| 摘要 |
PROBLEM TO BE SOLVED: To provide a pattern recognition device and a method for obtaining higher identification accuracy. SOLUTION: A pattern recognition method includes: a step (SB) of calculating an inclinationβ<SB POS="POST">i</SB>of a zero point proportion regression line y=β<SB POS="POST">i</SB>m of an i-th data set y<SB POS="POST">i1</SB>, y<SB POS="POST">i1</SB>..., y<SB POS="POST">ik</SB>to average values m<SB POS="POST">1</SB>, m<SB POS="POST">2</SB>..., m<SB POS="POST">k</SB>of item values of respective feature items as a first contracted feature item; a step (SC) of calculating a square sum Se<SB POS="POST">i</SB>'=Σ((y<SB POS="POST">ij</SB>-β<SB POS="POST">i</SB>m<SB POS="POST">j</SB>)/s<SB POS="POST">j</SB>)<SP POS="POST">2</SP>of a value obtained by dividing deviation between the zero point proportion regression line y=β<SB POS="POST">i</SB>m and the i-th data set by standard deviation s<SB POS="POST">j</SB>for the respective data sets, and calculatingσ<SB POS="POST">i</SB>' which is an amount in proportion to a square root of Se<SB POS="POST">i</SB>' as a second contracted feature item; and a step (SD) of calculating evaluation distance D on the basis of the first contracted feature item and the second contracted feature item to the respective data sets, and performing determination on the basis of the evaluation distance D. COPYRIGHT: (C)2013,JPO&INPIT |
