高级检索
    李超, 吕宪伟, 涂文俊, 张怡卓. 基于计算机视觉的实木表面智能化分选系统设计[J]. 北京林业大学学报, 2016, 38(3): 102-109. DOI: 10.13332/j.1000-1522.20150294
    引用本文: 李超, 吕宪伟, 涂文俊, 张怡卓. 基于计算机视觉的实木表面智能化分选系统设计[J]. 北京林业大学学报, 2016, 38(3): 102-109. DOI: 10.13332/j.1000-1522.20150294
    LI Chao, LÜ Xian-wei, TU Wen-jun, ZHANG Yi-zhuo. Design of an intelligent wood surface grading system based on computer vision[J]. Journal of Beijing Forestry University, 2016, 38(3): 102-109. DOI: 10.13332/j.1000-1522.20150294
    Citation: LI Chao, LÜ Xian-wei, TU Wen-jun, ZHANG Yi-zhuo. Design of an intelligent wood surface grading system based on computer vision[J]. Journal of Beijing Forestry University, 2016, 38(3): 102-109. DOI: 10.13332/j.1000-1522.20150294

    基于计算机视觉的实木表面智能化分选系统设计

    Design of an intelligent wood surface grading system based on computer vision

    • 摘要: 设计一种集实木传送、图像定位与采集、实木板材表面识别与分选的智能系统,系统通过传送带运送实木板材,CCD摄像头获取板材图像,在触摸屏工控机TPC700-9190T上应用MFC与OpenCV编写分选程序对板材图像进行分析,识别结果通过STM32单片机控制电磁阀完成实木板材的分类。在图像定位与识别算法中,采用积分投影算法确定板材边界,动态采集板材表面图像;在颜色分类方面,利用L*a*b*空间颜色分量的均值、方差和斜度3个低阶矩表达颜色;在缺陷检测方面,提出了基于纹理填充的缺陷分割方法,通过获取纹理掩膜图像,然后利用板材背景颜色淡化纹理,最后应用加权阈值法完成缺陷分割,分割后计算缺陷面积、边缘灰度均值、内部灰度均值和长宽比等特征表达缺陷信息;在纹理识别方面,提出了基于Contourlet变换的纹理特征提取方法,通过对纹理图像进行Contourlet变换3层分解,得到1个低频子带、6个中频子带和8个高频子带,分别计算低频和中频系数矩阵的均值和方差,并与高频系数矩阵的能量组成22个特征表达纹理信息;最后设计SVM分类器,分别对颜色、缺陷和纹理进行识别。采用300个柞木样本进行实验,板材传送速度在小于1.5 m/s范围内,颜色识别准确率为100%;活节、死结和裂纹识别准确率分别为92.2%、95.6%和93.3%;直纹、弯纹识别准确率分别为93.9%、92.8%。实验结果表明,分选系统具有实时、高效、准确的特点。

       

      Abstract: An intelligent system for wood surface detection is designed, which integrates plate transmission, image acquisition, image recognition and sorting equipment. The convey belt is used to carry plates, CCD is employed to acquire images, the recognition program is written by MFC and Open CV in the touch screen industrial computer, and the solenoid valve is controlled by STM32 according to recognition results. In the image localization process, the integral projection method is used to determine the boundary of the plates. In the color classification, the mean, variance and skewness features of the L*a*b* space are extracted to express color information. In the defect detection, a segmentation method based on texture filling is proposed. The texture of the image is extracted and the background color is used to fade the texture part which can reduce impact of texture’s effect, and then weighted threshold is used to segment the defects. After segmentation, the features of area, the edge gray value, the internal gray value and the length and width ratio are used to express defects. In texture recognition, the texture feature extraction method based on Contourlet transform is proposed. By Contourlet transform, one sub-band, six intermediate frequency sub-bands and eight high frequency sub-bands are obtained. By calculating the mean and variance of the low frequency and intermediate frequency coefficients, a 22-dimension feature vector is obtained with the energy of high frequency coefficient matrix. Finally, a SVM classifier is designed to recognize the color, defect and texture. A total of 300 samples are used in the test experiment, when convey speed is under 1.5 m/s, and the classification rate of color is 100%, the recognition rate of live knot, dead knot and crack are 92.2%, 95.6% and 93.3% respectively, and the recognition rate of radial texture and tangential texture are 93.9% and 92.8% respectively.

       

    /

    返回文章
    返回