Cardiac ultrasound image recognition method based on GLNET and self-attention

The invention discloses a cardiac ultrasound image recognition method based on GLNet and self-attention, and the method comprises the following steps: introducing Transform on the basis of the GLNet to help a local block to capture a global dependency relationship, including a plurality of GL layers, each GL layer including a GL block and a RepConv, the GL block including a plurality of RepConv and a Transform, modeling global and local information in input features, giving an input tensor, and carrying out self-attention on the input tensor; local space information is coded through n * n convolution, and then feature dimensions are expanded through point-by-point convolution; after training, a projection head is frozen, and different echocardiograms are classified only by using an encoder GLNet and representation. According to the method, the mixed CNN-Transform architecture GLNet is designed, and the problems that the global representation of the CNN is weak and the Transform lacks local information are sol