Multiple ANN-structure-based methods for improving image recognition performance under low physical and temporal space

摘要

The present invention is a multi-dimensional neural network, in order to compensate for the disadvantages of requiring a long time for learning and algorithm execution of the existing CNN (Convolution Neural Network) and lowering the accuracy, using a decision-making neural network learned from reconstruction data, and the result of a multi-dimensional neural network. It relates to a learning method using a new network structure that makes appropriate judgments about the network, the step of constructing a multi-dimensional neural network using an auxiliary neural network that has learned about images that are not recognized by the main network and the multi-dimensional neural network. Generating data to use the output result as an input of a decision making neural network; And constructing and learning a decision-making neural network to input the regenerated data.