Small sample self-coding neural network model method based on fusion spatial relationship

The invention discloses a small sample self-encoding neural network model method based on a fusion spatial relationship, and the method comprises the steps: collecting an image F1, and inputting the image F1 to a data input layer; the convolution layer 1 carries out convolution processing on the image F1 passing through the data input layer to obtain an image F2; performing feature extraction on the image F2 through an inverse residual convolutional layer to obtain a feature map F3; the SE inverse residual convolutional layer performs feature extraction on the feature map F3 and outputs a feature map F4; and the feature map F4 is fused with the spatial relationship through a spatial relationship self-coding module and is subjected to self-coding to obtain a feature map F5, and the feature map F5 is output to a coding space. According to the invention, a spatial relation self-coding module is fused in the small sample feature extraction network, a spatial position self-matching coding algorithm is designed, an