摘要 |
<p>PROBLEM TO BE SOLVED: To enable a soldering check device to be applied to a check for the shape of a soldered part, by a method wherein the soldering check device is equipped with a recognition means which is capable of learning the boundary conditions of the hierarchies of an object of recognition to classify receiving picture image data extracted from a measurement means and recognizing the shape of a soldered part imitating a human classifying ability basing on sampled learning data when a new picture image is inputted. SOLUTION: A recognition means which recognizes the shape of a soldered part is capable of calculating a relative relation between pattern hierarchies to classify for the shape of a soldered part 2 converting the optional sampled picture image data extracted from a color picture image acquisition section under specific lighting conditions into vector data for a check for a soldered part 2. A relative neural network 10 which is capable of learning the synapse weight of nerve cells through the result of a comparison between the relative value and the judgment value of a user and recognizing the classification condition of the sampled picture image data of the soldered part. A sampled pattern average error function is obtained comparing the vector relative value with the judgment result of a user, whereby the shape of a soldered part can be checked.</p> |