AI图像识别下的木材表面缺陷结构光视觉检测方法  

Structured light visual detection method for wood surface defects under AI image recognition

作　　者：曾惠霞 许龙铭 肖平 ZENG Huixia;XU Longming;XIAO Ping(Guangzhou City Institute of Technology,Guangzhou 510800,China)

机构地区：[1]广州城市理工学院,广州510800

出　　处：《激光杂志》2023年第7期223-228,共6页Laser Journal

基　　金：广东省科技创新战略专项资金资助项目(No.pdjh2022b0759)。

摘　　要：针对木材表面视觉特征复杂,嵌补平整过程不能适应木材以及节子的多样性变化的问题,提出基于AI图像识别下的木材表面缺陷结构光视觉检测方法。对结构光视觉技术加以分析并标定光平面,采用Faster-RCNN初步检测木材表面缺陷,定位其缺陷边框并识别缺陷类型,再通过非局部均值滤波法和改进的MSRCR处理图像,精细化分割区域图像,利用改进的最小二乘椭圆拟合方法生成最优椭圆轮廓,实现木材表面缺陷结构光视觉检测。实验结果表明,所提方法的结构光视觉图像预处理效果更好,召回率在95%以上,识别准确率最高可达99%左右,分类准确率最高可达97%左右,检测所用时间更短。Aiming at the problem that the visual characteristics of wood surface are complex and the process of inlay and leveling cannot adapt to the diversity of wood and knots,a structural light visual detection method of wood surface defects based on AI image recognition is proposed.The structural light vision technology is analyzed and the light plane is marked.The fast RCNN is used to preliminarily detect the wood surface defects,locate the defect frame and identify the defect type.Then the image is processed by the non local mean filtering method and the improved msrcr,the regional image is finely segmented,and the improved least square ellipse fitting method is used to generate the optimal ellipse contour to realize the structural light vision detection of wood surface defects.The experimental results show that the proposed method has better preprocessing effect of structured light vision image,with a recall rate of more than 95%,a recognition accuracy of up to 99%,a classification accuracy of up to 97%,and a shorter detection time.

关 键 词：AI图像识别 木材表面缺陷 结构光视觉 Faster-RCNN 最小二乘椭圆拟合 

分 类 号：TN247[电子电信—物理电子学]