Improved unmanned aerial vehicle small target detection method based on YOLO26s

By improving the backbone and neck network structures of the YOLO26s model, introducing windmill-shaped wavelet transform convolution, A2C2f modules, and DySample dynamic upsampling, and combining them with a context anchor residual attention feature fusion module, the problem of limited small target detection performance of YOLO26s in UAV scenarios was solved, achieving efficient and lightweight detection results.

CN121904640AActive Publication Date: 2026-04-21QUANZHOU INST OF EQUIP MFG +1
4 Cites 1 Cited by

Patent Information

Application Number
CN202610330787.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-18
Publication Date
2026-04-21
Estimated Expiration
2046-03-18

AI Technical Summary

Technical Problem

Existing YOLO26s models face limitations in small target detection performance in drone scenarios. Especially under the combined effects of complex background interference, extremely low target pixel ratio, and insufficient image resolution, it is difficult to achieve accurate recognition and localization. Furthermore, while lightweight convolutional solutions reduce computational load, they also lead to a significant decline in detection accuracy, making them unsuitable for the computational resource constraints of drones.

Method used

By introducing a windmill-shaped wavelet transform convolution module into the backbone network, an A2C2f module into the neck network, and DySample dynamic upsampling, and setting skip connections between the backbone and neck networks, combined with a context anchor residual attention feature fusion module, the feature extraction and fusion path is optimized, thereby enhancing the detection capability of small targets.

Benefits of technology

It significantly improves the accuracy and robustness of small target detection, reduces the computational complexity of the model, adapts to the computational resource limitations of UAV platforms, and achieves efficient small target detection.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the field of small target detection, in particular to an improved unmanned aerial vehicle small target detection method based on YOLO26s, and the method comprises the steps: constructing a YOLO26s-SOD model for the small target detection of an unmanned aerial vehicle, replacing a first C3k2 module and a second C3k2 module of an original backbone network with a context anchor point residual attention feature fusion module, replacing a third C3k2 module with an A2C2f module, and carrying out the detection of the small target of the unmanned aerial vehicle. Deleting modules behind the windmill-shaped wavelet transform convolution module and connecting the modules with a first convolution module; a first Upsample module in an original neck network is replaced by a first connection module, a context anchor point residual attention feature fusion module and a DySample module which are connected in sequence, a second Upsample module is replaced by the DySample module, and a C3k2 module in the neck network is replaced by the context anchor point residual attention feature fusion module. A DySample module, a second connection module and a context anchor point residual attention feature fusion module which are connected in sequence are added behind the second C3k2 module; and the space focusing capability on a tiny target is obviously improved.
Need to check novelty before this filing date? Find Prior Art