基于双重检测模型的农业气象科普视频镜头分割算法  

Shot Segmentation Algorithm of Agrometeorological Popular Science Video Based on Double Detection Model

作　　者：徐迎春 丰洪才 刘立 XU Ying-chun;FENG Hong-cai;LIU Li(Wuhan Meteorological Observatory;Network&Information Center,Wuhan Polytechnic University;Information Technology Department of Wuhan Dongxihu Vocational and Technical School,Wuhan 430023,China)

机构地区：[1]武汉市气象台 [2]武汉轻工大学网络与信息中心 [3]武汉市东西湖职业技术学校信息技术系,湖北武汉430023

出　　处：《软件导刊》2022年第9期159-166,共8页Software Guide

基　　金：武汉市气象局一般项目(WHY202205)。

摘　　要：为了提高农业科普气象视频的检索效率,以农业气象科普视频为例,提出一种基于双重检测模型(初检和复检)的镜头分割算法。初检阶段先按等面积矩形环的方式划分每一帧图像,从而突出视频图像中心主体部分,再采用自适应双阈值法对视频进行初次分割;复检阶段采用改进的sift算法进一步修正初次检测结果。通过对3段农业科普视频进行分割实验,结果表明,该算法能够较为有效地分割农业气象科普视频,其中平均查全率达到91.2%,平均查准率达到92.3%。该算法能够满足广大农民用户专业化与个性化的检索需求,提高农业气象视频的利用率,促进我国农业经济和气象科普事业的发展。In order to improve the retrieval efficiency of agricultural meteorological popular science video,this paper takes agricultural meteorological popular science video as an example,and proposes a shot detection algorithm based on dual detection model(initial detection and re detection). In the initial detection stage,each frame image is divided into equal area rectangular rings to highlight the main part of the video image center,and then the adaptive double threshold method is used to segment the video for the first time;The improved sift(scale invariant feature transform)algorithm is used to further modify the initial detection results. The segmentation experiments of three agricultural popular science videos show that the overall algorithm design in this paper can segment agricultural meteorological popular science videos effectively,with an average recall rate of 91.2% and an average precision rate of 92.3%. The algorithm meets the professional and personalized retrieval needs of farmers,improves the utilization of agrometeorological videos,and promotes the development of agricultural economy and meteorological science popularization.

关 键 词：农业气象科普视频 等面积矩形环 自适应双阈值 改进sift算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]