基于剪枝决策树的人造板表面缺陷识别
|
摘要
连续压机生产线的发展,使人造板实现自动化生产,但缺陷检测环节仍为人工.缺陷识别是检测中的一个重要环节,是根据缺陷特征值利用分类器进行识别的过程.由于人造板连续生产,实时性要求高,为实现缺陷的快速、准确识别,提出了一种基于剪枝的CART树对人造板进行缺陷识别.通过对已有的人造板缺陷图像进行预处理、分割,获得缺陷的形状、纹理特征作为输入,通过基于Gini指数的CART算法生成CART树.针对于自由生长的CART树容易出现过拟合的问题,利用代价复杂度算法对生成的CART树进行剪枝,通过十折交叉验证对剪枝前后的子树进行比较,获得最优子树.通过实验证明剪枝后的CART树缺陷识别正确率高达93%,满足人造板缺陷识别的实时性和正确率的要求,可以实现人造板在线缺陷检测.
The automatic production of wood-based panel has been realized with the development of continuous press production line, but the defect detection is still manual. As an important part of detection, defect recognition is a process of using a classifier to identify defects based on feature value. For the reason of the continuous production of wood-based panels, the defects need to be identified quickly and accurately. Therefore, a cart tree is proposed to identify the defects of the wood-based panel in this study. The defect features of shape and texture are firstly obtained using image preprocessing and image segmentation, and then the cart tree is generated by Gini exponent, at last defects are identified by using the cart tree. But it is easy to cause the problem of overfitting using cart tree without pruning, so the study obtains the optimal subtree by using the cost complexity algorithm and 10 cross-validations. The experimental results reflect that the accuracy rate of defect recognition reaches 93% with the proposed cart tree, which can satisfy the requirements of real-time and accuracy on defect identification.
The automatic production of wood-based panel has been realized with the development of continuous press production line, but the defect detection is still manual. As an important part of detection, defect recognition is a process of using a classifier to identify defects based on feature value. For the reason of the continuous production of wood-based panels, the defects need to be identified quickly and accurately. Therefore, a cart tree is proposed to identify the defects of the wood-based panel in this study. The defect features of shape and texture are firstly obtained using image preprocessing and image segmentation, and then the cart tree is generated by Gini exponent, at last defects are identified by using the cart tree. But it is easy to cause the problem of overfitting using cart tree without pruning, so the study obtains the optimal subtree by using the cost complexity algorithm and 10 cross-validations. The experimental results reflect that the accuracy rate of defect recognition reaches 93% with the proposed cart tree, which can satisfy the requirements of real-time and accuracy on defect identification.
引文
1张国梁,瞿国富,侯晓鹏,等.连续平压机入口角度动态控制方法.农业机械学报,2015,46(4):372-378,343.
2杜超,王志海,江晶晶,等.基于显露模式的数据流贝叶斯分类算法.软件学报,2017,28(11):2891-2904.[doi:10.13328/j.cnki.jos.005350]
3 Mu HH,Zhang MM,Qi DW,et al.The application of RBFneural network in the wood defect detection.International Journal of Hybrid Information Technology,2015,8(2):41-50.[doi:10.14257/ijhit]
4梁浩,曹军,林雪,等.基于贝叶斯神经网络的近红外光谱实木地板表面缺陷检测.光谱学与光谱分析,2017,37(7):2041-2045.
5 Tan M,Hu ZF,Wang BY,et al.Robust object recognition via weakly supervised metric and template learning.Neurocomputing,2016,181:96-107.[doi:10.1016/j.neucom.2015.04.123]
6 Fan MY,Wei L,He ZZ,et al.Defect inspection of solder bumps using the scanning acoustic microscopy and fuzzy SVM algorithm.Microelectronics Reliability,2016,65:192-197.[doi:10.1016/j.microrel.2016.08.010]
7黄新波,李文君子,宋桐,等.采用遗传算法优化装袋分类回归树组合算法的变压器故障诊断.高电压技术,2016,42(5):1617-1623.
8常辉,胡修林,张蕴玉.基于CART算法的卫星星座原始构型选择策略.华中科技大学学报(自然科学版),2011,39(6):1-5.
9张亮,宁芊.CART决策树的两种改进及应用.计算机工程与设计,2015,36(5):1209-1213.
10 Cui WC,Chen S,Yu TS,et al.Feature extraction of X-ray chest image based on KPCA.Proceedings of the 2nd International Conference on Computer Science and Network Technology.Changchun,China.2013.1263-1266.
11杜春蕾,张雪英,李凤莲.改进的CART算法在煤层底板突水预测中的应用.工矿自动化,2014,40(12):52-56.
12 Mburu JW,Kingwara L,Ester M,et al.Use of classification and regression tree(CART),to identify hemoglobin A1C(HbA1C)cut-off thresholds predictive of poor tuberculosis treatment outcomes and associated risk factors.Journal of Clinical Tuberculosis and Other Mycobacterial Diseases,2018,11:10-16.[doi:10.1016/j.jctube.2018.01.002]
13杨柳,王钰.泛化误差的各种交叉验证估计方法综述.计算机应用研究,2015,32(5):1287-1290,1297.[doi:10.3969/j.issn.1001-3695.2015.05.002]
2杜超,王志海,江晶晶,等.基于显露模式的数据流贝叶斯分类算法.软件学报,2017,28(11):2891-2904.[doi:10.13328/j.cnki.jos.005350]
3 Mu HH,Zhang MM,Qi DW,et al.The application of RBFneural network in the wood defect detection.International Journal of Hybrid Information Technology,2015,8(2):41-50.[doi:10.14257/ijhit]
4梁浩,曹军,林雪,等.基于贝叶斯神经网络的近红外光谱实木地板表面缺陷检测.光谱学与光谱分析,2017,37(7):2041-2045.
5 Tan M,Hu ZF,Wang BY,et al.Robust object recognition via weakly supervised metric and template learning.Neurocomputing,2016,181:96-107.[doi:10.1016/j.neucom.2015.04.123]
6 Fan MY,Wei L,He ZZ,et al.Defect inspection of solder bumps using the scanning acoustic microscopy and fuzzy SVM algorithm.Microelectronics Reliability,2016,65:192-197.[doi:10.1016/j.microrel.2016.08.010]
7黄新波,李文君子,宋桐,等.采用遗传算法优化装袋分类回归树组合算法的变压器故障诊断.高电压技术,2016,42(5):1617-1623.
8常辉,胡修林,张蕴玉.基于CART算法的卫星星座原始构型选择策略.华中科技大学学报(自然科学版),2011,39(6):1-5.
9张亮,宁芊.CART决策树的两种改进及应用.计算机工程与设计,2015,36(5):1209-1213.
10 Cui WC,Chen S,Yu TS,et al.Feature extraction of X-ray chest image based on KPCA.Proceedings of the 2nd International Conference on Computer Science and Network Technology.Changchun,China.2013.1263-1266.
11杜春蕾,张雪英,李凤莲.改进的CART算法在煤层底板突水预测中的应用.工矿自动化,2014,40(12):52-56.
12 Mburu JW,Kingwara L,Ester M,et al.Use of classification and regression tree(CART),to identify hemoglobin A1C(HbA1C)cut-off thresholds predictive of poor tuberculosis treatment outcomes and associated risk factors.Journal of Clinical Tuberculosis and Other Mycobacterial Diseases,2018,11:10-16.[doi:10.1016/j.jctube.2018.01.002]
13杨柳,王钰.泛化误差的各种交叉验证估计方法综述.计算机应用研究,2015,32(5):1287-1290,1297.[doi:10.3969/j.issn.1001-3695.2015.05.002]