Manifold learning and gradient lifting model-based picture multi-label classification method

Abstract

The invention discloses a manifold learning and gradient lifting model-based picture multi-label classification method. Constructing a weighted graph from the training data set, obtaining a non-negative weight matrix by solving the first minimization model, establishing a second minimization model according to the weighted graph, solving to obtain a reconstructed label matrix, constructing the training data set according to the reconstructed label matrix, training a binary correlation model, and predicting to obtain a label matrix; and establishing a regression device minimization solution forthe feature vector matrix of the picture, enhancing the feature vector matrix by using an iterative prediction result matrix, constructing a data set by combining a negative gradient matrix, trainingand learning to obtain weak regression devices, summing all the weak regression devices to obtain a final regression device, and processing and judging a pre-to-be-tested picture. According to the method, the multi-label classification prediction performance of the picture can be improved by fully utilizing the correlation between the partial multi-label data of the picture, the disambiguation ofthe partial label data can be realized, the accuracy and the robustness are improved, and the performance of the method is superior to that of the existing partial multi-label method of the picture.

Description

technical field [0001] The invention relates to the problem of over-many labels in label classification, in particular to a two-stage over-many label learning method based on manifold learning and gradient promotion. Background technique [0002] In the image multi-label classification problem, a picture can be associated with multiple labels at the same time. A common assumption in traditional image multi-label classification problems is that each image in the training dataset is accurately labeled, that is, supervised. Unfortunately, in many real image multi-label classification problems, noise-free labels are difficult to obtain. In contrast, obtaining a set of candidate labels is very easy. This type of problem is defined as PartialMulti-Label Learning (PML) for pictures. [0003] The basic assumption of PML is that the correct label of the image is hidden in the set of candidate labels, and it is invisible to the learner. The most intuitive approach to PML is to con...

AI Technical Summary

Problems solved by technology

Existing PML methods all focus on one aspect of candidate label set disambiguation or label correlation extraction, so these methods have limited predictive performance

Images