Aerial remote sensing image recognition method based on multi-layer and regional feature fusion

Abstract

The present invention discloses an aerial remote sensing image recognition method based on multi-layer and regional feature fusion, which belongs to the field of fine-grained image recognition. The backbone network Resnet‑50 is adopted, and targeted improvements are made according to the characteristics of aerial remote sensing images, such as fast update speed, blurred image details, and low resolution. The present invention adopts a new multi-layer fusion and regional feature fusion method, and the multi-layer fusion is accompanied by attention weights, giving the model the ability to autonomously allocate fusion weights, making the selection of details in the shallow feature map flexible, better retaining the use of local detail feature information of remote sensing images, and the extraction strategy of the region of interest reduces the computational overhead and parameter amount of regional feature fusion.

Images