| 摘要 |
<p>PROBLEM TO BE SOLVED: To obtain a method and an apparatus in which a detecting performance can be adjusted by a method wherein one neural netwrok is selected out of neural networks in which the relationship between a pattern used to detect an echo cignal and a desired pattern is learned in advance and whose detecting rate is different and a discriminating and processing operation is performed. SOLUTION: An input processing part 2 performs a beam phasing processing operation to an echo signal received by a transducer 1, and it outputs a time-series signal at an amplitude value. A preprocessing part 3 divides the time-series signal at the amplitude value into short-time section data so as to be frequency-converted, it converts the section data into time-series data (a frame) at a spectrum level, it cuts a frame section corresponding to the length of the echo signal from the target, and it generates a frame (the time), a frequency and a pattern, for three-dimensional detection, at a specrum level. In a neural network discriminating and processing part 4, the pattern for detection is input by a changeover processing part 41 to one neural network, designated by an operator, out out a plurality of neural networks which are learned in advance and whose detecting performance is different, whether the pattern is an echo signal from the target is discriminated, and a discrimination result is output.</p> |
