Supervised Classification Of Multispectral Image & Accuracy Assessment

DOI : 10.17577/IJERTV1IS9252

Download Full-Text PDF Cite this Publication

Text Only Version

Supervised Classification Of Multispectral Image & Accuracy Assessment

Supervised Classification Of Multispectral Image


Accuracy Assessment

1 B. Naga Jyothi, 2 K. S. R. Radhika , 3 Dr I. V. Murali Krishna

1 Assoc. Prof., Dept. of ECE, DMS SVHCE, Machilipatnam, A.P., India,

2 Assoc. Prof., Dept. of CSE, DMS SVHCE, Machilipatnam, A.P., India,

3 Director (Retd.,), IST/R&D Center, JNTU, Hyderabad, India,


  1. Multispectral image analysis in this paper is done using the 7 image files cp.tiff, cp. tiff, cp.tiff, ch4.tiff, cp. Tiff, cp.tiff, ch7.tiff of Copenhagen City, Denmark obtained from LANDSAT 5 TM which provides 7 bands of data. [eduspace Home 1]. The images used here are 512*512 pixel size. The LANDSAT system gives images with a high resolution of 30 m * 30m. The 7 individual image files are combined to a single ab.mat file.

    1. Band 1

      0.45-0.52 m

      visible light, blue

      Band 2

      0.52-0.60 m

      visible light, green

      Band 3

      0.63-0.69 m

      visible light, red

      Band 4

      0.76-0.90 m

      near infrared

      Band 5

      1.55-1.75 m

      middle infrared

      Band 6

      10.4 -12.5 m

      thermal infrared

      Band 7

      2.08 -2.35 m

      middle infrared

      Three-band composite images are created by using three LANDSAT spectral bands to control the amount of red, green and blue in a color image. True-color composite images with combinations of bands 1, 2 and 3 approximately match the spectral range of vision for the human eye, so these images appear to be close to what would be expected in a normal color photograph.

      A false-color image is an artificial representation of a multispectral image.

      The specific bands used in three-band composites are often identified by giving the band numbers used for red, green, and blue in a specific order. Thus, an image using band 7 for red, band 4 for green, and band 2 for blue would be designated (7,4,2). As can be seen in the simulation results [Fig: 1], the shortwave composite of the bands 7, 4, 2 shows the forest in red and the cultivated

      areas in different shades of red and pink. Roads are visible as straight light-colored lines. The river is in dark blue color.


In supervised classification ,the identity of land cover types are known priorly through some means such as aerial photography, map analysis and personal experience etc. The analyst locates specific sites that represent homogeneous examples in the remotely sensed data. These areas are referred as training samples and the spectral characteristics of these areas are used to train the classification algorithm of the remainder of the image. Multivariate statistical parameters such as mean, standard deviation, covariance matrices, correlation matrices are calculated for each training site. Every pixel both within and outside the training sites is then evaluated and assigned to the class of which it has the highest likelihood of being a member.

Thematic classification of an image involves the following steps:

  • the training area be a homogenous sample of the respective class, but at the same time include the range of variability for the class.

      1. To perform a classification accuracy or error assessment , two sources of information are to be compared.

        • Pixels in the remote sensing-derived classification map and

        • Ground reference test information

          The relationship between the two sets of information is summarized in an error matrix. The error matrix is used to assess the remote sensing classification accuracy of

          k classes. The central part of the error matrix is a square array of numbers k x k in size. The columns of the matrix represent the ground reference test information and the rows correspond to the classification generated from analysis of remotely sensed data. The intersection of the rows and columns summarize the number of sample units (pixels) assigned to a particular category(class) relative to the actual category as verified in the field. The total number of samples examined is N.

          The diagonal of the error matrix summarizes those pixels that were assigned to the correct class. Every error in the remote sensing classification relative to the ground

          reference information is summarized in the off-diagonal cells of the matrix.

          Each error is both an omission from the correct category and a commission to a wrong category. The column and row totals around the margin of the matrix are used to compute errors of inclusion (commission errors) and errors of exclusion (omission errors). The outer row and column totals are used to compute producers and users accuracy.

    Kappa analysis is a discrete multivariate

    technique of use in accuracy assessment.

    K hat Coefficient of Agreement: kappa analysis yields a statistic K , which is an estimate of Kappa. It is a measure of agreement between the remote sensing derived classification map and the reference data as indicated by a) the major diagonal and b) the chance agreement, which is indicated by the row and column totals (referred to as marginals).

    k k

    N xii – ( xi+ x x+i)

    i=1 i=1

    K = ——————————-


    N2 – ( xi+ x x+i)


    where k is the number of rows in the matrix, xii is the number of observations in row i and column i, and xi+ and x+i are the marginal totals for row i and column i respectively and N is the total number of observations. Values of K > 0.8 (80%) represent strong

    agreement, values between 0.4 and 0.8 represent moderate agreement and values less than 0.4 represent poor agreement.

    1. The problem is to segment the image into meaningful classes like forest, water, urban area , roads and agricultural areas. Simulation is done using Matlab [2]. Simulation involves the sequence of following steps:

  • The 7 individual image files cp.tiff, cp.tiff, cp.tiff, ch4.tiff, cp.tiff, cp.tiff, ch7.tiff are combined to a single ab.mat file using the matlab command savefile

  • The image data can be loaded into MATLAB using the command load ab

  • It is also possible to view three bands at a time using color or pseudo color. imtool command is used showing bands (7,4,2) converted to R,G,B

  • Training areas are marked using coordinates for rectangles and training labels are shown using the command imagesc(label_im)

  • Classification tools like classify, scatterplots are used for different bands. Individual bands are separated and classification result is observed using various band combinations using the command

  • Classification result is observed using 4,6 & 2,4,6,7 band combinations.

Finally, Accuracy Assessment is carried out taking all the pixels into consideration using a matlab program by comparing the classified output array with a reference image array generated on the basis of visual spectral inspection of the input image.

  1. Fg: 1 Input Image with Band Combination [7,4,2]

    Fig: 2 Training Labels

    Forest Water Agriculture

    Roads Urban

    Fig: 3 Classification using bands 4, 6

    Water Urban Roads

    Forest Agriculture

    Fig: 4 Classification using bands 2,4,6,7

  2. ———————————————– 29782 0 135 45854

    0 40151 358 4315

    1372 3776 10935 23595

    15615 0 93 86163

    ———————————————- CLASS 1 : 63.68 36.32

    CLASS 2 : 91.40 8.60

    CLASS 3 : 94.91 5.09

    CLASS 4 : 53.88 46.12

    ———————————————- CLASS 1 : 39.31 60.69

    CLASS 2 : 89.57 10.43

    CLASS 3 : 27.56 72.44

    CLASS 4 : 84.58 15.42

    Using Bands 2,4,6,7 ERROR MATRIX:

    ——————————————— 30293 2582 719 50024

    0 41107 18 2593

    280 238 9872 18732

    16196 0 912 88578

    ————————————————- CLASS 1 : 64.77 35.23

    CLASS 2 : 93.58 6.42

    CLASS 3 : 85.69 14.31

    CLASS 4 : 55.39 44.61

    ———————————————— CLASS 1 : 36.23 63.77

    CLASS 2 : 94.03 5.97

    CLASS 3 : 33.90 66.10

    CLASS 4 : 83.81 16.19

  3. Different bands reveal information about different areas and so some combination of bands might give good classification result for one set of classes and poor for the remaining classes.

    All the classes seem to be separated quite well, when bands 2, 4, 6 and 7 are used for classification. Overall Accuracy is also found to be high for this band combination.

    Softwares like Erdas can be used to compute Error matrix, Producers accuracy (Omission Error), Users accuracy (Commission Error), Overall accuracy and Kappa coefficient . But the limitation with these softwares is that the above parameters are computed considering only few randomly selected pixels. If this few randomly selected pixels are classified

    accurately then it results in 100% overall accuracy. If on the other hand , if all the randomly selected pixels are not classified correctly then it results in 0% overall accuracy. Thus the overall accuracy is determined based only on the classification result of few randomly selected pixels.

    However in this paper a separate matlab program is developed as mentioned in the methodology and then the Error matrix, Producers accuracy (Omission Error), Users accuracy (Commission Error), Overall accuracy and Kappa coefficient are computed by considering all the pixels in the image. Thus the overall accuracy is computed based on the classification result of all the pixels in the output classified image. Obviously this method yields in much accurate output results.

  1. eduspace Home, http://Copenhagen Landsat Images and Band Combinations

  2. Centre for image Analysis, Swedish University of Agricultural Sciences,Uppsala University

  3. Schowengerdt, Robert A. (1997). Remote Sensing: Models and Methods for Image Processing (2nd ed.). New York: Academic Press. 522 p


  4. Remote Sensing of Environment (RSE) withTNTmips® TNTview®

[5]. D. LU and Q. WENG A survey of image classification methods and techniques for improvingclassification performance International Journal of Remote Sensing Vol. 28, No. 5, 823870, 10 March 2007

[6]. Marçal, A.R.S., Borges, J.S., Gomes, J.

A. and Costa, J.F.Pinto Da,. Land cover update by supervised classification of segmented ASTER images. Int. J. Remote Sensing, 26(7),pp.1347-1362., 2005

[7]. Foody,G.M.,2002 Status of Land Cover Classification Accuracy Assessment Remote Sensing of Environment,80:185-201

[8]. John R.Jensen Introductory Digital Image Processing, A remote Sensing perspective, Third Edition, PPH, 338p, 499- 500p

[9]. Erdas,2003, Erdas Field Guide,

Atlanta:Leica GeoSystems

[10]. Congalton, R.G,1991, A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data, Remote Sensing of Environment,37;35-46

[11]. Hudson,W and C.Ramm,1987,Correct Formulation of the Kappa Coefficient of Agreement, Photogrammetric engineering & Remote Sensing,53(4):421-422

[12].B.Naga Jyothi, Dr. G. R. Babu, Dr.I.V.Murali Krishna, Thematic classification of multispectral imagery, International Journal of Electronics and computer Science Engineering , V1N2-181- 190,2012

[13].KSR Radhika, B.Naga Jyothi, Dr. Ch.Venkateswara Rao, Dr.V.Kamakshi Prasad, Accuracy Assessment of per pixel Based Classification Conference Proceedings, NC-Velasiem-2k12: Pg.73-77

Leave a Reply