Map 
  
indcovers 
"Vers 
ated by the 
ation. 
((k,i))^ 
    
k: Study site ID 
N(k): Number of landcovers in the study site (k) 
A(k,i): Estimated area ratio of landcover(i) in the study 
site(k) by classification 
Ar(k,i):Actual area ratio of landcover(i) in the study 
site(k) 
By introducing the estimated area ratios of 
landcovers as a priori probabilities, the 
accuracy of  landcover classification was 
significantly improved. The RMS errors of 
landcover area ratios in each small study site 
by Bayes' classification were less than those 
estimated only by maximum likelihood 
estimation. 
5. Discussion & Conclusion 
The results of the experiment to estimate 
the  landcover area ratios as a priori 
probabilities of landcovers showed that the 
quality of training data affects the accuracy 
of landcover area ratios. So, a new algorithm 
like an experiment[Yoshino(1995)] is needed 
to be developed in order to determine proper 
training data sets. 
As a result of this research, the 
following are concluded. 
1) An algorithm to decompose mixed data can be 
used to estimate the area ratios of landcovers 
in the study area as a priori probabilities 
for the Bayes' classifier. 
2) The area ratios of landcovers were 
estimated fairly well by the algorithm of the 
decomposition of mixed data. 
3) Introducing estimated area ratios of 
landcovers as a priori probabilities for the 
Bayes' classifier resulted in better accuracy 
of landcover classification by maximum 
likelihood classification. 
4) It was found that this new algorithm to 
estimate area ratios of landcovers in the 
study area strongly requires very good 
training data in the study area. 
5) Estimating the number of proper landcovers 
for the study area and selecting good training 
signatures are very important. 
6) A new procedure to determine proper 
training datasets for the estimation of 
landcover area ratios must be developed. 
7) The results of this research will be very 
useful for hyper multispectral images like 
AVIRIS data. 
This research was supported by the 
Sasakawa Scientific Research Grant from the 
Japan Science Society in 1995. We are greatly 
indebted to many students in the department of 
agricultural engineering of the University of 
Tokyo for their assistance at the time of the 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
DN OMEN Je. 
ground truth. 
Reference 
1)Inamura,Minoru: Analysis of Remotely Sensed 
Image Data by means of Category Decomposition, 
Trans. of the Institute of Electronics, 
Information and Communication Engineers, 
Vol.J70-C(2),pp.241-250, 
1987 
2)Ito,Tadashi and Sadao Fujimura: Estimation 
of Cover Area of Each Category in a Pixel by 
Pixel Decomposition into Categories, Trans. of 
the Society of Instrument and Control 
Engineers Vol.23(8),p.20-25,1987 
3)Matsumoto,Masao, Kouki Fujiku, Kiyoshi 
Tsuchiya,Kohei Arai: Category Decomposition 
based on Maximum Likelihood Estimation, 
Journal of the Japan Society of 
Photogrammetry, Vol.30(2), p.25-34,1991 
4)strahler, Alan H,:. The Use of Prior 
Probabilities in Maximum Likelihood 
Classification of Remotely Sensed Data, Remote 
Sensing of Environment 10, pp.135-163, 1980 
5)Yoshino, Kunihiko: One procedure to 
determine landcover categories for landcover 
supervised classification in remotely sensed 
image, Proc. of the annual conference of Japan 
Society of Irrigation and Drainage, 
Reclamation Engineering, p.82-83, 1995 
997