基于视觉语义分割算法的电力信息化平台多媒体元素自动提取模型  

A Multi-media Element Automatic Extraction Model for Power Information Platform Based on Visual Semantic Segmentation Algorithm

作　　者：夏立典 章超 李晨 周诗苇 胡修宇 王丽乃 XIA Lidian;ZHANG Chao;LI Chen;ZHOU Shiwei;HU Xiuyu;WANG Linai(Zhejiang Huayun Information Technology Co.,Ltd.,Hangzhou 310000,China)

机构地区：[1]浙江华云信息科技有限公司,浙江杭州310000

出　　处：《微型电脑应用》2024年第10期116-119,共4页Microcomputer Applications

摘　　要：在电力信息化平台中,提取所需多媒体元素时面临图像滤波去噪效果不佳、图像质量粗糙以及提取效果受偏态指标影响较大的问题,因此,提出一种基于视觉语义分割算法的电力信息化平台多媒体元素自动提取模型。该模型采用图像滤波去噪增强算法,通过脉冲检测多媒体图像极值找出多媒体图像噪声点,结合自适应中值滤波,完成电力信息化平台中多媒体图像滤波去噪处理;构建由全卷积分割网络和区域建议网络组成的视觉语义分割网络,将处理后的多媒体图像作为视觉语义分割网络输入,通过全卷积分割网络分割预处理后多媒体图像,提取多媒体元素,结合区域建议网络获取的包含类别标记信息的区域建议框,优化全卷积分割网络多媒体元素提取效果。实验结果表明:该模型预处理后的多媒体图像清晰度较高,可有效提取所需多媒体元素,且不同偏态程度下的多媒体元素提取时间在2.2 s至2.4 s之间,提取效率更高。In the power information platform,the extraction of required multi-media elements faces problems such as poor image filtering and denoising effect,rough image quality,and significant influence of skewness indicators on the extraction effect.Therefore,a multi-media element automatic extraction model based on visual semantic segmentation algorithm is proposed for the power information platform.This model adopts an image filtering denoising enhancement algorithm,which detects multi-media image noise points through pulse detection of extreme values,and combines adaptive median filtering to complete multi-media image filtering denoising processing in the power information platform,This paper constructs a visual semantic segmen-tation network consisting of a fully convolutional segmentation network and a regional suggestion network.The processed multi-media images are used as inputs to the visual semantic segmentation network.The preprocessed multi-media images are segmented through the fully convolutional segmentation network,and multi-media elements are extracted.Combined with the regional suggestion box containing category marker information obtained by the regional suggestion network,the multi-media element extraction effect of the fully convolutional segmentation network is optimized.The experimental results show that the pre-processed multi-media images of this model have higher clarity and can effectively extract the required multi-media ele-ments.The extraction time of multi-media elements under different degrees of skewness is between 2.2 s and 2.4 s,which in-dicates higher extraction efficiency.

关 键 词：语义分割 信息化平台 多媒体元素 全卷积分割网络 区域建议网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]