Binary classification of pornographic and non-pornographic materials using the sAI 0.4 model and the modified sexACT database

Purpose: Neural networks may be used to solve problems in the field of psychology and sexology. In particular, it seems that neural networks may be important to limit the unintentional contact of minors with pornographic material. The aim of the study was to create the neural networks model for the classification of pornographic (also fetishist) materials from non-pornographic materials. Methods: In order to create a new model, the sAI 0.3 model was used as the basic model. The fast.ai library version 1.0.55 was used. A modified version of ResNet152 was adopted as the neural network architecture. The sexACT database was modified to include new training material – 1630 non-pornographic photos of women. A total of 1304 photos (80% of the set) were used to train the network and the remaining 326 photos (20% of the set) were used for its later validation. Results: As a result of the research, the sAI 0.4 model was created, enabling binary classification of pornographic and non-por-nographic materials with 96% accuracy. The model tends to make more the first type of error than the second type of errors. The model has a high precision (0.94) and high sensitivity (0.88). The final validation loss was 0.1314. Conclusions: The potential benefits of using the discussed model from a clinical perspective were discussed. The application of the discussed model could prevent the negative effects of contact of minors with pornographic material, which could consequently limit the prevalence of risky sexual behavior or negative psychosocial effects.