Instance segmentation method and system based on multi-scale features and context attention

The invention relates to an instance segmentation method and system based on multi-scale features and context attention. The method comprises the following steps: performing data preprocessing including data enhancement and standardization processing on images in a training set; constructing a multi-scale feature fusion and context attention aggregation module to enhance feature representation by using a pyramid structure, and focusing a small target object in a pyramid feature map; constructing an instance segmentation network based on multi-scale feature fusion and context attention aggregation; training the instance segmentation network by using the images in the training set, generating an instance segmentation result and calculating a loss function, and performing back propagation to optimize parameters of the whole network to obtain a trained instance segmentation network; and inputting a to-be-processed image into the trained instance segmentation network to obtain an instance segmentation result. The