首页|基于近红外高光谱成像技术的塑料分类(特邀)

基于近红外高光谱成像技术的塑料分类(特邀)

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
塑料因其可塑性与低成本在日常生活与工业中被广泛使用,然而这也带来环境污染与资源浪费等问题,因此塑料分类成为重要研究课题.为验证高光谱成像技术在塑料分类中的可行性,采用近红外高光谱成像技术(NIR-HSI),比较了1100～1650 nm波段数据在9种常见塑料分类中的效果.涵盖K邻近法(K-NN)、支持向量机(SVM)、粒子群算法训练的SVM(PSO-SVM)、遗传算法优化的SVM(GA-SVM)等机器学习方法.通过验证数据筛选模型准确率后,将其应用于高光谱图像,通过可视化分类对比原始图像评估模型效果.结果显示,基于欧氏距离、余弦相似度的K-NN和GA-SVM分类效果最佳,验证数据的精度分别达到96.14%、96.21%和98.67%,在可视化分类上也呈现出良好效果.高光谱成像技术在塑料分选中具有很高的应用价值,只需获取特定塑料的光谱数据并进行适当处理,即可对不同颜色、形状、工艺的同类塑料制品进行有效区分.
Classification of Plastics Based on Near-Infrared Hyperspectral Imaging Technology(Invited)
Plastics are widely used in daily life and industry because of their plasticity and low costs.However,they cause problems,such as environmental pollution and resource waste,and plastic classification has become an important research topic.Near-infrared hyperspectral imaging(NIR-HSI)is used to compare the effect of 1100‒1650 nm band data in classifying nine common plastics to verify the feasibility of hyperspectral imaging in plastic sorting.Machine learning methods such as the K-neighborhood method(K-NN),support vector machine(SVM),SVM trained by particle swarm algorithm(PSO-SVM),and SVM optimized by genetic algorithm(GA-SVM)are used.After verifying the accuracy of the data screening model,it is applied to hyperspectral images,and the model effect is evaluated by comparing the original images through visual classification.The results show that the K-NN and GA-SVM based on the Euclidean distance and cosine similarity are the most effective in classification,and the accuracy of the validation data reaches 96.14%,96.21%,and 98.67%,respectively.Good results are also presented in the visualization classification.The experiment demonstrates that hyperspectral imaging technology has high application value in plastic sorting.This can effectively differentiate similar plastic products based on color,shape,and process by acquiring the spectral data of specific plastics and processing them appropriately.

near-infrared hyperspectral imagingplastic classificationmachine learningvisual classification

胡锡敦、尹禄、杨钦晨、王乐

展开 >

中国计量大学光学与电子科技学院,浙江 杭州 310018

近红外高光谱成像 塑料分类 机器学习 可视化分类

国家自然科学基金国家自然科学基金国家自然科学基金国家自然科学基金国家重点研发计划浙江省科技计划浙江省科技计划浙江省自然科学基金

51832005620752036230532012100420182021YFC33404002022C011272021C05005LQ23A040007

2024

10.3788/LOP232402

激光与光电子学进展
中国科学院上海光学精密机械研究所

激光与光电子学进展

CSTPCD北大核心
影响因子：1.153
ISSN：1006-4125
年,卷(期)：2024.61(2)
  • 18