摘要
电视台标检测是网络视频审核的常用方法,但传统台标检测方法检测成功率较低,文章将基于深度学习的SSD算法应用于台标检测。首先选取79类237个常见台标视频作为台标基准库,以每秒一帧的速率对台标基准库进行处理,共127 980张图像帧,其中95 586张用于台标检测模型训练,32 394张用于模型测试,算法每训练2 000次进行模型迭代,共训练90 000次并选取最优模型。经过大量的台标样本实验测试,准确率可达98.2%,优于权威文献中经典方法。表明该方法具有较高准确率和高扩充性。
TV logo detection is a common method of network video audit,but traditional logo detection exists low accuracy rate.In this paper,we apply SSD algorithm based on deep learning to logo detection.Firstly,it selects 79 kinds of 237 common logo video as a benchmark database,the logo is processed to benchmark rate per frame,a total of 127 980 frames,95 586 of them are used to train the logo detection model,32 394 sheets for model testing.The algorithm is iterated for each training 2 000 times,training 90 000 times and selecting the optimal model.In many experiments of sample test station,the accuracy is up to 98.2%,it is superior to the classical method in the authoritative literature.It shows that the method in this paper has high accuracy and high scalability.
作者
刘琨
Liu Kun(School of Information and Control Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China)
出处
《无线互联科技》
2018年第15期36-38,共3页
Wireless Internet Technology
