基于BP神经网络的双能X射线透射的金属识别算法

Metal Identification Algorithm Based on BP Neural Network for Dual Energy X-ray Transmission

双能X射线透射识别技术可识别物质种类,是一种能用于废金属回收的新方法。在废金属识别算法中,目前的曲线拟合识别算法只能在物质厚度较小时能较好地拟合,识别准确性和厚度范围不能满足废金属分选要求,并且不能解决X射线源扇形效应的影响。针对现有算法的不足,基于α曲线提出α识别特征,提高了识别的厚度范围;提出将物料位置作为识别特征,解决了扇形效应;结合α特征和位置特征提出基于BP神经网络算法的废金属分类模型。通过铜和铝物料实验对比,结果表明识别准确率从81.4%提高到了94%。

Dual-energy X-ray transmission identification technology can identify the types of substances and is a new method that can be used for scrap metal recycling.In the scrap metal recognition algorithm,the current curve fitting recognition algorithm can only fit well when the thickness of the material is small,the recognition accuracy and thickness range cannot meet the requirements of waste metal sorting,and it cannot solve the fan effect of the X-ray source impact.Aiming at the shortcomings of the existing algorithms,the recognition features based on the curve are proposed to increase the thickness range of recognition;the material position is used as the recognition feature to solve the fan-shaped effect;the feature and position feature are combined to propose a waste metal classification model based on BP neural network algorithm.The experimental comparison between copper and aluminum materials shows that the recognition accuracy has been improved from 81.4%to 94%.