Identification network classification node self-increasing method

Abstract

The invention relates to an identification network classification node self-increasing method. The identification network based on a deep neural network adds the following three modes: an identification connection mode, a training point increasing mode and a sleep forgetting mode. In the process of continuously inputting the data flow into the identification network, automatic increase and automatic labeling of the classification nodes in the identification network are realized through automatic switching of the three modes.

Description

technical field [0001] The invention relates to the field of electrical digital data processing, in particular to a self-incrementing method for identifying network classification nodes. Background technique [0002] Under this artificial wave, intelligent applications such as unmanned driving, industrial robots, service robots, and intelligent robots have not achieved the expected results. The reason is that supervised learning has limitations in various aspects. For example, the limitation of labeled data, the data collected and labeled manually are highly discrete data, the network after training is static, and classification nodes cannot be added dynamically. This is the biggest difference from human intelligence. [0003] The existing unsupervised learning algorithm, the effect of the algorithm cannot be compared with the effect of the existing supervised learning algorithm. The newly proposed semi-supervised learning algorithm and self-supervised learning algorithm a...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a recognition network in order to solve the problem in the prior art that the deep learning recognition network classification nodes based on the deep neural network are formed through batch training. After the training is completed, new classification nodes cannot be added immediately. The classification node self-increment method realizes the self-generation and automatic labeling of classification nodes, so that the recognition network can open the training mode at any time, add new classification nodes and do not need to use manually labeled data

Images