熵平均密度分段压缩感知反射光谱重建方法  

An Entropy Density Segment Compressed Sensing Method for Reflectance Spectrum Reconstruction

作　　者：赵首博 ZHAO Shou-bo(School of Mechanical Engineering,Guangdong Ocean University,Zhanjiang 524088,China;School of Measurement and Control Technology and Communication Engineering,Harbin University of Science and Technology,Harbin 150080,China)

机构地区：[1]广东海洋大学机械工程学院,广东湛江524088 [2]哈尔滨理工大学测控技术与通信工程学院,黑龙江哈尔滨150080

出　　处：《光谱学与光谱分析》2024年第11期3090-3094,共5页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(61801148);黑龙江省高校青年创新人才培养计划项目(UNPYSCT-2020187);广东海洋大学科研启动经费资助项目(060302062301)资助。

摘　　要：反射光谱作为物体表面重要特征被广泛用于远程遥感目标识别、物质成分的含量检测、农业作物成熟度检测、医学影像的疾病诊断等领域。为解决反射光谱数据冗余,实现全光谱数据稀疏表达和提高光谱重建精度,将压缩感知技术应用于光谱数据分析和处理。针对全局光谱压缩感知重建方法中各波段数据稀疏度的差异性对采样率的限制条件不同,提出熵平均密度分段压缩感知反射光谱重建方法。首先定义熵平均密度作为光谱分段参考量来寻找光谱分段断点和判定各分段光谱的熵密度高低。而后依据有限等距约束条件重新分配各分段光谱的采样率,生成测量矩阵和稀疏矩阵完成各局部反射光谱稀疏感知。采用正交匹配追踪算法求最优解,分配各分段光谱的迭代次数,用感知矩阵中的列原子和稀疏信号进行迭代匹配重构各局部反射光谱,将各重构的局部反射光谱缝合为全局反射光谱。用全局光谱压缩感知方法和该方法对标准色块24Munsell ColorChecker的反射光谱进行对比实验。实验结果表明,较之于全局光谱压缩感知方法,该方法重建光谱曲线高熵密度区重建精度更高,低熵密度区压缩效率更高,在总压缩采样率不变的情况下,RMSE和MAPE统计数据得到改善,提升了整体曲线重建效果。Reflectance spectrum,as a significant characteristic of the object surface,is widely used in various fields such as remote sensing target identification,content detection of material components,agricultural crop maturity detection,and disease diagnosis in medical imaging.However,while the reflectance spectrum enriches target information,it also brings data redundancy,causing great difficulties in acquiring,processing,and transmitting spectral data.To settle these difficulties,our team has focused on spectral data analysis and processing utilizing compressed sensing technology.It was found that sparse representation of global spectral data was achieved,and spectral reconstruction accuracy was improved.Various sparsities of data in each spectral band constrain different sampling rates in spectral compressed sensing reconstruction methods.This paper proposes an entropy density segment compressed sensing method for reflectance spectrum reconstruction.Specifically,entropy average density is defined as the segmenting reference in the search for breakpoints.Based on the reference,the decision on whether the entropy density of each segmented spectrum is high or low can be given.After that,the sampling rates of each segmented spectrum are reassigned according to the limited equidistant constraint condition.The measurement and sparse matrices are generated for sparsity sensing of segmented reflectance spectrum.The optimal solution is obtained using the orthogonal matching pursuit algorithm.Iteration times of each segmented spectrum are reassigned.Each segmented reflectance spectrum is iteratively matched and reconstructed using the columns in the sensing matrix and sparse signals.Finally,the reconstructed segmented reflectance spectrums are stitched.A comparative experiment was conducted on the reflectance spectrum of the standard color block(24 Munsell ColorChecker)using the global spectral compressed sensing method and our proposed method.The experimental results show that compared with the global spectral compressed sen

关 键 词：反射光谱函数 压缩感知 信息熵 光谱分段 

分 类 号：O433.4[机械工程—光学工程]