基于CiteSpace的深度卷积神经网络的研究述评

A Review of Research on Deep Convolutional Neural Networks Based on CiteSpace

摘要: 为了全面和系统地考察中国深度卷积神经网络的研究现状、重点热点以及未来展望，基于中国知网数据库，以“深度卷积神经网络”为主题词，筛选出2014~2025年间共3181条相关文献进行综述和知识图谱分析。结果表明：2014~2025年的10年期间，中国深度卷积神经网络的研究大体经历了初步认识(2014~2016年)，快速发展(2017~2022年)和平稳发展(2023~2025年)的三个阶段。在研究关键词中，深度学习、目标检测、图像处理、迁移学习、特征提取、故障诊断、图像分类等为该领域的高频关键词。在关键词演进分析中，该研究领域热点逐步向调制识别、轻量级、图像复原、扩张卷积等分类等进行转变。在发文作者上，以王鑫、张强、李明、郑宗生、杨军和汤一平为各自团队的核心取得丰硕的发表成果，在研究机构上，形成了以浙江工业大学计算机科学与技术学院、中国科学院大学、南京信息工程大学遥感与测绘工程学院为代表的科研院所占据重要地位。在发表期刊上，《计算机工程与应用》《中国图像图形学报》《计算机应用》等期刊为主要的见刊期刊。

Abstract: In order to comprehensively and systematically investigate the research status, key hotspots, and future prospects of deep convolutional neural networks (DCNN) in China, this study is based on the China National Knowledge Infrastructure (CNKI) database. Using “Deep Convolutional Neural Network” as the subject term, 3181 relevant publications from the period 2014~2025 were selected for review and knowledge mapping analysis using CiteSpace. The results indicate that over the decade from 2014 to 2025, DCNN research in China generally underwent three stages: initial recognition (2014~2016), rapid development (2017~2022), and stable development (2023~2025). Regarding research keywords, deep learning, object detection, image processing, transfer learning, feature extraction, fault diagnosis, and image classification are high-frequency keywords in this field. The keyword evolution analysis shows that the research hotspots have gradually shifted towards areas such as modulation recognition, lightweight models, image restoration, dilated convolution, and related classifications. In terms of prolific authors, Xin Wang, Qiang Zhang, Ming Li, Zongsheng Zheng, Jun Yang, and Yiping Tang, as cores of their respective teams, have achieved abundant publication results. Regarding research institutions, schools such as the College of Computer Science and Technology at Zhejiang University of Technology, the University of Chinese Academy of Sciences, and the School of Remote Sensing and Geomatics Engineering at Nanjing University of Information Science and Technology have emerged as leading scientific research institutions. For publication venues, journals such as Computer Engineering and Applications, Journal of Image and Graphics, and Computer Applications are the primary publication channels.