Human-object interaction detection method and device based on affordability-scene knowledge prompt

The invention discloses a human-object interaction detection method and device based on affordability-scene knowledge prompt, which can improve the recognition capability in a real scene and predict a more accurate HOI detection result. The method comprises the following steps: (1) constructing an affordability-scene knowledge graph ASKG; (2) building a human-object interaction detection model: in a training stage, inputting an RGB image, in the first stage, detecting people and objects, and identifying scene categories and encoding original features; in the second stage, the priori knowledge embedding module encodes specific knowledge embedding, the instance domain adaptive fusion module encodes instance features and knowledge embedding to obtain instance perception features, the scene domain adaptive fusion module encodes scene features and knowledge embedding to obtain scene perception features, and finally a result is output through the HOI recognition module; and (3) executing human-object interaction de