CPC G16H 30/40 (2018.01) [G06F 18/251 (2023.01); G06F 18/253 (2023.01); G06N 3/045 (2023.01); G06V 10/764 (2022.01); G06V 10/806 (2022.01); G06V 10/82 (2022.01); G16H 50/20 (2018.01)] | 15 Claims |
1. A method for fusing features applied to small target detection, the method comprising:
acquiring feature maps output by convolutional layers in a Backbone network;
performing a convolution on the feature maps to obtain input feature maps of feature layers, the feature layers representing resolutions of the input feature maps; and
fusing, based on densely connection feature pyramid network features, the input feature maps of each feature layer to obtain output feature maps of the feature layers, the fusing comprising:
sampling an input feature map of an ith feature layer and input feature maps of other feature layers having resolutions lower than a resolution corresponding to the ith feature layer, i being a positive integer;
scaling, after the sampling, the input feature maps of the other feature layers to a same size as the input feature map of the ith feature layer; and
superimposing the input feature maps of the other feature layers after scaling and the input feature map of the ith feature layer one by one, and using a superimposed final result as an output feature map of the ith feature layer.
|