The invention discloses a method for detecting salient regions in sequence images based on an improved visual attention model. The method mainly aims at solving the problems that an existing method for detecting salient regions based on the visual attention model is complex in process and poor in real-time performance. The method includes the implementation steps that firstly, a watching graph of salient regions of a study image is generated, a feature saliency graph and weight vectors of the feature saliency graph of the study image are generated, and salient point coordinates are recorded; secondly, a saliency graph of a test image is generated, salient point coordinates of the saliency graph of the test image are recursively predicted through the salient point coordinates of the study image, and a restraint core function is established to highlight the regions where salient points are located; thirdly, the salient point coordinates are updated, and salient regions of a next test image are predicted through a salient point coordinate recurrence relation and the restraint core function; fifthly, the salient regions of the sequence images are detected by cyclically executing the third step and the fourth step. The salient regions in the sequence images can be detected in real time, the model is simple and effective, and the method can be used for target recognition.