Spectral Classifier Research Articles

An Adaptive Atrous Spatial Pyramid Pooling Network for Hyperspectral Classification

Hyperspectral imaging (HSI) offers rich spectral and spatial data, beneficial for a variety of applications. However, challenges persist in HSI classification due to spectral variability, non-linearity, limited samples, and a dearth of spatial information in conventional spectral classifiers. While various spectral–spatial classifiers and dimension reduction techniques have been developed to mitigate these issues, they are often constrained by the utilization of handcrafted features. Deep learning has been introduced to HSI classification, with pixel- and patch-level deep learning (DL) classifiers gaining substantial attention. Yet, existing patch-level DL classifiers encounter difficulties in concentrating on long-distance dependencies and managing category areas of diverse sizes. The proposed Self-Adaptive 3D atrous spatial pyramid pooling (ASPP) Multi-Scale Feature Fusion Network (SAAFN) addresses these challenges by simultaneously preserving high-resolution spatial detail data and high-level semantic information. This method integrates a modified hyperspectral superpixel segmentation technique, a multi-scale 3D ASPP convolution block, and an end-to-end framework to extract and fuse multi-scale features at a self-adaptive rate for HSI classification. This method significantly enhances the classification accuracy of HSI with limited samples.

Towards automated detection of hyperhydricity in plant in vitro culture

Hyperhydricity (HH) is one of the most important physiological disorders that negatively affects various plant tissue culture techniques. The objective of this study was to characterize optical features to allow an automated detection of HH. For this purpose, HH was induced in two plant species, apple and Arabidopsis thaliana, and the severity was quantified based on visual scoring and determination of apoplastic liquid volume. The comparison between the HH score and the apoplastic liquid volume revealed a significant correlation, but different response dynamics. Corresponding leaf reflectance spectra were collected and different approaches of spectral analyses were evaluated for their ability to identify HH-specific wavelengths. Statistical analysis of raw spectra showed significantly lower reflection of hyperhydric leaves in the VIS, NIR and SWIR region. Application of the continuum removal hull method to raw spectra identified HH-specific absorption features over time and major absorption peaks at 980 nm, 1150 nm, 1400 nm, 1520 nm, 1780 nm and 1930 nm for the various conducted experiments. Machine learning (ML) model spot checking specified the support vector machine to be most suited for classification of hyperhydric explants, with a test accuracy of 85% outperforming traditional classification via vegetation index with 63% test accuracy and the other ML models tested. Investigations on the predictor importance revealed 1950 nm, 1445 nm in SWIR region and 415 nm in the VIS region to be most important for classification. The validity of the developed spectral classifier was tested on an available hyperspectral image acquisition in the SWIR-region.

Machine learning for underwater laser detection and differentiation of macroalgae and coral

A better understanding of how spatial distribution patterns in important primary producers and ecosystem service providers such as macroalgae and coral are affected by climate-change and human activity-related events can guide us in anticipating future community and ecosystem response. In-person underwater field surveys are essential in capturing fine and/or subtle details but are rarely simple to orchestrate over large spatial scale (e.g., hundreds of km). In this work, we develop an automated spectral classifier for detection and classification of various macroalgae and coral species through a spectral response dataset acquired in a controlled setting and via an underwater multispectral laser serial imager. Transferable to underwater lidar detection and imaging methods, laser line scanning is known to perform in various types of water in which normal photography and/or video methods may be affected by water optical properties. Using off the shelf components, we show how reflectance and fluorescence responses can be useful in differentiating algal color groups and certain coral genera. Results indicate that while macroalgae show many different genera and species for which differentiation by their spectral response alone would be difficult, it can be reduced to a three color-type/class spectral response problem. Our results suggest that the three algal color groups may be differentiated by their fluorescence response at 580 nm and 685 nm using common 450 nm, 490 nm and 520 nm laser sources, and potentially a subset of these spectral bands would show similar accuracy. There are however classification errors between green and brown types, as they both depend on Chl-a fluorescence response. Comparatively, corals are also very diverse in genera and species, and reveal possible differentiable spectral responses between genera, form (i.e., soft vs. hard), partly related to their emission in the 685 nm range and other shorter wavelengths. Moreover, overlapping substrates and irregular edges are shown to contribute to classification error. As macroalgae are represented worldwide and share similar photopigment assemblages within respective color classes, inter color-class differentiability would apply irrespective of their provenance. The same principle applies to corals, where excitation-emission characteristics should be unchanged from experimental response when investigated in-situ.

Identification of acute myeloid leukemia by infrared difference spectrum of peripheral blood

Acute myeloid leukemia (AML) is a high mortality and recurrence rates hematologic malignancy. Thus, whatever early detection or subsequent visit are both of high significance. Traditional AML diagnosis is conducted via peripheral blood (PB) smear and bone marrow (BM) aspiration. But BM aspiration is a painful burden for patients especially in early detection or subsequent visit. Herein, the use of PB to evaluate and identify the leukemia characteristics will be an attractive alternative source for early detection or subsequent visit. Fourier transform infrared spectroscopy (FTIR) is a time- and cost-effective approach to reveal the disease-related molecular features and variations. However, to the best of our knowledge, there is no attempts using infrared spectroscopic signatures of PB to replace BM for identifying AML. In this work, we are the first to develop a rapid and minimally invasive method to identify AML by infrared difference spectrum (IDS) of PB with only 6 characteristic wavenumbers. We dissect the leukemia-related spectroscopic signatures of three subtypes of leukemia cells (U937, HL-60, THP-1) by IDS, revealing biochemical molecular information about leukemia for the first time. Furthermore, the novel study links cellular features to complex features of blood system which demonstrates the sensitivity and specificity with IDS method. On this basis, BM and PB of AML patients and healthy controls were provided to parallel comparison. The IDS of BM and PB combined with principal component analysis method revealing that the leukemic components in BM and PB can be described by IDS peaks of PCA loadings, respectively. It is demonstrated that the leukemic IDS signatures of BM can be replaced by the leukemic IDS signatures of PB. In addition, the IDS signatures of leukemia cells are reflected in PB of AML patients with peaks of 1629, 1610, 1604, 1536, 1528 and 1404 cm−1 for the first time as well. To this end, we access the leukemic signatures of IDS peaks to compare the PB of AMLs and healthy controls. It is confirmed that the leukemic components can be detected from PB of AML and distinguished into positive (100%) and negative (100%) groups successfully by IDS classifier which is a novel and unique spectral classifier. This work demonstrates the potential use of IDS as a powerful tool to detect leukemia via PB which can release subjects’ pain remarkably.

On-chip hyperspectral image segmentation with fully convolutional networks for scene understanding in autonomous driving

Most of current computer vision-based advanced driver assistance systems (ADAS) perform detection and tracking of objects quite successfully under regular conditions. However, under adverse weather and changing lighting conditions, and in complex situations with many overlapping objects, these systems are not completely reliable. The spectral reflectance of the different objects in a driving scene beyond the visible spectrum can offer additional information to increase the reliability of these systems, especially under challenging driving conditions. Furthermore, this information may be significant enough to develop vision systems that allow for a better understanding and interpretation of the whole driving scene. In this work we explore the use of snapshot, video-rate hyperspectral imaging (HSI) cameras in ADAS on the assumption that the near infrared (NIR) spectral reflectance of different materials can help to better segment the objects in real driving scenarios. To do this, we have used the HSI-Drive 1.1 dataset to perform various experiments on spectral classification algorithms. However, the information retrieval of hyperspectral recordings in natural outdoor scenarios is challenging, mainly because of deficient color constancy and other inherent shortcomings of current snapshot HSI technology, which poses some limitations to the development of pure spectral classifiers. In consequence, in this work we analyze to what extent the spatial features codified by standard, tiny fully convolutional network (FCN) models can improve the performance of HSI segmentation systems for ADAS applications. In order to be realistic from an engineering viewpoint, this research is focused on the development of a feasible HSI segmentation system for ADAS, which implies considering implementation constraints and latency specifications throughout the algorithmic development process. For this reason, it is of particular importance to include the study of the raw image preprocessing stage into the data processing pipeline. Accordingly, this paper describes the development and deployment of a complete machine learning-based HSI segmentation system for ADAS, including the characterization of its performance on different embedded computing platforms, including a single board computer, an embedded GPU SoC and a programmable system on chip (PSoC) with embedded FPGA. We verify the superiority of the FPGA-PSoC over the GPU-SoC in terms of energy consumption and, particularly, processing latency, and demonstrate that it is feasible to achieve segmentation speeds within the range of ADAS industry specifications using standard development tools.

Multi-Scale Superpixel-Guided Structural Profiles for Hyperspectral Image Classification.

Hyperspectral image classification has received a lot of attention in the remote sensing field. However, most classification methods require a large number of training samples to obtain satisfactory performance. In real applications, it is difficult for users to label sufficient samples. To overcome this problem, in this work, a novel multi-scale superpixel-guided structural profile method is proposed for the classification of hyperspectral images. First, the spectral number (of the original image) is reduced with an averaging fusion method. Then, multi-scale structural profiles are extracted with the help of the superpixel segmentation method. Finally, the extracted multi-scale structural profiles are fused with an unsupervised feature selection method followed by a spectral classifier to obtain classification results. Experiments on several hyperspectral datasets verify that the proposed method can produce outstanding classification effects in the case of limited samples compared to other advanced classification methods. The classification accuracies obtained by the proposed method on the Salinas dataset are increased by 43.25%, 31.34%, and 46.82% in terms of overall accuracy (OA), average accuracy (AA), and Kappa coefficient compared to recently proposed deep learning methods.

The local universe in the era of large surveys - II. multi-wavelength characterization of activity in nearby S0 galaxies

ABSTRACT This is the second paper in a series using data from tens of thousands S0 galaxies of the local Universe (z ≲ 0.1) retrieved from the NASA-Sloan Atlas. It builds on the outcomes of the previous work, which introduced a new classification scheme for these objects based on the principal component analysis (PCA) of their optical spectrum and its projections on to the first two eigenvectors or principal components (the PC1–PC2 diagram). We provide a comprehensive characterization of the activity of present-day S0s throughout both the broad-band PC1–PC2 spectral classifier and the conventional narrow-line BPT/WHAN ones, contrasting the different types of activity classes they define, and present an alternative diagram that exploits the concordance between WHAN and PCA demarcations. The analysis is extended to the mid-infrared, radio and X-ray wavelengths by crossmatching our core sample with data from the WISE, FIRST, XMM–Newton, and Chandra surveys. This has allowed us to carry out a thorough comparison of the most important activity diagnostics in the literature over different wavebands, discuss their similarities and differences, and explore the connections between them and with parameters related to star formation and black hole accretion. In particular, we find evidence that the bulk of nebular emission from radio and X-ray detected S0–Seyfert and LINER systems is not driven by star birth, while the dominant ionising radiation for a number of LINERs might come from post-AGB stars. These and other outcomes from the present work should be transferable to other morphologies.

Double Attention Based Multilevel One-Dimensional Convolution Neural Network for Hyperspectral Image Classification

The large spectral variability and nonlinearity of hyperspectral images (HSIs) make classification a challenging task. Hence, the powerful capacities for feature extraction and nonlinear representation provided by deep learning have made it a widely-used tool for HSI classification. However, most of these methods involve complex spectral-spatial networks with a mass of parameters to achieve high-performance results, which are limited by the small number of training samples and large computational costs in practice. Simple networks harnessing spectral information alone effectively alleviate these problems, but these approaches generally underutilize channel correlations and hierarchical complementarity, which degrades discriminability and lowers performance. To solve these problems, a novel end-to-end double attention based multilevel one-dimensional convolution neural network (DAMN) is proposed, that fully exploits the rich spectral information contained in HSIs to boost an accurate and efficient classification with a limited number of samples. An effective multilevel one-dimensional convolution neural network is constructed to hierarchically mine local-channel correlations and comprehensively utilize shallow and deep features. Then, a multilayer perceptron attention module is utilized to extract informative features from raw spectral signatures as network feeds, with an advanced ultra-lightweight subspace attention module employed to learn complementary cross-channel dependencies to further enhance the discriminability of features. The effectiveness of DAMN was validated on three frequently-used hyperspectral datasets, with the results showing that DAMN outperforms the other state-of-the-art spectral classifiers and can be a competitive tool in practical applications.

Spectral Feature Fusion Networks With Dual Attention for Hyperspectral Image Classification

Recent progress in spectral classification is largely attributed to the use of convolutional neural networks (CNNs). While a variety of successful architectures have been proposed, they all extract spectral features from various portions of adjacent spectral bands. In this article, we take a different approach and develop a deep spectral feature fusion method, which extracts both local and interlocal spectral features, capturing thus also the correlations among nonadjacent bands. To our knowledge, this is the first reported deep spectral feature fusion method. Our model is a two-stream architecture, where an intergroup and a groupwise spectral classifier operate in parallel. The interlocal spectral correlation feature extraction is achieved elegantly, by reshaping the input spectral vectors to form the so-called nonadjacent spectral matrices. We introduce the concept of groupwise band convolution to enable the efficient extraction of discriminative local features with multiple kernels adopting the local spectral content. Another important contribution of this work is a novel dual-channel attention mechanism to identify the most informative spectral features. The model is trained in an end-to-end fashion with a joint loss. Experimental results on real datasets demonstrate excellent performance compared with the current state of the art.

Error removal by energy scaling from hyperspectral images for performance improvement of spectral classifiers

ABSTRACT In the remote sensing community, HyperSpectral (HS) images (HSI) are becoming increasingly popular as the advancement of technology and the consequent reduction of cost make them financially more accessible. The reason for their success is the higher capability they can offer, with respect to multispectral data, to discriminate classes that are spectrally similar. A series of pre-processing steps such as geometric, radiometric and atmospheric corrections are carried out on raw data captured by HS sensors. The processed data is then passed to the data analyst, whose work generally relies on the assumption that the received HS data is ‘clean,’ as all possible corrections have already been implemented; however, corrections are hardly perfect and residual disturbances can still bias the quality of results. At this stage, however, all corrections based on ancillary information have already been made and the possibilities for ‘exogenous’ correction of data are exhausted. More could be possibly done by sourcing additional information from the data itself. In this paper, we propose a simple yet effective additional step for error suppression through energy scaling, termed ‘Error Removal by Energy Scaling’ (ERES). In classification problems, the absolute value of wavelength is often overlooked, except for, e.g. removal of strong absorption bands; yet the value can actually further support the classification process if their physical meaning is tapped. The proposed ERES method is indeed a non-linear scaling method, derived from physical phenomena linked with radiation extinction properties. In ERES, each band is associated with an energy level, that is inversely related to its own wavelength. The associated nonlinear energy information in HSI, neglected in most classification strategies, prevents optimal separation of class-specific spectral signatures, that are generated by the physics of wave-matter interaction. This is especially true for linear classifiers such as Support Vector Machines (SVM). Removing this physics-linked information makes data more suitable to be classified with physics-unaware classification strategies, typically used in down stream remotely sensed data processing. The relevance of the issue, and the benefits of ERES, are discussed and validated in this work over three different datasets, using accuracy improvements on the popular Spectral Angle Mapper and SVM classifiers as a means to gauge the effectiveness of the correction strategy. Results clearly reveal the positive impact of applying ERES to the data before proceeding to classification.

Abstract 2821: Spectral classifier as a complement to magnetic resonance spectroscopy in the management of patients with brain tumors of the central nervous system

Abstract Introduction: In the last decades molecular imaging, more precisely Magnetic Resonance Spectroscopy (MRS), has taken an important role in medical decision-making. The MRS provides biochemical information on metabolites on tissues, it also provides information regarding tissue metabolism, allowing to characterize some brain metabolites of a certain volume of the brain. With due experience, patterns of relative metabolite levels can be visually interpreted to arrive at radiological guidance, but the most robust result is obtained from quantitative analysis of the spectra using pattern recognition methods. Within these methods we find the Spectral Classifier (SC), it is a Java solution for designing and implementing MRS based classifiers. The main goal of SC is to allow users with minimum background knowledge of multivariate statistics to perform a fully automated pattern recognition analysis, it is able to read low resolution in-vivo MRS (single-voxel and multi-voxel) and high resolution tissue MRS. Objective: To implement the use of the SC as a tool for the processing of Magnetic Resonance Spectroscopy (MRE) studies in patients with brain tumors of the central nervous system treated at the National Cancer Institute, Bogotá, Colombia. Methods: The spectral acquisition parameters in the Siemens 1.5T resonator were homogenized, the spectra acquisition protocol for ERM was modified by including new single and multivoxel voxel sequences (10x10 voxel mesh). 63 studies were carried out using single and multivoxel voxel, which were processed with the jMRUI and jMRUI2xml plugin of the SC program, a pattern recognition program, which takes the information and introduces it into a glial tumor database, allowing it to classify the masses depending on the pattern of their metabolites, and the scVisualizer. The reading was performed by two neuroradiologists, to whom only the spectroscopic data and the reference image in the axial plane and T2-weighted were provided. Results: From the first radiological reading without SC, 36.5% by single voxel and 36.5% by multivoxel corresponded to glial tumor patterns. By using SC, 76.2% per single voxel and 63.5% per multivoxel corresponded to glial tumor patterns. From the follow-up and post-radiological diagnosis, it was confirmed that of the 63 studies, 77.8% corresponded to recurrences. 100% and 87.2% of the lesions with glial tumor patterns corresponded to recurrences according to the post-radiological diagnosis for single-voxel and multi-voxel SC, respectively. Conclusions: Through the implementation of the SC, additional information is obtained that allows to determine and confirm the presence of glial tumor patterns, which could contribute to better decision-making in the management of this type of patients. Citation Format: Milena Acosta, Gina Catalina Malaver, Oscar Gamboa, Cesar Rodriguez, Julian Beltran, Pablo Moreno-Acosta. Spectral classifier as a complement to magnetic resonance spectroscopy in the management of patients with brain tumors of the central nervous system [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2821.

Plant identification of Beijing Hanshiqiao wetland based on hyperspectral data

Hyperspectral data play an important role in monitoring and protecting wetlands because it not only offers high resolution and wide observation range but also reveal plant information. To explore the hyperspectral characteristics of wetland plants and their applicability in identification, in situ hyperspectral data for 15 species of wetland plants were collected from the Hanshiqiao Wetland Nature Reserve in Beijing, a constructed wetland. Then, first derivative, second derivative, and logarithmic transformation of the reciprocal of the hyperspectral reflectance data were performed using three data conversion methods (principal component analysis, vegetation index, and full-band spectrum) to identify characteristic hyperspectral variables. Four classifiers were compared, namely, support vector machine, decision tree, back-propagation neural network, and random forest. The results demonstrated that: (1) for a comprehensive comparison of the four spectral features data, logarithmic treatment of the reciprocal significantly improved the identification accuracy; (2) for the three data conversion methods for identifying spectral characteristics, the identification accuracy based on the full-band spectrum was better than the other two data forms, and the identification effectiveness of principal component analysis was better than that of the vegetation index; (3) because of differences in plant types and data formats, none of the four classifiers had obvious advantages or disadvantages over the others; and (4) the identification results for all plant statistics further verified the applicability of full-band spectrum data for plant identification. The results showed that the vegetation index and principal component variables reduced the calculation time but did not improve the identification accuracy. Overall, the spectral variables and classifiers examined in this study could meet the needs of wetland plant identification to a certain extent.

Multispatial Filtering Module Cascaded System for Hyperspectral Image Classification

This article presents a multispatial filtering module cascaded system (MSFMCS) for hyperspectral image classification (HSIC), which can serve as a paradigm to improve spectral–spatial classification. It includes multiple spatial filtering modules (SFMs) that are cascaded to particularly capture spatial information from the classification maps generated from the preceding modules. As a result, any spectral classifier (SC) can be used as an input to an initial/input module (IM). Through MSFMCS, its classification performance keeps improving as more SFMs are processed. To terminate MSFMCS, an automatic stopping rule is particularly designed by support vector machine (SVM) which is used not only as a classifier but also as a decision-maker. So, once an SC cannot be further improved, MSFMCS is terminated. One major benefit resulting from MSFMCS is its framework which can implement any arbitrary SC as its initial classifier in IM. Another is its ability in capturing additional spatial classification information module by module as the process progresses. A third one is no weights connected between modules so that no training phase is required like a feedforward neural network. Finally, the number of modules used in MSFMCS can be automatically determined by its stopping rule not predetermined empirically. To illustrate full advantages of MSFMCS in HSIC, three types of heterogeneous classifiers, pure-pixel-based SVM, mixed-pixel-based constrained energy minimization (CEM), and feature-extraction-based classifier—orthogonal total variation component analysis (OTVCA)—are used for experiments to demonstrate how MSFMCS can improve their classification performance.

Spectral-Spatial Hyperspectral Image Classification Using Robust Dual-Stage Spatial Embedding

Recently, many spectral-special classification models have emerged one after another in the remote sensing community. These models aim to introduce the spatial information of the pixel to improve the accuracy of the class attribute of the pixel. However, for the spectral-spatial classification algorithms, not all pixels need to introduce the corresponding spatial information since the use of a large amount of spatial information has a costly time. To solve this problem, this paper proposes a robust dual-stage spatial embedding (RDSSE) model for spectral-spatial classification of hyperspectral image, which is composed of the following several main steps: First, an over-segmentation algorithm is employed to cluster original hyperspectral image into many superpixel blocks with shape adaptive characteristics. Then, we design a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula> -peak criterion to fuse the spectral feature of pixels within and between superpixels. Next, a low time-consumption spectral classifier is introduced to conduct primary classification for a testing pixel to achieve the corresponding probability distribution. Specifically, the difference between the probability of the largest class and that of the second-largest class is served as class confidence. Finally, the predicted label of the low-confidence testing pixels is reclassified based on a high-accuracy spectral-spatial classification method. Experimental results on several real images illustrated that the proposed RDSSE method can obtain superior performance with respect to several competitive methods.

HSIGAN: A Conditional Hyperspectral Image Synthesis Method With Auxiliary Classifier

In this article, we explore a conditional hyperspectral image (HSI) synthesis method with generative adversarial networks (GAN). A new multistage and multipole generative adversarial network, which is suitable for conditional HSI generation and classification (HSIGAN), is proposed. For HSIs synthesis, it is crucial to learn a great deal of spatial-spectral distribution features from source data. The multistage progressive training makes the generator effectively imitate the real data by fully exploiting the high-dimension learning capability of GAN models. The coarse-to-fine information extraction method helps the discriminator to understand the semantic feature better while the multiscale classification prediction presents a positive impact on results. A spectral classifier joins the adversarial network, which offers a helping hand to stabilize and optimize the model. Moreover, we apply the 3-D DropBlock layer in the generator to remove semantic information in a contiguous spatial-spectral region and avoid model collapse. Experimental results of the quantitative and qualitative evaluation show that HSIGAN could generate high-fidelity, diverse hyperspectral cubes while achieving top-ranking accuracy for supervised classification. This result is encouraging for using GANs as a data augmentation strategy in the HSI vision task.

Multilevel Structure Extraction-Based Multi-Sensor Data Fusion

Multi-sensor data on the same area provide complementary information, which is helpful for improving the discrimination capability of classifiers. In this work, a novel multilevel structure extraction method is proposed to fuse multi-sensor data. This method is comprised of three steps: First, multilevel structure extraction is constructed by cascading morphological profiles and structure features, and is utilized to extract spatial information from multiple original images. Then, a low-rank model is adopted to integrate the extracted spatial information. Finally, a spectral classifier is employed to calculate class probabilities, and a maximum posteriori estimation model is used to decide the final labels. Experiments tested on three datasets including rural and urban scenes validate that the proposed approach can produce promising performance with regard to both subjective and objective qualities.

Fusion of Dual Spatial Information for Hyperspectral Image Classification

The inclusion of spatial information into spectral classifiers for fine-resolution hyperspectral imagery has led to significant improvements in terms of classification performance. The task of spectral-spatial hyperspectral image classification has remained challenging because of high intraclass spectrum variability and low interclass spectral variability. This fact has made the extraction of spatial information highly active. In this work, a novel hyperspectral image classification framework using the fusion of dual spatial information is proposed, in which the dual spatial information is built by both exploiting pre-processing feature extraction and post-processing spatial optimization. In the feature extraction stage, an adaptive texture smoothing method is proposed to construct the structural profile (SP), which makes it possible to precisely extract discriminative features from hyperspectral images. The SP extraction method is used here for the first time in the remote sensing community. Then, the extracted SP is fed into a spectral classifier. In the spatial optimization stage, a pixel-level classifier is used to obtain the class probability followed by an extended random walker-based spatial optimization technique. Finally, a decision fusion rule is utilized to fuse the class probabilities obtained by the two different stages. Experiments performed on three data sets from different scenes illustrate that the proposed method can outperform other state-of-the-art classification techniques. In addition, the proposed feature extraction method, i.e., SP, can effectively improve the discrimination between different land covers.

Fusion of Spectral–Spatial Classifiers for Hyperspectral Image Classification

A spectral-spatial (SS) hyperspectral classifier generally implements a spectral classifier (SC) followed by a spatial filter (SF) for classification. This article develops a new approach to fusing multiple SC-SF classifiers for hyperspectral image classification (HSIC) as to improve classification performance. To accomplish this goal an iterative process is particularly designed to fuse the spatial-filtered classification maps (SFMaps) produced by each of SC-SF classifiers into one single SFMap via maximum a posteriori (MAP) criterion. Such fused SFMaps are then fed back and added to the current data cube to create a new data set for next round SC-SF classifier fusion. The same process is repeated iteratively until it satisfies an automatic stopping rule. To further fuse more than two SS methods, two approaches are also developed, called simultaneous multiple SC-SF fusion (SMSSF) method and progressive multiple SC-SF fusion (PMSSF) method. Experimental results demonstrate that fusing multiple SC-SF classifiers can indeed perform better than using an individual single SC-SF classifier alone without fusion.

Unified compressive sensing paradigm for the random demodulator and compressive multiplexer architectures

A major challenge in spectrum sensing for cognitive radio (CR) applications is the very high sampling rates involved, which imposes significant demands on the signal acquisition technology. This has given impetus to applying compressive sensing (CS) as a sub-Nyquist sampling paradigm for CR-type wireless signals which exhibit sparsity in certain domains. CS architectures like the random demodulator (RD) and compressive multiplexer (CM) can be used for CR spectral sensing, though both are inherently restricted in terms of the signal classes they can effectively process. To address these limitations, this paper presents two unified RD and CM-based CS architectures that seamlessly integrate precolouring and the multitaper spectral estimator into their respective structures to facilitate efficient sensing of both digitally modulated and narrowband signals, along with popular CR-access technologies like orthogonal frequency division multiplexing. A significant feature of these unified CS architectures is they do not require a priori knowledge of either the input signal or modulation scheme, while a tristate spectral classifier is introduced to afford notably enhanced spectrum access opportunities for unlicensed secondary users. A critical performance evaluation corroborates that both unified architectures demonstrate consistently superior CS results and robustness across a broad range of CR-type signals, modulations and access technologies.

Remote sensing image classification using subspace sensor fusion

The amount of remote sensing and ancillary datasets captured by diverse airborne and spaceborne sensors has been tremendously increased, which opens up the possibility of utilizing multimodal datasets to improve the performance of processing approaches with respect to the application at hand. However, developing a generic framework with high generalization capability that can effectively fuse diverse datasets is a challenging task since the current approaches are usually only applicable to two specific sensors for data fusion. In this paper, we propose an accurate fusion-based technique called SubFus with capability to integrate diverse remote sensing data for land cover classification. Here, we assume that a high dimensional multisensor dataset can be represented fused features that live in a lower-dimensional space. The proposed classification methodology includes three main stages. First, spatial information is extracted by using spatial filters (i.e., morphology filters). Then, a novel low-rank minimization problem is proposed to represent the multisensor datasets in subspaces using fused features. The fused features in the lower-dimensional subspace are estimated using a novel iterative algorithm based on the alternative direction method of multipliers. Third, the final classification map is produced by applying a supervised spectral classifier (i.e., random forest) on the fused features. In the experiments, the proposed method is applied to a three-sensor (RGB, multispectral LiDAR, and hyperspectral images) dataset captured over the area of the University of Houston, the USA, and a two-sensor (hyperspectral and LiDAR) dataset captured over the city of Trento, Italy. The land-cover maps generated using SubFus are evaluated based on classification accuracies. Experimental results obtained by SubFus confirm considerable improvements in terms of classification accuracies compared with the other methods used in the experiments. The proposed fusion approach obtains 85.32% and 99.25% in terms of overall classification accuracy on the Houston (the training portion of the dataset distributed for the data fusion contest of 2018) and trento datasets, respectively.