红外小目标伪标签生成方法、系统、存储介质及设备
By using a centroid offset calibration network and a pseudo-label image fine generation network in infrared weak target detection, high-precision pseudo-label images are automatically generated, solving the problems of low efficiency and insufficient accuracy of traditional annotation methods and improving the performance of infrared image analysis tasks.
CN118736290BActive Publication Date: 2026-07-17XIDIAN UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2024-06-19
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In infrared detection of small targets, traditional manual labeling methods are inefficient and lack accuracy. How to automatically generate high-precision and high-efficiency pseudo-label images has become an urgent problem to be solved.
Method used
By acquiring infrared images and manually labeled images, the target centroid and coarse label images are extracted. After processing with Gaussian blur kernel, a centroid offset calibration network and a pseudo-label image fine generation network are trained to gradually generate high-precision pseudo-label images.
Benefits of technology
It improves the accuracy and detail of pseudo-labeled images, thereby enhancing the performance of infrared image analysis tasks, reducing annotation complexity, and increasing efficiency.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN118736290B_ABST
Abstract
本发明实施例公开了一种红外小目标伪标签生成方法，该方法包括：采集红外图像和与其对应的人工标签图像；从人工标签图像中提取目标形心标签图像和粗标签图像，并使用高斯模糊核处理二者，获取模糊形心标签图像和模糊粗标签图像；根据红外图像和模糊粗标签图像训练形心偏移校准网络直至收敛；预设该网络的权值，并使用形心偏移校准网络确定中心区域预测结果；获取该结果的形心，并根据高斯模糊核得到模糊形心；根据红外图像和模糊形心训练伪标签图像精细生成网络直至收敛；根据伪标签图像精细生成网络确定红外图像对应的最终伪标签图像。本发明通过形心偏移校准和伪标签图像精细生成，提高伪标签图像准确度和精细度，有利于红外图像分析。
Need to check novelty before this filing date? Find Prior Art