Unmanned aerial vehicle target detection method and system based on improved YOLOv12m

By improving the DPCM-YOLO network of YOLOv12m and combining a multidimensional collaborative attention module and a dynamically weighted W-IOU loss function, the problems of low accuracy in small target detection and sensitivity to background interference in UAV target detection are solved, achieving efficient, robust, and lightweight detection.

CN120808208APending Publication Date: 2025-10-17JISHOU UNIVERSITY
0 Cites 2 Cited by

Patent Information

Application Number
CN202510859523.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-25
Publication Date
2025-10-17

AI Technical Summary

Technical Problem

Existing drone target detection technology has low small target detection accuracy in low-altitude complex scenes, is sensitive to background interference, and has an imbalance between model efficiency and accuracy, and lacks a systematic solution.

Method used

We employ a DPCM-YOLO network based on an improved YOLOv12m, combined with a multidimensional collaborative attention module (MCAM) and a dynamically weighted W-IOU loss function. Through a dual-path heterogeneous convolutional structure and residual connection design, we improve feature extraction efficiency and small target detection accuracy.

Benefits of technology

It improves the detection accuracy of small targets, suppresses interference from complex background noise, enhances the robustness and efficiency of the model, and achieves high-precision, low-latency, lightweight detection.

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

Abstract

The invention discloses an unmanned aerial vehicle target detection method and system based on improved YOLOv12m, and belongs to the technical field of computer vision and deep learning. The method comprises the following steps: designing a dual-path context module DPCM to replace an A2C2f module in an original model; a double-path heterogeneous convolution structure is adopted in Aera Attention in a module attention branch ABlock to replace traditional 7 * 7 convolution, and the small target feature expression ability is enhanced; an improved residual multidimensional cooperative attention module is embedded in a feature extraction branch, residual features are fused through a channel-height-width three-direction feature extraction branch, and background noise is suppressed; a dynamic weighting W-IOU loss function is introduced into a model detection head, and the regression weight is adaptively adjusted based on the target size, so that the core problems of low small target detection precision, background interference sensitivity, model efficiency-precision imbalance and the like in the prior art can be solved.
Need to check novelty before this filing date? Find Prior Art