Short video multi-label classification method based on depth matrix decomposition

• Main Authors: SUN XIAOWEI, ZHAO LI, MA HAIYAN, WANG XIAOHUA, FAN FUGUI, ZHOU SHUWANG, JING PEIGUANG, SU YUTING, XIAO JING, SHI WEILI, WANG SHAOYI, SONG YUXIA, WANG SHAOCHU, WANG YUFAN
• Format: Patent
• Language: Chinese; English
• Published: 18.11.2022
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a short video multi-label classification method based on depth matrix decomposition, and the method comprises the steps: taking depth matrix decomposition as a basic framework, mining advanced semantic features hidden in short videos and labels through each layer of parameter matrix and potential representation matrix in the depth matrix decomposition; the expression ability of the potential representation matrix is enhanced through the idea of coding and decoding; label relevance information is embedded into short video content at a source end through an inverse matrix analysis method, and finally learning of a parameter matrix and a potential representation matrix is guided through supervision constraint terms. Compared with the prior art, the short video multi-label classification method has the advantages that complex structured and hierarchical effective information can be learned from the short video, and advanced semantic features and label relevance of the short video can be ef