Research on the detection model of film and television music copyright infringement based on contrastive learning and isolation forest

With the expansion of film and television music distribution channels, copyright infringement has become increasingly covert, while existing detection methods suffer from insufficient feature extraction and low anomaly recognition accuracy. This paper proposes an infringement-detection model based on contrastive learning and Isolation Forests, given the prevalence of copyright infringement in film and television music. Traditional detection methods that rely on manual comparison struggle to meet efficiency and accuracy requirements. Key innovations include: 1) Constructing a feature representation network for film and television music using contrastive learning to enhance feature discrimination in dimensions such as melody and rhythm; 2) Integrating an Isolation Forest into the detection process to achieve unsupervised anomaly recognition of infringing segments; 3) Designing a feature fusion module to reduce feature bias across music file formats. Experiments conducted on a dataset containing 2,000 film and television music tracks and 300 infringement samples demonstrated a model accuracy of 94.2% and an F1 score of 93.7%. This study provides an efficient technical solution for copyright protection in film and television music.