The invention relates to the technical field of
crop fine-grained classification, in particular to a
crop fine-grained classification method based on global and local
feature fusion, comprising the following steps: S1, constructing a dual-path classification network which comprises a first
backbone network, a second
backbone network, a
detector, a feature compression module, a
feature fusion module and a classifier, s2, inputting the
crop image into the first
backbone network, and extracting global features of the crop image; and meanwhile, inputting the crop image into a
detector, screening out local target images of which the confidence ranks top k, inputting the local target images into a second backbone network, and extracting local features of the local target images. According to the crop fine-grained classification method based on global and local
feature fusion, feature fusion is used for improving precision, global features are extracted through an original image, local features are obtained through a
detector, the overall structure and detail difference are considered after fusion, and the problem that the intra-class difference of crops is large, and the inter-class difference is small is solved.