基于非均匀采样和超分辨率重建的小目标检测方法及系统
By combining super-resolution reconstruction and instance-level non-uniform sampling with a collaborative mechanism of reversible feature anti-distortion, the problems of feature loss and semantic inconsistency in small object detection are solved, thereby improving the efficiency and accuracy of small object detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INSPUR GENERSOFT CO LTD
- Filing Date
- 2025-09-05
- Publication Date
- 2026-07-17
AI Technical Summary
In existing small target detection methods, uniform downsampling leads to the loss of small target features, non-uniform sampling methods fail to effectively solve the problems of target scale differences and semantic inconsistencies, and super-resolution reconstruction lacks effective means of utilizing information.
By constructing a collaborative mechanism of super-resolution reconstruction, instance-level non-uniform sampling, and reversible feature anti-distortion, and by using instance-level saliency calculation and separable non-uniform sampling transformation formula, a distorted sampling grid is generated. Combined with reversible inverse transformation to recover the feature image, the detection of small targets is enhanced.
It effectively restores the texture features of small targets, ensures spatial consistency between the detection results and the original labels, and improves the accuracy and efficiency of small target detection.
Smart Images
Figure CN121033396B_ABST