基于卷积神经网络的数字电视劣质画面检测算法设计

Design of Digital Television Inferior Picture Detection Algorithm Based on Convolutional Neural Network

为降低无线电视发射台值班人员的工作强度,提高监测效率,本文提出一种基于卷积神经网络的数字电视劣质画面检测方法。基于Pytorch深度学习框架,选取Mobile Net V2网络进行简化、优化来设计新的网络模型从而实现目标。截取了多套不同时段的数字电视节目画面作为数据集对网络模型进行训练、验证和测试。新网络模型在测试集上达到了95.59%的准确率和6.80毫秒的平均处理时间。创新性地验证了使用人工智能深度学习技术来检测数字电视劣质画面的可行性。

In order to reduce the working intensity of the watchman in the radio TV transmitting station and improve the monitoring efficiency,this paper proposes a method of digital TV inferior image detection using convolutional neural network.Based on the Pytorch deep learning framework,the MobileNetV2 network is selected for simplification and optimization to design a new network model to achieve the goal.The various pictures from different types of TV programs at different time periods are captured to be used as datasets to train,verify and test the network model.The new network model achieves an accuracy of 95.59%and an average processing time of 6.80 ms on the test set.It innovatively verifies the feasibility of using the artificial intelligence deep learning technology to detect inferior pictures of digital TV.