DEVICE AND METHOD FOR RECOGNIZING PICTURE BY USING NEURAL NETWORK, AND CONVERTER

摘要

PROBLEM TO BE SOLVED: To enable to learn in a short period and to reduce erroneous recognition in a picture recognizing device using neural network. SOLUTION: In picked up picture data 1, the coordinate of centroid P of oil spill region detected in picture processing is obtained and a fixed range region 1a (attentional region 1a) centering around this coordinate P is segmented. Picture data of this segmented region of interest 1a is density converted and compressed and inputted to an input layer 11 of a neural network 10. The neural network 10 is made to learn data that includes factors causing erroneous recognition.