Model generator, learning model and subject recognition device

摘要

PROBLEM TO BE SOLVED: To reduce the burden on an administrator who prepares teacher data. A subject recognition device 1 uses a type of subject included in an input image G as a label based on teacher data in which an image G and a label indicating the type of the subject included in the image G are associated with each other. It has a model generation unit 122 that generates a learning model M of a convolutional neural network to be output, and the learning model M includes a plurality of transformation layers C that perform different predetermined transformations on the input input data, and a plurality of transformations. It has a convolutional layer M21 that applies a convolutional filter F to a plurality of converted data converted by the layer C. [Selection diagram] Fig. 2