【摘要】 针对人工检测中存在的漏检、误判、成本偏高等问题,提出一种采用改进决策树结合前馈神经网络(FFN)选择最优分类属性的方法,实现了皮革表面缺陷自动检测分类;有效解决神经网络分类处理时间长及"黑盒"性弊端,决策树构建、剪枝计算量大,寻找最优树难等问题;实验表明对于各种缺陷的正确识别率均高于90%,该方法稳定可靠,检测精度较高,可满足实际生产需要。更多还原

【Abstract】 In allusion to artificial problem such as undetected,misjudgment,high cost,this paper presents a new method that select optimal attribute by Feed-forward Neural Networks(FNN),combined with improved decision tree,which effectively achieved leather surface defect inspection and classification.This method also solve disadvantages of neural network and decision tree,like long processing time,shortcomings ’Black box’,large calculation of construction and pruning,and difficult to find optimal tree.Experiments show that the recognition rate for various defects was higher than 90%.The method is stable and reliable detection of high precision,could meet the needs of the actual production.更多还原