中图分类号：TP391 文献标志码：A DOI: 10.16157/j.issn.0258-7998.256320 中文引用格式： 李文钊，董晓炜，赵芳，等. 基于改进MobileNetV3-SSD的数据可视化区域实时检测方法[J]. 电子技术应用，2025，51(12)：103-109. 英文引用格式： Li Wenzhao，Dong Xiaowei，Zhao Fang，et al. Real-time detection method for data visualization regions based on improved MobileNetV3-SSD[J]. Application of Electronic Technique，2025，51(12)：103-109.
Real-time detection method for data visualization regions based on improved MobileNetV3-SSD
Li Wenzhao1，Dong Xiaowei2，Zhao Fang1，Yang Cai1
1.Chongqing Meteorological Information and Technology Support Center；2.Chongqing Ceprei Industrial Technology Research Instutute Co.， Ltd.
Abstract： Data visualization plays an important role in meteorological big data. Through the display of data panel, users can visually and conveniently understand the basic situation of complex data. However, due to the variability of the location and size of different data sections of the panel, it is difficult for users to quickly find the location of various data sections. Based on this, this study proposes a real-time detection method of data visualization area based on improved MobileNetV3-SSD.On the basis of standard SSD, the attention mechanism is combined, and the improved FPN module is used to fuse the feature information, which can efficiently locate the “Title” position of the target data section, and then display that area for users. The proposed model can significantly reduce the number of parameters, and the precision ,recall and mAP indexes reach 83.05%, 85.02%, 74.35% respectively, meanwhile, the inference time based on CPU is reduced to 244 ms per image.
Key words : data visualization；SSD algorithm；feature pyramid；automatic generation of datasets
