反射光干扰激光图像的多层次特征点识别研究  

Multi-level feature point recognition of laser images with reflected light interference

作　　者：林苏美 胡克用[2] 时月梅 LIN Sumei;HU Keyong;SHI Yuemei(Chengdu University of Arts and Sciences,Chengdu 610401,China;Hangzhou Normal University,Hangzhou 310018,China)

机构地区：[1]成都文理学院,成都610401 [2]杭州师范大学,杭州310018

出　　处：《激光杂志》2024年第9期164-170,共7页Laser Journal

基　　金：浙江省自然科学基金联合基金资助项目(No.LHY21E090004)。

摘　　要：由于反射光干扰激光图像复杂性较强,噪声较多,导致反射光干扰激光图像多层次特征点识别精度和效率较低。为此,提出了一种反射光干扰激光图像的多层次特征点识别研究。根据反射光干扰激光图像的特点计算相邻点的斜率,筛选有效激光数据。采用数据曲率算法改变反射光干扰激光图像之间的距离,剔除有效反射光干扰激光图像噪声,完成滤波平滑处理。利用Mask R-CNN对滤波处理后的反射光干扰激光图像有效激光数据噪声进行检测,通过损失函数提取Mask R-CNN不同层级的特征点信息,以此完成反射光干扰激光图像多层次特征点识别。测试结果表明,所提方法的多层次特征点识别结果较好,可以识别出全部的多层次特征点,有效激光数据筛选最大误判率为0.4%,多层次特征点识别率高于95.25%,最高识别时间为22 ms。Due to the strong complexity and high noise of reflected light interference in laser images,the accuracy and efficiency of multi-level feature point recognition in laser images are low.Therefore,a multi-level feature point recognition study for laser images with reflected light interference is proposed.Calculate the slope of adjacent points based on the characteristics of reflected light interfering with laser images,and screen effective laser data.Using data curvature algorithm to change the distance between reflected light interfering laser images,eliminate effective reflected light interfering laser image noise,and complete filtering and smoothing processing.Using Mask R-CNN to detect the effective laser data noise of filtered reflected light interference laser images,the loss function is used to extract feature point information at different levels of Mask R-CNN,thereby completing multi-level feature point recognition of reflected light interference laser images.The test results show that the proposed method has good multi-level feature point recognition results,and can recognize all multi-level feature points.The maximum misjudgment rate for effective laser data filtering is 0.4%,and the multi-level feature point recognition rate is higher than 95.25%.The maximum recognition time is 22 ms.

关 键 词：反射光干扰 激光图像 Mask R-CNN 特征点识别 多层次特征 

分 类 号：TN249[电子电信—物理电子学]