Water obstacle detection system and method based on unknown target attention

The invention relates to an water obstacle detection system based on attention to an unknown target. The system comprises a feature extraction module which is used for carrying out feature extraction on an input training image and generating a corresponding feature map; a candidate region acquisition module used for selecting K candidate regions with target existence possibilities from the feature map; a feature pooling module used for converting the input features of the candidate region into feature vectors with fixed lengths; a weighting module used for weighting the feature vectors; an unknown class clustering and recognition module used for clustering the weighted feature vectors and recognizing unknown classes; an unknown attention weight generation module used for generating a weight attention matrix; and a classifier module used for classifying the input feature vectors and outputting an obstacle detection result, and compared with the prior art, the invention has the advantages that the learning abil