一种用于电视节目播出异态识别的人工智能模型训练方法  

Training Method of AI Model for Television Broadcast Abnormal Image Recognition

作　　者：汤冻[1] 奚晓轶[1] 闫涛[1] TANG Dong;XI Xiaoyi;YAN Tao(Jiangsu Broadcasting Corporation,Nanjing 210013,China)

机构地区：[1]江苏省广播电视总台,江苏南京210013

出　　处：《电视技术》2023年第1期61-65,共5页Video Engineering

摘　　要：提出一种用于电视节目画面播出异态识别的人工智能模型训练方法,解决没有针对广播电视安全播出监看领域图像识别的神经网络模型,以及自有模型训练异态样本量少的问题。最终训练集可以达到97%的训练精度,实际测试检测准确度超过90%,可以满足应用需求。This paper proposes an AI model training method for the recognition of abnormal picture of TV program broadcast, which solves the problem that the number of abnormal samples is small, and there is no neural network model for image recognition in the field of broadcast and television broadcast monitoring. The final training set can reach 97% of the training accuracy, and the actual test detection accuracy exceeds 90%, which can meet the application requirements.

关 键 词：人工智能 图像识别 模型训练 

分 类 号：TN931.3[电子电信—信号与信息处理]