LIBS结合GA-BP神经网络与GA-SVM对塑料分类研究

Plastic classification research by LIBS combined with GA-BP-NN and GA-SVM

塑料制品回收力度小、重复利用率低,造成环境污染和资源浪费,因此对废旧塑料精确分类是提高塑料回收的关键。本文采用激光诱导击穿光谱(LIBS)技术结合遗传算法优化误差反向传播神经网络(GA-BP)和支持向量机(GA-SVM)对常见的10种塑料进行分类识别。利用LIBS技术对塑料样品进行处理,分别采集每种塑料样品100组光谱。对采集到的原始光谱做滤波和归一化处理,提取光谱中14条主要的特征谱线,分别建立GA-BP神经网络和GA-SVM模型。实验结果表明,GA-BP神经网络对塑料的识别性能优于GA-SVM,其中GA-BP神经网络识别精度为99.25%,原因是GA-SVM利用升维算法实现对数据集的分类,在塑料样品种类多的情况,分类效果不及GA-BP神经网络。因此,利用LIBS技术结合不同的识别算法,可以实现对多种塑料样品的分类,也为研究不同算法对塑料样品分类识别提供研究思路。

Low recycling and reuse rate of plastic products cause environmental pollution and resource waste.Therefore,accurate classification of waste plastics is the key to improving plastic recycling.In this paper,laser-induced breakdown spectroscopy(LIBS)technology combined with genetic algorithm optimization error back propagation neural network(GA-BP-NN)and support vector machine(GA-SVM)are used to classify and identify 10 common plastics.The plastic samples are processed by LIBS technology,and 100 sets of spectra of each plastic sample are collected.After filtering and normalizing the collected original spectrum,14 main spectral in the spectrum are extracted as characteristic spectral,and GA-BP neural network and GA-SVM models are established respectively.The experimental results show that the GA-BP neural network model is better than the GA-SVM model in identifying plastics.The GA-BP neural network has a recognition accuracy of 99.25%,because GA-SVM uses the ascending algorithm to classify the data set.In the case of a large number of plastic samples,the classification effect is not as good as the GA-BP neural network.Therefore,the use of LIBS technology combined with different recognition algorithms can realize the classification of a variety of plastic samples,and also provide research ideas for the classification and recognition of plastic samples by different algorithms.