摘要：The traditional computed tomography(CT)reconstruction methods are noisy,low resolution,poor contrast,and generally not suitable to detect the smaller ***,the filter design is also *** CT characteristics reconstruction technology was brought forward to improve in these aspects,which is defined to directly reconstruct the characteristics of the projection for the best requirements not the overall image *** two-dimension(2D)and three-dimension(3D)CT characteristics reconstruction algorithm were firstly introduced,then by detailed analysis,experimental results and comparsion of parameters calculated,its advantages in keeping better high-frequency feature,better noise immunity,short time-consuming and easier design are verified.

摘要：X射线成像是解决复杂构件内部零件装配正确性识别的最有效方法,现有特征识别方法是以图像中的连通区域形状、长宽比、面积等特征为目标.虽然检测效果较好,但受机械精度、装配公差、零件错位等因素影响,识别鲁棒性较差.基于此,综合卷积神经网络目标分类识别技术与X射线多视角成像技术,首先设计了一个深度卷积神经网络模型,通过深度学习的方法提取特征、训练分类器,对工件内部零件进行分类,输出坐标框,完成工件零件漏装检测.进而针对所识别零件坐标信息;依据CT投影正弦特性,找到与当前检测工件投影角度相符的标准工件投影,完成零件换位、错位等识别.通过实验验证,此方法在自行建立的数据集上完成了对工件内部零件漏缺和换位的识别,鲁棒性较高.