摘要 |
PROBLEM TO BE SOLVED: To provide a pattern recognition device for recognition and classification by detecting and eliminating objects whose recognition and classification are difficult and which are eliminated from recognition objects even by human being. SOLUTION: A characteristic quantity of a recognition object is obtained (S6), a size in the characteristic quantity is noticed and an object with a very small size is excluded from the processing (S7), an output of each output element of a neural network is referenced to obtain a parameter denoting the reliability of the recognition, it is compared with a threshold level for reliability evaluation to evaluate the recognition reliability and when the reliability is low, the object is excluded from the recognition classification object (S8, S9). Through the processing above, objects causing low reliability even from automatic recognition are detected and excluded from the recognition processing. Thus, the processing efficiency is improved. |