Semi-supervised learning pornographic image detection system and method based on mask self-coding

• Main Authors: SANG YONGSHENG, LYU JIANCHENG, NIU CHAOQUN, XU ZI'AO, CHEN YONGZHI
• Format: Patent
• Language: Chinese; English
• Published: 16.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a semi-supervised learning pornographic image detection system and method based on mask self-encoding. The system comprises a mask encoder, a decoder and a classifier, wherein the decoder and the classifier are connected with the mask encoder; the mask encoder is used for preprocessing and encoding an original picture to obtain a multi-level image feature map and a mask map; the decoder is used for splicing and decoding the multi-level image feature graph and the mask graph, and calculating the restoration loss of the multi-level image feature graph and the mask graph; the loss is used for training the mask encoder and decoder; the classifier is used for receiving the multi-layer feature map output by the trained mask encoder and carrying out classification judgment on the multi-layer feature map to complete pornographic image detection, model training can be completed by only needing a small part of label data in the training process, and the method can adapt to the situation that the