Apparatus for Recognizing Object based on Deep Learning In Embedded Device

摘要

A deep learning-based object recognition device in an embedded device is disclosed. An apparatus for deep learning-based object recognition in an embedded device according to an embodiment of the present invention includes a feature extraction neural network that outputs a feature vector for an input image and a classifier that predicts a class of an input image from the feature vector, wherein the feature extraction neural network is a convolutional neural network that is scale-learned in advance based on deep learning so that feature vectors output for a plurality of input images belonging to the same class are approximated in the feature vector space, and the classifier is Among at least one class to which a predetermined number of reference samples belong, a class having the largest number of adjacent reference samples belonging to each class is determined as a class for the input image, wherein the reference sample is at least one input image belonging to each class. may be a feature vector obtained in advance by inputting to the feature vector extraction neural network.