Partial mark learning method for multi-class noise separation

• Main Authors: WEI HUANGZHAO, LIU ZHENBING
• Format: Patent
• Language: Chinese; English
• Published: 04.08.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a partial label classification method for multi-class noise separation, and the method comprises the steps: firstly obtaining an initial correct label through an iterative propagation technology, then fully recognizing different noises (impulse noise and Gaussian noise) in reality, building different noise classifiers, and finally separating out the corresponding noises. And the real label-preserving classifier completes the identification of the true value labels. According to the multi-class noise separation framework constructed in the invention, the problem of multi-class noise in current partial mark learning is solved, the classification precision performance of partial mark learning is improved, and the generalization performance of a partial mark learning algorithm is also improved. 本发明公开一种多类噪声分离的偏标记分类方法，所述方法包括：首先通过迭代传播技术得到最初的正确标签，然后充分识别现实中的不同噪声(脉冲噪声和高斯噪声)，由此建立不同噪声分类器，并在最后分离出所对应的噪声，保有真实标签分类器完成真值标签的识别。本发明中构建的多类噪声分离框架，解决目前偏标记学习中多类噪声的问题，提高了偏标记学习在分类精度上的性能，同时还提高了偏标记学习算法的泛化性能。