Abstract: Due to the high redundancy of image data, traditional image classification methods have low accuracy. Although the deep learning structures earn outperformance, they are hard to be trained. This paper proposes a shallow fuzzy K-means clustering network to do image classification. This network uses fuzzy K-means clustering to help build feature vectors, which is applied to train the shallow network classifier to do image classification tasks. The proposed algorithm can achieve better performance compared with the traditional methods. A solid foundation for further research is laid through the analysis and explanation of the experimental data.

Key words: image classification, deep learning, K-means clustering, shallow network

摘要： 由于图像数据的冗余性较高，传统的图像分类方法的分类准确率较低，深度学习方法较传统方法提高了图像分类的准确率，但其训练较为复杂。提出了一种浅层模糊K均值图像分类网络，其基本思想是利用模糊K均值聚类求出的聚类中心构造图像特征向量，再利用特征向量训练浅层网络分类器，最后利用训练好的分类器完成图像分类。通过与传统方法的对比，验证了该方法能够较好地完成图像分类任务，并对实验结果进行了分析，为以后的工作奠定了基础。

关键词: 图像分类, 深度学习, K均值聚类, 浅层网络