RGB-T image feature fusion model construction method and system based on frequency domain enhancement

The invention provides an RGB-T image feature fusion model construction method and system based on frequency domain enhancement, and the method comprises the steps: carrying out the down-sampling of an original image through an image frequency domain information enhancement module, and carrying out the discrete cosine transformation, image high-frequency information enhancement, discrete cosine inverse transformation and image up-sampling of the image after the down-sampling; an image feature fusion module is used to extract image features in a blocking manner, a convolutional neural network is used to carry out dimension transformation on the image features, an attention mechanism is used to process the image features, and RGB and infrared image feature fusion is carried out. The technical problems that frequency domain information is not fully applied, downstream tasks cannot guide fusion weights, feature fusion parameters cannot be adaptively adjusted, and image features are not rich enough are solved. 本发明