PURPOSE: To provide a hierarchical neural network picture identification device capable of automatically setting the teacher signals of picture patterns for which setting is difficult.

CONSTITUTION: This hierarchical neural network picture identification device is provided with a picture image pickup device 201 for picking up the images of the pictures of an identification object, a picture processor 202 for generating the picture patterns to a hierarchical neural network from the image picked up pictures, a picture pattern storage device 203 for storing the generated picture patterns, a hierarchical neural network arithmetic unit 206 for fetching he picture patterns taken out from the storage device 203 and the teacher signals to the picture patterns in pairs and learning them and a learned result storage device 207 for storing the correction values of connection coefficients among the respective layers of the neural network obtained by learning. Then, a teacher signal generator 204 for calculating a distance between the picture patterns from the picture patterns stored in blue picture pattern storage device 203 and preparing the teacher signals based on the distance is provided.