一种目标检测方法、装置、存储介质及电子设备

By introducing the dynamic lateral inhibition mechanism of dragonfly compound eyes into the target detection algorithm, and combining the joint inhibition of spatial and channel dimensions, the feature fusion process is optimized, which solves the accuracy bottleneck of small target detection in complex scenes and achieves higher recognition accuracy and robustness.

CN122200205BActive Publication Date: 2026-07-17INSPUR (SHANDONG) COMPUTER TECH CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INSPUR (SHANDONG) COMPUTER TECH CO LTD
Filing Date
2026-05-18
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing target detection algorithms suffer from a bottleneck in the accuracy of small target detection in complex scenes, especially with high false negative and false positive rates under varying illumination and background interference. Furthermore, the lateral inhibition mechanism of existing biological vision systems fails to effectively combine the dynamic characteristics of multi-scale features, leading to noise accumulation and insufficient accuracy.

Method used

A dynamic side-suppression mechanism for dragonfly compound eyes is introduced. By jointly suppressing the spatial and channel dimensions of the backbone network, and combining the cross-scale semantic mask and channel suppression factor of the feature pyramid network, the feature fusion process is optimized to achieve local sharpening and global noise reduction.

Benefits of technology

It improves the target detection accuracy in complex scenarios, reduces noise interference, and enhances the recognition accuracy and robustness of small target detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122200205B_ABST
    Figure CN122200205B_ABST
Patent Text Reader

Abstract

本公开实施例提供了一种目标检测方法、装置、存储介质及电子设备,应用于计算机视觉技术领域,方法包括:获取待检测图像;利用检测模型识别所述待检测图像,得到所述待检测图像的识别结果,所述识别结果包括所述待检测图像中目标对象的类别;其中,所述检测模型至少包括:主干网络、特征金字塔网络;所述主干网络至少用于对待检测图像进行特征图的提取;以及,对所述特征图进行空间维度和通道维度的联合抑制,基于抑制后的特征图进行图像识别;所述特征金字塔网络至少用于利用高层特征对低层特征的非目标区域施加侧抑制,将抑制后的所述低层特征与高层特征融合。如此,实现局部锐化和全局降噪,降低噪声干扰,提高识别精度。
Need to check novelty before this filing date? Find Prior Art