Semi-supervised image classification method based on dictionary deep learning

Abstract

The invention discloses a semi-supervised image classification method based on dictionary deep learning. The method comprises the following steps of: preparing the raw materials; constructing a cost function L1 (Zl, Yl) of the labeled data and a cost function Lu (Zu, P) of the unlabeled data according to a Softmax cost function of the deep neural network; constructing an overall model function according to the cost function L1 (Zl, Yl) of the labeled data and the cost function Lu (Zu, P) of the unlabeled data; And using an alternating optimization algorithm to train the overall model, whereinthe training optimization process comprises joint class estimation based on dictionary learning and Softmax network information, and joint learning of a neural network and unlabeled class estimation.According to the method, dictionary learning is combined with the Softmax classifier of the deep neural network, so that the exploration capability of the deep neural network on unlabeled data is enhanced, and the network feature learning capability and the classifier learning capability are greatly improved. The method is suitable for the field of computer vision or mode recognition.

Description

technical field [0001] The invention relates to the fields of computer vision and pattern recognition, and more specifically, to a semi-supervised image classification method based on dictionary deep learning. Background technique [0002] In the field of computer vision and pattern recognition, image classification is the basis of vision technology. With its powerful feature representation ability and similar structure to the human visual system, deep neural networks have achieved great success in image classification tasks and have attracted extensive attention from researchers. In order to improve the accuracy of image classification, it is necessary to continuously develop more powerful and complex deep neural networks. However, the success of deep neural networks relies heavily on a large amount of labeled data, and the more complex the neural network architecture, the greater the demand for labeled data. However, the acquisition of labeled data requires a lot of manp...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem that the deep neural network can obtain the effective features of the labeled data, but cannot obtain the unlabeled data and the semi-supervised dictionary method can absorb the unlabeled data into the discriminative dictionary learning, but cannot extract the effective features of the unlabeled data The problem is to provide a semi-supervised image classification method based on dictionary deep learning, which uses dictionary learning to complement the Softmax classifier of traditional deep neural network, and can simultaneously perform feature learning and classifier learning on unlabeled data, which greatly improves network performance

Images