基于鞍山式铁矿成像光谱的融合算法研究

Fusion Algorithm Research Based on Imaging Spectrum of Anshan Iron Ore

铁矿资源是我国经济发展和社会进步的物质基础。在铁矿开采过程中,快速精准地确定铁矿品位,对矿山开采决策及经济效益具有重要影响。高光谱成像技术具有影像覆盖范围广、精度高等优势,已广泛应用于矿石分类及成分反演等领域。然而目前高光谱成像传感器的波段范围主要为可见短近红外(Vis-SWIR)和近红外(NIR)两类,且两类数据多为独立获取,缺乏连续性,采用单一数据所建模型的精度往往偏低。因此融合多传感器所获光谱数据,可有效解决单一传感器波段范围小、包含目标特征波段少等问题,提高基于高光谱成像技术的铁矿品位反演精度。使用Pika L与Pika NIR-320高光谱成像仪,分别在Vis-SWIR与NIR两个波段范围内采集鞍山式铁矿的成像光谱数据,提出了基于互信息(MI)的光谱串联融合方法,该方法首先对两组光谱数据进行预处理,然后对处理后的数据进行互信息计算以此对光谱数据进行串联融合。最后分别以Vis-SWIR、NIR以及基于不同波段串联融合的光谱数据为数据源,建立RBF神经网络品位反演模型,并以融合前后光谱数据所建模型的准确性与精度为融合算法有效性的判别指标。结果表明,光谱数据串联融合后所建模型的准确性与精度高于单独使用Vis-SWIR、NIR光谱数据所建模型的准确性与精度。与基于其余波段串联融合的光谱数据相比,在基于互信息计算得出的959.89nm处串联融合后光谱数据所建模型的准确性与精度最高,R2为0.88,RPD为2.97,RMSE为4.464,MAE为3.32。该研究针对多传感器光谱融合提出了一种新思路,对成像光谱技术应用于铁矿品位反演具有现实指导意义。

Iron ore resources are the foundation of our economic development and social progress.In the process of iron ore mining,the rapid and accurate determination of iron ore grade has an important influence on the mining decision and economic benefit.Hyperspectral imaging technology has the advantages of wide image coverage and high accuracy and has been widely used in ore classification and composition inversion.However,the band range of existing hyperspectral imaging sensors mainly includes visible and shortwave infrared(Vis-SWIR)and near-infrared(NIR).The two data types are mostly acquired independently,lacking continuity,and the accuracy of the model built with single data is often low.Therefore,the fusion of spectral data obtained by multiple sensors can effectively solve the problems of the small band range of a single sensor and few bands containing target characteristics and improve the accuracy of iron ore grade inversion based on hyperspectral imaging technology.In this study,Pika L and Pika NIR-320hyperspectral imagers were used to collect imaging spectral data of Anshan iron ore in Vis-SWIR and NIR bands,respectively,and a spectral series fusion method based on mutual information(MI)wasproposed.Firstly,the two groups of spectral data were preprocessed.Then,mutual information is calculated on the processeddata to conduct a series fusion of spectral data.Finally,Vis-SWIR,NIR,and spectral data based on the series fusion of differentbands were used as data sources to establish RBF neural network grade inversion models,and the accuracy and precision of themodels based on spectral data before and after fusion were used as the discrimination index of the effectiveness of the fusionalgorithm.The results show that the accuracy and precision of the model built after series fusion of spectral data is higher thanthat built using Vis-SWIR and NIR spectral data alone.Compared with the spectral data based on series fusion of other bands,the accuracy and precision of the model established based on the mutual information calculation of series fusion spectral data at959.89nm are the highest,R2 0.88,RPD 2.97,RMSE 4.464,MAE 3.32.This study proposes a new idea for multi-sensorspectral fusion,which has practical significance for the application of imaging spectrum technology in the inversion of iron oregrade.