A multi-label object recognition method based on convolutional neural network

Abstract

The invention discloses a multi-label object recognition method based on a convolutional neural network. The method utilizes the inclusion relationship between labels, constructs the CNN feature extraction part of each label in the order of inclusion, and continuously abstracts and extracts through the convolution operation. The features of each layer, and the classifiers of each label are set at different depths of the network, and the feature map extracted by the CNN feature extraction part of the corresponding label is input to the corresponding classifier, and multiple classifiers are used for error back propagation. , supervise and train the network weight parameters of the corresponding layers, and finally obtain each label category to complete the recognition. The multi-label convolutional neural network adopted in the present invention can well solve the information fusion between multiple labels, solve the problem of low accuracy of traditional multi-label object recognition, and improve the efficiency of training and recognition at the same time.

Description

technical field [0001] The invention belongs to the technical field of image processing, and particularly relates to a multi-label object recognition method based on a convolutional neural network. Background technique [0002] With the rapid development of artificial intelligence technology, automatic object recognition has become a hot research topic at home and abroad in recent years, and has broad application prospects in the fields of intelligent monitoring, telemetry and remote sensing, robotics, and medical image processing. In real life, there are many kinds of objects, and the individual similarity is high. Humans can visually perceive information such as shape, color and distance, and synthesize these information to accurately determine the category of the object, but this is often more difficult for computers. difficulty. Therefore, how to make the computer have the recognition ability similar to or even surpassing that of human has become an important direction ...

AI Technical Summary

Problems solved by technology

[0006] Purpose of the invention: Aiming at the problem that the accuracy of the object recognition algorithm is not high due to the repeated extraction of image features by a single-label convolutional neural network and the lack of connection between each label network in the prior art, it provides a method that utilizes the relationship between labels. Inclusion relationship, feature extraction and classification at different levels, multi-label object recognition method based on convolutional neural network to solve the technical problem of low accuracy of traditional object recognition

Images