Scholarlink

This study presents an intelligent framework for proactive detection of video codec anomalies, designed to safeguard uninterrupted broadcasting at TV Laayoune, part of Morocco’s SNRT network. Unlike conventional validation tools, the proposed system employs a multi-stage pipeline: data augmentation techniques enhance robustness to rare codec variations, while anomaly detection is handled through autoencoders. A hybrid deep learning model—integrating CNNs, LSTMs, and an attention mechanism—extracts both spatial patterns and temporal dependencies from video streams. Key metadata attributes, identified as critical for playback quality, are prioritized in the training process. Embedded into the broadcasting workflow, the framework enables real-time error identification and automated alerts, reducing reliance on manual inspection and minimizing service interruptions. Experimental results show a detection accuracy of 97%, consistently outperforming traditional ML classifiers. The findings highlight the potential of advanced ML/DL architectures to transform reliability standards in modern television networks by offering a scalable and efficient codec validation solution.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid CNN-LSTM with Attention and Autoencoder-Based Anomaly Detection for Enhanced Pre-Broadcast Video Codec Validation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Khalid El Fayq,
  • Said Tkatek,
  • Lahcen Idouglid

摘要

This study presents an intelligent framework for proactive detection of video codec anomalies, designed to safeguard uninterrupted broadcasting at TV Laayoune, part of Morocco’s SNRT network. Unlike conventional validation tools, the proposed system employs a multi-stage pipeline: data augmentation techniques enhance robustness to rare codec variations, while anomaly detection is handled through autoencoders. A hybrid deep learning model—integrating CNNs, LSTMs, and an attention mechanism—extracts both spatial patterns and temporal dependencies from video streams. Key metadata attributes, identified as critical for playback quality, are prioritized in the training process. Embedded into the broadcasting workflow, the framework enables real-time error identification and automated alerts, reducing reliance on manual inspection and minimizing service interruptions. Experimental results show a detection accuracy of 97%, consistently outperforming traditional ML classifiers. The findings highlight the potential of advanced ML/DL architectures to transform reliability standards in modern television networks by offering a scalable and efficient codec validation solution.