Journal of Frontiers of Computer Science and Technology ›› 2021, Vol. 15 ›› Issue (1): 173-184.DOI: 10.3778/j.issn.1673-9418.1911057

• Graphics and Image • Previous Articles     Next Articles

Multi-grained Fusion Image Feature Learning with Fuzzy Rule System

MA Xiang, DENG Zhaohong, WANG Shitong   

  1. School of Digital Media, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2021-01-01 Published:2021-01-07

多粒度融合的模糊规则系统图像特征学习

马翔邓赵红王士同   

  1. 江南大学 数字媒体学院,江苏 无锡 214122

Abstract:

Currently, the most popular image feature learning method is deep neural network, which can automatically extract efficient features through feature learning for tasks such as classification and recognition without human     involvement. However, the deep neural network image feature extraction method currently faces many challenges. Its effectiveness relies heavily on large data, and it is usually regarded as a black box model with poor interpretability. Aiming at the above challenges, a more interpretable and scalable feature learning method, multi-grained fusion image feature learning with fuzzy rule system, is proposed in the paper. The method is based on Takagi-Sugeno-Kang fuzzy system (TSK-FS) with fuzzy rule inference. This method extracts image features through rule-based TSK-FS, so the feature learning process can be explained by rules. Then, multi-granularity scanning also further enhances its feature learning ability. Extensive experiments have been conducted on image datasets of different scales, and the results show the proposed method is effective on image datasets.

Key words: feature learning, fuzzy system, non-linear model, image classification, interpretability

摘要:

当前最流行的图像特征学习方法是深度神经网络,该类方法无需人工参与即可自动地通过特征学习提取高效的特征,用于分类识别等任务。然而,深度神经网络图像特征抽取方法目前也面临着诸多挑战,其有效性严重依赖大规模的数据,且通常被视为黑盒模型,解释性较差。针对上述挑战,以基于模糊规则推理的TSK模糊系统(TSK-FS)为基础,提出了一种适用于不同规模数据集且易于理解的特征学习方法——多粒度融合的模糊规则系统图像特征学习算法。该方法通过基于规则的TSK-FS抽取图像特征,因而特征学习过程是可以利用规则进行解释的。其次,多粒度扫描也使得其特征学习能力进一步提升。在不同规模的图像数据集上进行了充分的实验,实验结果表明该方法在图像数据集上具有较好的有效性。

关键词: 特征学习, 模糊系统, 非线性模型, 图像分类, 可解释性