Abstract
With the development of big data, artificial intelligence has provided many intelligent solutions to people's life. However, the image in real life is mostly unlabeled, so the performance of many intelligent models show limitations. Therefore, how to use a large amount of unlabeled image data to build an efficient and high-quality model has been an urgent research topic. In this paper, we propose an unsupervised feature extraction method referred to as a stacked multi-granularity convolution denoising auto-encoder (SMGCDAE). The algorithm is based on a convolutional neural network (CNN) yet introduces a multi-granularity kernel. This approach resolved issues with image unicity by extracting a diverse category of high-level features. In addition, the denoising auto-encoder ensures stability and improves classification accuracy by extracting more robust features. The algorithm was assessed using the conventional MNIST, CIFAR-10, and CIFAR-100 dataset, achieving an average accuracy of 98.97%, 65.33%, and 46.06%, respectively. It was also used to classify a series of meningitis images, achieving an accuracy of 84.02% and outperforming conventional methods. These results suggest the algorithm is capable of extracting more discriminative high-level features and thus offers superior performance compared to existing methodologies.
Funding Information
  • National Natural Science Foundation of China (61876166, 61663046)
  • Yunnan Applied Fundamental Research Project (2016FB104)
  • Yunnan Provincial Young academic and technical leaders reserve talents (2017HB005)
  • Program for Yunnan High Level Overseas Talent Recruitment
  • Yunnan University

This publication has 42 references indexed in Scilit: