The invention provides an abnormal behavior intelligent detection method and system based on a similarity graph neural network, and relates to the technical field of behavior detection, and the method comprises the following steps: an information obtaining step: shooting the abnormal behavior of a person in a monitoring video to obtain a training video sequence; a network training step: extracting human skeleton points in the training video sequence to obtain a skeleton point sequence, constructing a graph network structure, learning the skeleton point sequence by using a similarity graph neural network, and training the network; an anomaly detection step: recognizing human body skeleton points in the training video sequence, constructing a graph network structure, performing feature extraction on the skeleton point sequence by using a similarity graph neural network, and performing abnormal behavior recognition; and an intelligent recording step: automatically intercepting abnormal video clips, marking abnormal behavior types, and storing the abnormal behavior types in a database. According to the invention, the credibility of abnormal behavior recognition can be greatly improved, the recognition process is greatly simplified, the recognition time is shortened, and the real-time recognition effect is achieved.