基于深度神经网络技术的台标识别模型设计与训练

Design and Training of Logo Recognition Model Based on Deep Neural Network Technology

在监测监管工作中,为了快速、准确检测识别电视节目的播出机构,基于传统图像识别算法的台标识别工具得到广泛应用。随着图像视觉技术的快速发展,深度神经网络技术在近几年的图像识别大赛中大放异彩,相较于传统算法表现出更高的准确性和泛化性。本文基于深度神经网络技术详细论述了台标识别模型的设计与训练,通过尝试比较不同的网络结构、算法、训练超参数和训练集,最终训练出适合业务场景的识别模型。测试结果表明,该模型可以在不同场景中准确、高效地识别各种变形、半透明等传统算法难以准确识别的台标,具备很高的实用价值。

In the monitoring and supervision work,in order to quickly and accurately detect and identify the broadcasting organization of TV programs,the station logo recognition tools based on traditional image recognition algorithms have been widely used.With the rapid development of image vision technology,deep neural network technology has shined in image recognition competitions in recent years,which shows higher accuracy and generalization than traditional algorithms.Based on deep neural network technology,this paper discusses the design and training of the logo recognition model in detail.By trying to compare different network structures,algorithms,training hyperparameters and training sets,a recognition model suitable for business scenarios is finally trained.The test results show that the model can accurately and efficiently identify various deformations,translucency and other logos which are difficult to be accurately identified by traditional algorithms in different scenarios,and it has high practical value.