An anchor-free target detection method for remote sensing images

Abstract

The present invention provides a remote sensing image target detection method based on an anchor-free frame, comprising the following steps: establishing a remote sensing image target detection model without an anchor frame, including: a feature extraction network, a feature pyramid, and an anchor-free frame detector; acquiring a remote sensing image, and Carry out sliding partitions to obtain small-size images; input the small-size images into the feature extraction network, and use the feature pyramid structure to obtain feature maps of three different scales; input the feature maps into the anchor-free frame detector for prediction, Obtain the target prediction result; design the loss function of multi-index fusion, use the target prediction result and the loss function of multi-index fusion to complete the training of the target detection model, and obtain the trained target detection model; use the trained target detection model to perform remote sensing image target detection. The beneficial effect provided by the invention is that a remote sensing image target detection model with less hyperparameters, low model complexity and high detection accuracy is established.

Description

technical field [0001] The invention relates to the field of target detection, in particular to a remote sensing image target detection method without anchor frame. Background technique [0002] Most of the existing remote sensing image target detection methods use two methods: [0003] (1) Use the anchor frame to predict the remote sensing image target, match the anchor frame with the real frame, and select the anchor frame with the highest matching degree as the final prediction result; [0004] (2) Use the horizontal frame to mark the target of the remote sensing image, and its expression is the coordinates of the center point and the length and width. [0005] The existing pre-defined anchor box detectors mainly use the generated anchor boxes and offsets to match the target real boxes, and finally predict the bounding box as the anchor box with the highest matching degree plus the offset of the center point. However, on remote sensing images, using anchor boxes to dete...

AI Technical Summary

Problems solved by technology

[0006] In view of this, the present invention mainly solves the problem of too many hyperparameters related to the anchor frame and the multi-direction of the target in the remote sensing image target detection, and proposes a remote sensing image target detection method based on no anchor frame

Images