Method for improving detection network model based on addition of attention mechanism

The invention discloses a method for improving a detection network model based on addition of an attention mechanism, and the method comprises the steps: S10, decomposing an image through a global average pooling layer, carrying out the compression coding of the image in a horizontal coordinate direction and a vertical coordinate direction, and generating two image feature codes in different coordinate directions; s20, performing aggregation in a convolution mode to generate an intermediate feature map f; s30, segmenting the f based on the spatial dimension to obtain two independent vectors, and then converting the sizes of the two vectors by using two different convolution kernels to obtain the same number of channels as the beginning input; and S40, processing the two vectors gh and gw after size conversion to obtain an attention weight containing position information, and then acting the attention weight on an original image feature map. After the attention module is used for improving the original network