatial data. law not only have direct contact with the flame flickering frequency, but also have a great difference 2. PREVIOUS WORK with interference source. According to statistical In earlier work, an efficient forest fire detection system analysis, in the sequence of frames in the image, 70% used fuzzy logic. The primary intention of this work is of the energy of the fire flame images concentrates on to extract useful information from spatial data and low frequency range, approximately from 1 to 8Hz. The employ them for locating the region exposed to forest energy distribution can reflect the characteristics of the fire using image processing and remote sensing flame beating, and this article mainly analyzes static techniques. This paper presents an intelligent system and dynamic features of the flame, such as the flame that is capable of detecting forest fires. Here the radial flickering frequency and the contour of the flame. The basics function neural network is employed in the rapid progress in scientific data collection has led to design of the presented intelligent system[2]. enormous and continuously ascending amount of data making it infeasible to be interpreted manually. 3. PROPOSED WORK Therefore, the development of new techniques and tools In this work, we have presented an efficient system to support of humans, assisting in the conversion of data detect fires using spatial data collected from forest. into useful knowledge, Data mining is the vital step in Image processing and Fuzzy C-Means clustering KDD, knowledge discovery in data bases can be techniques were utilized in the design of the present defined as “The non-trivial extraction of implicit, system. Anisotropic diffusion is employed for previously unknown, and potentially useful information segmentation process and Fuzzy C-means is employed from databases .Fires have been a source of trouble for for fire detection process. The images are converted a long time now. Fires have remarkable influence in the from RGB into L*a*b color space and segmentation is environmental and economic utilities of the wild and performed. The Fuzzy C-Means algorithm is
www.ijiser.com
238
Vol 2 Issue 12 DEC 2015/101
International Journal of Innovations in Scientific and Engineering Research (IJISER)
ISSN: 2347-971X (online) ISSN: 2347-9728(print)
implemented in the proposed work which is successfully useful to detect the forest fires in the spatial data. [5]
smoothing the image interiors to emphasize boundaries for segmentation and by eliminating the spurious detail besides eradicating noise from images efficiently. However there is a property that the low frequency objects that are complex to be dissolved without overprocessing the image are left out by the relaxation process that implement anisotropic diffusion [11,23,34]. Anisotropic diffusion in image processing discreteness the family of continuous partial differential equations, which combines the physical process of diffusion and the Laplacian, Provided that there are no shrinks of sources that exist the following equation. Formulates the above mentioned process for any dimension [23]
4. METHODOLOGY The concepts utilized in the presented intelligent system for effective detection of forest fire such as color space, anisotropic diffusion segmentation and Fuzzy C-means clustering are detailed in this section. figure 1 shows the present system in forest fire detection. Color Separation. Anisotropic Diffusion Segmentation. Gaussian Kernel Density Estimation. 4.1 Color Separation Convert image from RGB color space into L*a*b* color space. The color space enables the quality of the visual diffusion. The L*a*b* color space is derived from the CIEXYZ tristimulus values the L*a*b* space consists of a luminosity layer denoted as ‘L*’ chromaticity layer denoted as ‘a*’ indicates where color range along the red-green axis and chromaticity layer denoted as ‘b*’ indicates where the color falls range the blue-yellow axis. The color information is available in the ‘a*’ and ‘b*’ layers. Measuring the difference between two colors is done using the Euclidean distance metric by converting the image to L*a*b*
4.3 Feature selection Using Gaussian Kernel Density Estimation This research work aims to find the relevant features from the large dimensional data set. The feature selection methods are used to preserve the primary information and meaning of features in the selected subset. The proposed work which uses the efficient Gaussian Kernel Density Estimation to select the relevant feature from the large data set. The purpose of this method is to remove noisy and redundant features from the original feature subspace. This is a filter approach method, which chooses more informative features according to their probability density function (pdf) using the non-parametric of Gaussian Kernel Density Estimation relations. This method does not use the class label of each pattern (document), thus being suited for unsupervised learning problems. The main idea of the this method is firstly approximating the Probability Density function using Gaussian Kernel Density Estimation of each feature independently in an unsupervised manner and then removing those features which their probability density functions have higher covering areas with the pdfs of other features which are termed as redundant features. The general form of the non-parametric estimation of probability density function using the Kernel Density Estimation,
4.2 Anisotropic Different Segmentation A low-level image processing task image segmentation is used for splitting an image into identical regions. Using the segmentation results the regions of interest and objects in the scene that is very advantageous to the consequent Image Analysis can be identified. Color image segmentation is more monotonousthan the grey image segmentation owing to the reason that the inherent multi features contain nonlinear relation individually and encompass inter-feature dependency between R, G and B.[3] In our system we have used anisotropic diffusion approach for the segmentation of images. The segmentation is performed on the L*a*b color space converted image. Literature studies have discussed several models of linear and nonlinear diffusion in order to accomplish image smoothing and segmentation. The nonlinear anisotropic diffusion has been employed by many researchers in their works. The anisotropic diffusion improves the response of edge detection algorithms by
Where k(x) is the kernel function. www.ijiser.com
239
Vol 2 Issue 12 DEC 2015/101
International Journal of Innovations in Scientific and Engineering Research (IJISER)
ISSN: 2347-971X (online) ISSN: 2347-9728(print)
Gaussian kernel
publicly available of fire was used in the formation of estimation.
The Gaussian Kernel density estimation for pdfs for unsupervised feature selection which is a filter approach includes following steps which describes the whole process of the feature selection method. . STEPS: Step 1: Load the dataset from the database. Step 2: Scale the each feature in data set within the range of 0 to 1 interval because the range of various features may be different. Step 3: Estimate the probability density function for every feature using the Kernel density estimation formula
Using the Gaussian Kernel
(X, X Step 4: Having estimated the probability density functions for each feature, calculate the difference in similarity between each features. Step 5: Two features are said to be similar, if the Mean Square Error (MSE) of their pdfs is less than a user specified threshold. Step 6: Similar features contain the same information because their pdfs are sufficiently similar. Hence one similar feature can be removed without ample loss of information
Figure2: a) Scene Image b) Diffused Segmentation Image c) Forest Image d) Cluster Image e) Forest Fire Detection Image Step 7: Among similar features in the whole feature space, features which are similar to other feature are removed. Step 8: Therefore, by removing this feature the loss of information will be at least.
6. CONCLUSION Forest fires cause noteworthy environmental dimidiation while menacing human lives. The explosive growth of spatial data and wide spread use of spatial databases emphasize the need for the automated discovery of spatial knowledge fire plays a vital role in a majority of the forest eco-systems. In this paper, we have presented an intelligent system for effective detection of forest fire using spatial data. The proposed
5. EXPERIMENTAL RESULTS This section contains the result of our experiments. The proposed system is implemented in MATLAB. The
www.ijiser.com
240
Vol 2 Issue 12 DEC 2015/101
International Journal of Innovations in Scientific and Engineering Research (IJISER)
ISSN: 2347-971X (online) ISSN: 2347-9728(print)
system made use of image processing and Feature selection Using Gaussian Kernel Density Estimation. The images utilized by the presented system are used for the detection of forest fires. REFERENCES [1] Li, Deren, Song Xia, Haigang Sui and Xiaodong Zhang. "Change Detection Based on Spatial Data Mining", Wuhan University, white paper, pages-8(2007) [2] Kargupta, Hillol, Anupam Joshi, Krishnamoorthy Sivakumar and Yelena Yesha. Data mining: next generation challenges and future directions. Aaai Press, 2004. [3] Ester Martin, Hans-Peter Kriegel and Jörg Sander, "Algorithms and applications for spatial data mining", Geographic Data Mining and Knowledge Discovery 5, no. 6 (2001) [4] Park Jooyoung and Irwin W. Sandberg, "Universal approximation using radial-basis-function networks", Neural computation 3, no. 2 (1991): 246-257. [5] Frawley, William J, Gregory Piatetsky-Shapiro and Christopher J. Matheus. "Knowledge discovery in databases: An overview", AI magazine 13, no. 3 (1992): 57. [6] Tadesse, Tsegaye, Jesslyn F. Brown and Michael J. Hayes. "A new approach for predicting drought-related vegetation stress: Integrating satellite, climate, and biophysical data over the US central plains", ISPRS Journal of Photogrammetry and Remote Sensing 59, no. 4 (2005): 244-253. [7] Julisch, Klaus. "Data mining for intrusion detection", In Applications of data mining in computer security, PP. 33-62, Springer US, 2002 [8] Tang, Hong and Simon McDonald, "Integrating GIS and spatial data mining technique for target marketing of university courses", In ISPRS Commission IV, Symposium. PP. 9-12, 2002 [9] Mukhlash, Imam and Benhard Sitohang, "Spatial data preprocessing for mining spatial association rule with conventional association mining algorithms", In Proceedings of the International Conference on Electrical Engineering and Informatics. Institut Teknologi Bandung, 2007. [10] Ding, Qin, William Perrizo, Qiang Ding and Amalendu Roy, "On Mining Satellite and other Remotely Sensed Images", In DMKD. 2001. [11] Gonzalez, José Ramon, Marc Palahi, Antoni Trasobares, and Timo Pukkala. "A fire probability model for forest stands in Catalonia (north-east Spain)."Annals of Forest Science 63, no. 2 (2006): 169-176. [12] Cortez, Paulo, and Aníbal de Jesus Raimundo Morais. "A data mining approach to predict forest fires using meteorological data", (2007).
www.ijiser.com
241
Vol 2 Issue 12 DEC 2015/101