Cardiac ultrasound image recognition method based on query contrast learning framework

The invention discloses a heart ultrasound image recognition method of a query-based contrast learning framework. The method comprises the following steps: establishing a query-based visual representation contrast learning framework; finding an image which is the same as the image and a sample which is similar to the image in the data set for direct alignment; in the training process, contrast learning is carried out by reducing the distance between directly facing samples; after training, a projection head is frozen, and different echocardiograms are classified only by using an encoder GLNet and representation; when data enhancement is carried out on the same image, two data enhancement embedding parts are used as positive alignment, and negative samples are from other images in the same batch. According to the method, the positive sample is constructed through the query set, the similar feature vectors are found in the potential space, and better learning feature representation is achieved; a multi-layer CN