基于图像增强技术的运动目标检测算法  

Moving object detection algorithm based on image enhancement technology

作　　者：李树壮 祖国明 李余光 翟双[1] LI Shuzhuang;ZU Guoming;LI Yuguang;ZHAI Shuang(School of Computer Science&Engineering,Changchun University of Technology,Changchun 130102,China)

机构地区：[1]长春工业大学计算机科学与工程学院,吉林长春130102

出　　处：《长春工业大学学报》2024年第1期73-80,共8页Journal of Changchun University of Technology

基　　金：吉林省教育厅基金项目(JJKH20210739KJ)。

摘　　要：利用多权值联合损失函数进行图像细节恢复来解决图像对比度不高导致的细节模糊问题。针对光流网络在图像特征提取上存在输入参数多的问题,提出空洞残差网络作为图像的特征提取模块,通过增大卷积核的感受野提高特征提取速度。使用COCO数据集进行训练和测试。实验结果表明,与已有同类算法对比,文中算法得到的光流场图像有更高的清晰度,有效提高了检测精确度,降低特征提取时间。The multi-weight joint loss function is used to restore image details to solve the problem of blurred details caused by low image contrast;Aiming at the problem that the optical flow network has many input parameters in image feature extraction,a hollow residual network is proposed as the feature extraction module of the image,and the feature extraction speed is improved by increasing the receptive field of the convolution kernel.In this paper,the COCO data set is used for training and testing.The experimental results show that compared with the existing similar algorithms,the optical flow field image obtained by the algorithm in this paper has higher definition,which effectively improves the detection accuracy and reduces the feature extraction time.

关 键 词：图像增强 目标检测 空洞残差网络 联合损失函数 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]