搜索跟踪转台高分辨率图像的小目标检测技术研究  

Research on Object Detection Technology of High Resolution Image of Photoelectric Turntable

作　　者：黄玺 吕耀文[1] HUANG Xi;LV Yaowen(School of Opto-Electronic Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学光电工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2024年第5期22-29,共8页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技厅自然科学基金(YDZJ202201ZYTS419)。

摘　　要：针对搜索跟踪转台高分辨率图像的要求,结合搜索跟踪转台的运动特点,提出了一种基于背景建模和YOLOv7的图像目标检测方法。首先,根据转台的运动参数,将提前采样的背景图像,由投影透视校正得到连续的背景图像,进而由背景差分法得到运动前景目标;其次,以YOLOv7作为该算法的目标检测器,结合转台背景信息半固定的特点,将背景图像作为训练数据进一步训练得到优化的网络参数;最后,搭建实验平台,建立了行人检测实验数据集。实验结果表明,本模型相比于原YOLOv7模型的召回率提升了6.56%,精确率提升了5.36%,可有效应用于高分辨率搜索跟踪转台的目标检测任务中。According to the requirements of high-resolution images of search tracking turntable,combined with the motion characteristics of search tracking turntable,this paper proposes an image object detection method based on background modeling and YOLOv7.Firstly,according to the motion parameters of the turntable,the pre-sampling background image is corrected by projection perspective to obtain a continuous background image,and then the moving foreground object is obtained by the background difference method.Secondly,taking YOLOv7 as the object detector of the algorithm,combined with the characteristics of semi-fixed background information of the turntable,the background image is used as training data to further train the optimized network parameters.Finally,an experimental platform is set up and an experimental data set is established.Experimental results show that compared with the original YOLOv7 model,the recall rate of the proposed model is improved by 6.56%,and the accuracy is improved by 5.36%.It can be effectively applied to the object detection task of high-resolution search tracking turntable.

关 键 词：目标检测 搜索跟踪转台 YOLOv7 背景差分 

分 类 号：O482.3[理学—固体物理]