The Spatial LASSO With Applications to Unmixing Hyperspectral Biomedical Images

Hyperspectral imaging (HSI) is a spectroscopic method that uses densely sampled measurements along the electromagnetic spectrum to identify the unique molecular composition of an object. Traditionally HSI has been associated with remote sensing-type applications, but recently has found increased use in biomedicine, from investigations at the cellular to the tissue level. One of the main challenges in the analysis of HSI is estimating the proportions, also called abundance fractions of each of the molecular signatures. While there is great promise for HSI in the area of biomedicine, large variability in the measurements and artifacts related to the instrumentation has slow adoption into more widespread practice. In this article, we propose a novel regularization and variable selection method called the spatial LASSO (SPLASSO). The SPLASSO incorporates spatial information via a graph Laplacian-based penalty to help improve the model estimation process for multivariate response data. We show the strong performance of this approach on a benchmark HSI dataset with considerable improvement in predictive accuracy over the standard LASSO. Supplementary materials for this article are available online.     

Application of Vis–NIR hyperspectral imaging in classification between fresh and frozen-thawed pork Longissimus Dorsi muscles

Fresh and frozen-thawed (F-T) pork meats were classified by Vis–NIR hyperspectral imaging. Eight optimal wavelengths (624, 673, 460, 588, 583, 448, 552 and 609 nm) were selected by successive projections algorithm (SPA). The first three principal components (PCs) obtained by principal component analysis (PCA) accounted for over 99.98% of variance. Gray-level-gradient co-occurrence matrix (GLGCM) was applied to extract 45 textural features from the PC images. The correct classification rate (CCR) was employed to evaluate the performance of the partial least squares-discriminate analysis (PLS-DA) models, by using (A) the reflected spectra at full wavelengths and (B) those at the optimal wavelengths, (C) the extracted textures based on the PC images, and (D) the fused variables combining spectra at the optimal wavelengths and textures. The results showed that the best CCR of 97.73% was achieved by applying (D), confirming the high potential of textures for fresh and F-T meat discrimination.     

Mapping nonnative plants using hyperspectral imagery

Nonnative plant species are causing enormous ecological and environmental impacts from local to global scale. Remote sensing images have had mixed success in providing spatial information on land cover characteristics to land managers that increase effective management of invasions into native habitats. However, there has been limited evaluation of the use of hyperspectral data and processing techniques for mapping specific invasive species based on their spectral characteristics. This research evaluated three different methods of processing hyperspectral imagery: minimum noise fraction (MNF), continuum removal, and band ratio indices for mapping iceplant (Carpobrotus edulis) and jubata grass (Cortaderia jubata) in California's coastal habitat. Validation with field sampling data showed high mapping accuracies for all methods for identifying presence or absence of iceplant (97%), with the MNF procedure producing the highest accuracy (55%) when the classes were divided into four different densities of iceplant.     

Optical scattering in beef steak to predict tenderness using hyperspectral imaging in the VIS-NIR region

The objective of this research is to develop a non-destructive method for predicting cooked beef tenderness using optical scattering of light on fresh beef muscle tissue. A hyperspectral imaging system (λ = 496–1,036 nm) that consists of a CCD camera and an imaging spectrograph, was used to acquire beef steak images. The hyperspectral image consisted of 120 bands with spectral intervals of 4.54 nm. Sixty-one fresh beef steaks, including 44 strip loin and 17 tenderloin cuts, were collected. After imaging, the steaks were cooked and Warner-Bratzler shear (WBS) force values were collected as tenderness references. The optical scattering profiles were derived from the hyperspectral images and fitted to the modified Lorentzian function. Parameters, such as the peak height, full scattering width at half maximum (FWHM), and the slope around the FWHM were determined at each wavelength. Stepwise regression was used to identify 7 key wavelengths and parameters. The parameters were then used to predict the WBS scores. The model was able to predict WBS scores with an R = 0.67. Optical scattering implemented with hyperspectral imaging shows limited success for predicting current status of tenderness in beef steak.     

Target detection based on a dynamic subspace

For hyperspectral target detection, it is usually the case that only part of the targets pixels can be used as target signatures, so can we use them to construct the most proper background subspace for detecting all the probable targets? In this paper, a dynamic subspace detection (DSD) method which establishes a multiple detection framework is proposed. In each detection procedure, blocks of pixels are calculated by the random selection and the succeeding detection performance distribution analysis. Manifold analysis is further used to eliminate the probable anomalous pixels and purify the subspace datasets, and the remaining pixels construct the subspace for each detection procedure. The final detection results are then enhanced by the fusion of target occurrence frequencies in all the detection procedures. Experiments with both synthetic and real hyperspectral images (HSI) evaluate the validation of our proposed DSD method by using several different state-of-the-art methods as the basic detectors. With several other single detectors and multiple detection methods as comparable methods, improved receiver operating characteristic curves and better separability between targets and backgrounds by the DSD methods are illustrated. The DSD methods also perform well with the covariance-based detectors, showing their efficiency in selecting covariance information for detection.     

高光谱影像的BDT-SVM地物分类算法与应用

面对海量数据的特征空间高维性及训练样本的有限性,高光谱遥感影像若采用常规统计模式的分类方法难以获得较好的分类结果。因此探讨支持向量机(SVM)分类器的基本原理,针对EO-1Hyperion高光谱影像的分类特点及现有多类SVM算法所存在的训练时间长及分类精度低等问题,引入二叉决策树SVM(BDT-SVM)分类算法,并提出一种新的类间分离度定义方法及相应的客观确定二叉树结构的策略,由此生成改进的BDT-SVM算法。实验结果表明:与其他多类分类方法相比,基于改进的BDT-SVM算法的高光谱影像地物分类效果更好,总体精度达到90.96%,Kappa系数为0.89,该算法还解决了经典SVM多类分类可能存在的不可分区域问题。     

基于Hyperion数据的铜矿信息直接和间接提取方法研究

针对不同的地表覆盖条件,研究基于Hyperion星载高光谱数据的砂页岩型铜矿信息提取方法。首先对Hyperion L1级数据进行波段筛选、坏线修复、大气纠正和几何纠正等预处理,然后分别针对不同的植被覆盖情况使用不同的信息提取方法。在岩石裸露区,直接使用光谱角制图法;在植被覆盖区,使用铜金属离子的积累导致的植被生理异常作为间接标志来识别铜矿信息,生理异常使用高光谱植被指数来计算。结果表明,综合使用这两种方法的互补信息能够便于提取复杂地表覆盖情况下的铜矿信息。     

基于改进独立分量分析的高光谱数据分类研究

高光谱数据波段数目多,光谱信息量大,采用传统的分类方法无法取得较好的精确分类效果.针对上述问题,结合独立分量分析(ICA)和概率神经网络(PNN)在高光谱数据特征提取及分类中的优势,提出了一种改进的独立分量分析与概率神经网络相结合的高光谱数据分类方法,首先采用改进的独立分昔分析对高光谱数据进行降维,然后采用概率神经网络对提取的独立分量进行分类.通过仿真实验结果表明,方法可以在获得较高分类精度的同时大大节省分类的时间.     

低功耗异构计算架构的高光谱遥感图像分类研究

高光谱图像分类算法通常需要逐点对图像中的像素点进行迭代处理,计算复杂度及并行程度存在较大差异。随着高光谱遥感图像空间、光谱和辐射分辨率的不断提升,这些算法无法满足实时处理海量遥感图像数据的需求。通过分析NPU存储计算一体化模式与遥感图像分类算法的实现步骤,设计低功耗CPU+NPU异构资源计算架构的低秩稀疏子空间聚类（LRSSC）算法,将数据密集型计算转移至NPU,并利用NPU数据驱动并行计算和内置AI加速,对基于机器学习算法的海量遥感数据进行实时分类。受到big.LITTLE计算范式的启发,CPU+NPU异构资源计算架构由8 bit和低精度位宽NPU共同组成以提高整体吞吐量,同时减少图网络推理过程中的能量损耗。实验结果表明,与CPU计算架构和CPU+GPU异构计算架构的LRSSC算法相比,CPU+NPU异构计算架构的LRSSC算法在Pavia University遥感数据集下的计算速度提升了3~14倍。