Computer vision (CV) attempts to mimic human eyes for image processing and identifications of detailed visual information, such as object positions, features of appearances, and even human emotions and behaviors. In this research, more than one hundred literatures, relating to applying deep learning (DL) methodologies in advanced computer visions (2010~2018), are reviewed and analyzed. The objective is to discover the state-of-the-art DL methods, topics, and trends for CV and their practical applications. DL algorithms aim at representing multi-levels of distributed neural networks. Because of the enhancement of high speed computational power, DL modeling, based on accumulated big data analytics, has found practical applications for non-supervised intelligent decision supports, such as detection of product defects and prognosis of machine malfunctions based on real-time signal or feature data analyses. There are a vast number of literature, describing DL related researches, developments, and implementations for problem solving. For the comprehensive mining of the related literature, we integrate Latent Dirichlet Allocation (LDA), K-means (Clustering), and normalized term frequency-inverse document frequency (NTF-IDF) approaches to discover, or called technology mining, of the major trends in DL for computer visions, specifically for key applications in object detection, semantic segmentation, image retrieval, and human pose estimation.
|主出版物標題||2019 IEEE International Conference on Systems, Man, and Cybernetics (IEEE SMC 2019)|
|出版狀態||Published - 2019|