Spatial Descriptors Research Articles

Spectral-Spatial Global Graph Reasoning for Hyperspectral Image Classification.

Convolutional neural networks (CNNs) have been widely applied to hyperspectral image classification (HSIC). However, traditional convolutions can not effectively extract features for objects with irregular distributions. Recent methods attempt to address this issue by performing graph convolutions on spatial topologies, but fixed graph structures and local perceptions limit their performances. To tackle these problems, in this article, different from previous approaches, we perform the superpixel generation on intermediate features during network training to adaptively produce homogeneous regions, obtain graph structures, and further generate spatial descriptors, which are served as graph nodes. Besides spatial objects, we also explore the graph relationships between channels by reasonably aggregating channels to generate spectral descriptors. The adjacent matrices in these graph convolutions are obtained by considering the relationships among all descriptors to realize global perceptions. By combining the extracted spatial and spectral graph features, we finally obtain a spectral-spatial graph reasoning network (SSGRN). The spatial and spectral parts of SSGRN are separately called spatial and spectral graph reasoning subnetworks. Comprehensive experiments on four public datasets demonstrate the competitiveness of the proposed methods compared with other state-of-the-art graph convolution-based approaches.

Determining the density and spatial descriptors of atomic scale defects of 2H–WSe2 with ensemble deep learning

We have demonstrated atomic-scale defect characterization in scanning tunneling microscopy images of single crystal tungsten diselenide using an ensemble of U-Net-like convolutional neural networks. Coordinates, counts, densities, and spatial extents were determined from almost 16 000 defect detections, leading to the rapid identification of defect types and their densities. Our results show that analysis aided by machine learning can be used to rapidly determine the quality of transition metal dichalcogenides and provide much needed quantitative input, which may improve the synthesis process.

FE modeling to generate composite RVEs with high volume fractions and various shapes of inclusions

This work proposes a novel multi-step dynamic FE compression method to generate Representative Volume Elements (RVEs) of composites containing a variety of inclusions. This method is actualized through a sequential and four-stage procedure: (1) Sparse and periodic inclusions exhibiting a predefined orientation distribution are generated by implementing a modified random sequential adsorption algorithm, (2) Sparse inclusions undergo biaxial compression into the region of the targeted RVE via a dynamic FE analysis, (3) Periodic inclusions are compressed into the region utilizing a dynamic FE analysis with periodic boundary conditions, and (4) Positions and orientations of the compressed inclusions are extracted and the RVE in computer-aided design format is then generated. The proposed method confers advantages from four distinct perspectives: (1) applicability to various inclusion geometries, including ellipses, lobules, polygons, kidneys and stars, (2) ability to generate periodic RVEs of composites with high inclusion volume fractions (up to 80.0% for circular inclusions), (3) simple and straightforward numerical implementation without explicitly considering inclusion intersection check and (4) capacity to predefine inclusion orientation distribution. Statistical analyses utilizing multiple spatial descriptors, i.e., inclusion orientation angle, local volume fraction, Voronoi cell area, nearest-neighbor distance and orientation, second order intensity function, radial distribution function and two-point probability function, confirm randomness of inclusion distribution in the generated RVEs. The elastic properties and damage behaviors of composites via the FE homogenization method are predicted based on the generated RVEs and are compared with those of available experimental data, the literature and the mean-field homogenization models to demonstrate effectiveness of the proposed multi-step dynamic FE compression method.

Enhanced prediction of transverse mechanical behavior of unidirectional fiber-reinforced composites using new spatial descriptors based on deep neural networks

In this study, we developed a data-driven approach using a deep neural network (DNN) to predict the nonlinear stress–strain behavior of fiber-reinforced composites under transverse tensile loading. We explored the impact of different input features characterizing the composite microstructure on the prediction accuracy of the DNN model. Specifically, we introduced a spatial descriptor representing fiber arrangement and proposed new spatial descriptors based on fiber area to improve input feature quality. Incorporating widely-used spatial descriptors such as the second-order intensity function and radial distribution function improved the prediction performance compared to previous methods. Combining spatial descriptors for both fiber distribution and alignment further enhanced prediction accuracy. To address the limitations of conventional spatial descriptors, we developed new spatial descriptors using a continuous function. Our findings demonstrate the importance of selecting appropriate input features for improved DNN model performance, even with the same dataset. Moreover, the proposed fiber area-based spatial descriptors offer insights into the micromechanical behavior of composite materials.

Spatial relationship recognition via heterogeneous representation: A review

Spatial relationship between objects in an image can help to gain a deep understanding of the image. At present, spatial relationship recognition has received more and more attentions and has been applied to many computer vision tasks. Although many substantial studies have been continuing, there is few review for investigating spatial relationship recognition. Therefore, we make this comprehensive survey that covers major literatures in spatial relationship recognition. Especially, we focus on spatial description which occurs in forms of spatial descriptors, definitional descriptions, probabilistic descriptions, and model descriptions. We analyse its impacts on the state-of-the-art research on spatial relationship recognition in recent years. In this survey, we introduce a classification of methodologies, and show strengths and weaknesses of methods in each category. Besides, we compare the performance of methods and algorithms most of which are learning styled and based on popular public datasets. In addition, many future research directions are also discussed, such as technology trends, and dataset creation, etc.

Spherical correlation as a similarity measure for 3-D radiation patterns of musical instruments

We investigate the use of spherical cross-correlation as a similarity measure of sound radiation patterns, with potential applications for their study, organization, and manipulation. This work is motivated by the application of corpus-based synthesis techniques to spatial projection based on the radiation patterns of orchestral instruments. To this end, we wish to derive spatial descriptors to complement other audio features available for the organization of the sample corpus. Considering two directivity functions on the sphere, their spherical correlation can be computed from their spherical harmonic coefficients. In addition, one can search for the 3-D rotation matrix which maximizes the cross-correlation, i.e. which offers the optimal spherical shape matching. The mathematical foundations of these tools are well established in the literature; however, their practical use in the field of acoustics remains relatively limited and challenging. As a proof of concept, we apply these techniques both to simulated radiation data and to measurements derived from an existing database of 3-D directivity patterns of orchestral instruments. Using these examples we present several test cases to compare the results of spherical correlation to mathematical and acoustical expectations. A range of visualization methods are applied to analyze the test cases, including multi-dimensional scaling, employed as an efficient technique for data reduction and navigation. This article is an extended version of a study previously published in [Carpentier and Einbond. 16th Congrès Français d’Acoustique (CFA), Marseille, France, April 2022, pp. 1–6.https://openaccess.city.ac.uk/id/eprint/28202/].

Benchmarking structural evolution methods for training of machine learned interatomic potentials

When creating training data for machine-learned interatomic potentials (MLIPs), it is common to create initial structures and evolve them using molecular dynamics (MD) to sample a larger configuration space. We benchmark two other modalities of evolving structures, contour exploration (CE) and dimer-method (DM) searches against MD for their ability to produce diverse and robust density functional theory training data sets for MLIPs. We also discuss the generation of initial structures which are either from known structures or from random structures in detail to further formalize the structure-sourcing processes in the future. The polymorph-rich zirconium-oxygen composition space is used as a rigorous benchmark system for comparing the performance of MLIPs trained on structures generated from these structural evolution methods. Using Behler–Parrinello neural networks as our MLIP models, we find that CE and the DM searches are generally superior to MD in terms of spatial descriptor diversity and statistical accuracy.

Scan Context++: Structural Place Recognition Robust to Rotation and Lateral Variations in Urban Environments

Place recognition is a key module in robotic navigation. The existing line of studies mostly focuses on visual place recognition to recognize previously visited places solely based on their appearance. In this article, we address structural place recognition by recognizing a place based on structural appearance, namely from range sensors. Extending our previous work on a rotation invariant spatial descriptor, the proposed descriptor completes a generic descriptor robust to both rotation (heading) and translation when roll–pitch motions are not severe. We introduce two subdescriptors and enable topological place retrieval followed by the 1-degree of freedom semimetric localization, thereby bridging the gap between topological place retrieval and metric localization. The proposed method has been evaluated thoroughly in terms of environmental complexity and scale. The source code is available and can easily be integrated into existing light detection and ranging simultaneous localization and mapping.

Diatom Red List Species Reveal High Conservation Value and Vulnerability of Mountain Lakes

Mountain lakes are unique and often isolated freshwater habitats that harbour a rich biotic diversity. This high conservation value may be reflected by diatoms, a group of algae that is known for its reliability as a bioindicator, but which has not been studied extensively in mountain lakes of the northern European Alps. In this study, the conservation value of these lakes was assessed by characterizing the number, share, and abundance of diatom Red List (RL) taxa and their relationship with environmental variables, diatom α and β diversity (assemblage uniqueness). For this purpose, linear regression models, generalized linear models, and generalized additive models were fitted and spatial descriptors were included when relevant. Of the 560 diatom taxa identified, 64% were on the RL and half of these were assigned a threat status. As hypothesized, a decreasing share of RL species in sediment and littoral samples at higher trophic levels was reflected by higher total phosphorous content and lower Secchi depth, respectively. Species-rich lakes contained a high number of RL taxa, contrasting our hypothesis of a logarithmic relationship. In turn, RL abundance increased with uniqueness, confirming our initial hypothesis. However, some of the most unique sites were degraded by fish stocking and contained low abundances of RL species. The results demonstrate the importance of oligotrophic mountain lakes as habitats for rare freshwater biota and their vulnerability in light of human impact through cattle herding, tourism, damming, and fish stocking. Additional conservation efforts are urgently needed for mountain lakes that are still underrepresented within legal conservation frameworks. Species richness and uniqueness reflect complementary aspects of RL status and thus should be applied jointly. Uniqueness can indicate both pristine and degraded habitats, so that including information on human impacts facilitates its interpretation.

Identification of Representative Equivalent Volumes on the Microstructure of 3D-Printed Fiber-Reinforced Thermoplastics Based on Statistical Characterization.

The present work describes a methodology to compute equivalent volumes representing the microstructure of 3D-printed continuous fiber-reinforced thermoplastics, based on a statistical characterization of the fiber distribution. In contrast to recent work, the methodology herein presented determines the statistically equivalent fiber distribution directly from cross-section micrographs, instead of generating random fiber arrangements. For this purpose, several regions, with different sizes and from different locations, are cropped from main cross-section micrographs and different spatial descriptor functions are adopted to characterize the microstructures in terms of agglomeration and periodicity of the fibers. Detailed information about the adopted spatial descriptors and the algorithm implemented to identify the fiber distribution, as well as to define the location of cropped regions, are given. From the obtained statistical characterization results, the minimum size of the equivalent volume required to be representative of the fiber distribution, which is found in the cross-section micrographs of 3D-printed composite materials, is presented. To support the findings, as well as to demonstrate the effectiveness of the proposed methodology, the homogenized properties are also computed using representative equivalent volumes obtained in the statistical characterization and the results are compared to those experimentally measured, which are available in the literature.

A Spatial-Spectral Feature Descriptor for Hyperspectral Image Matching

Hyperspectral Images (HSIs) have been utilized in many fields which contain spatial and spectral features of objects simultaneously. Hyperspectral image matching is a fundamental and critical problem in a wide range of HSI applications. Feature descriptors for grayscale image matching are well studied, but few descriptors are elaborately designed for HSI matching. HSI descriptors, which should have made good use of the spectral feature, are essential in HSI matching tasks. Therefore, this paper presents a descriptor for HSI matching, called HOSG-SIFT, which ensembles spectral features with spatial features of objects. First, we obtain the grayscale image by dimensional reduction from HSI and apply it to extract keypoints and descriptors of spatial features. Second, the descriptors of spectral features are designed based on the histogram of the spectral gradient (HOSG), which effectively preserves the physical significance of the spectral profile. Third, we concatenate the spatial descriptors and spectral descriptors with the same weights into a new descriptor and apply it for HSI matching. Experimental results demonstrate that the proposed HOSG-SIFT performs superior against traditional feature descriptors.

Economic costs of invasive alien species in the Mediterranean basin

Invasive alien species (IAS) negatively impact the environment and undermine human well-being, often resulting in considerable economic costs. The Mediterranean basin is a culturally, socially and economically diverse region, harbouring many IAS that threaten economic and societal integrity in multiple ways. This paper is the first attempt to collectively quantify the reported economic costs of IAS in the Mediterranean basin, across a range of taxonomic, temporal and spatial descriptors. We identify correlates of costs from invasion damages and management expenditures among key socioeconomic variables, and determine network structures that link countries and invasive taxonomic groups. The total reported invasion costs in the Mediterranean basin amounted to $27.3 billion, or $3.6 billion when only realised costs were considered, and were found to have occurred over the last three decades. Our understanding of costs of invasions in the Mediterranean was largely limited to a few, primarily western European countries and to terrestrial ecosystems, despite the known presence of numerous high-impact aquatic invasive taxa. The vast majority of costs were attributed to damages or losses from invasions ($25.2 billion) and were mostly driven by France, Spain and to a lesser extent Italy and Libya, with significantly fewer costs attributed to management expenditure ($1.7 billion). Overall, invasion costs increased through time, with average annual costs between 1990 and 2017 estimated at $975.5 million. The lack of information from a large proportion of Mediterranean countries, reflected in the spatial and taxonomic connectivity analysis and the relationship of costs with socioeconomic variables, highlights the limits of the available data and the research effort needed to improve a collective understanding of the different facets of the costs of biological invasions. Our analysis of the reported costs associated with invasions in the Mediterranean sheds light on key knowledge gaps and provides a baseline for a Mediterranean-centric approach towards building policies and designing coordinated responses. In turn, these could help reach socially desirable outcomes and efficient use of resources invested in invasive species research and management.

The importance of spatial configuration of neighbouring land cover for explanation of surface temperature of individual patches in urban landscapes

ContextRelationships between spatial configuration of urban form and land surface temperature (LST) in the excess heat mitigation context are studied over larger tracts of land not allowing for micro-scale recommendations to urban design.ObjectivesTo identify spatial configuration descriptors (SCDs) of urban form and the size of zone of influence conducive to the formation of the coldest and hottest land cover (LC) patches of different types (buildings, grass, paved and trees) from 2 m resolution LC and 2 and 100 m resolution LST maps at two time-steps in the summer.MethodsRandom Forest regression models were deployed to explain the LST of individual LC patches of different types based on SCDs of core LC patches and patches in their neighbourhoods. ANOVA was used to determine significantly different values of the most important SCDs associated with the coldest and hottest LC patches, and analysis of quartiles informed specification of their ranges.ResultsUrban form in the immediate neighbourhood to the core LC patches had a strong influence on their LST. Low elevation, high proximity to water, and high aggregation of trees, being important to the formation of the coldest patches of all types. High resolution of LST contributed to a higher accuracy of results. Elevation and proximity to water gained in importance as summer progressed.ConclusionsSpatial configuration of urban form in the nearest proximity to individual LC patches and the use of fine resolution LST data are essential for issuing heat mitigation recommendations to urban planners relevant to micro-scales.

Modeling Rain Isotopic Composition under Orographic Control: A Landscape Approach for Hydrogeological Applications

Oxygen isotopic composition is useful for individuating recharge areas of groundwater bodies by the comparison with those of local rainfalls. While on a global scale general relationships, such as the isotopic vertical gradient or continentality effects, efficiently describe spatial variations of the isotopic signature, hydrogeological applications need spatial models that are more focused on the effects of local topographic structures and/or subsoil geology. This work presents a case study in northeastern Sicily (Italy) characterized by complex geological and orographic structures, in which isotopic composition of rainfalls is governed by orographic effects and the varying initial composition of humid air masses. We used a black box approach, comparing the average isotopic composition of rain collected from a network of eight samplers with their spatial descriptors (elevation, latitude and longitude). We obtained the best correlation with the simultaneous use of all these variables, applying their multiple linear correlation equation to transform the 1 × 1 km digital elevation model (DEM) of the study area into a digital isotopic model (DIM). The reliability of the DIM was confirmed by its good agreement with the oxygen isotopic composition contour map of the local groundwater.

Biodiversity Patterns of Macroinvertebrate Assemblages in Natural and Artificial Lentic Waters on an Oceanic Island

The Azorean islands have been historically affected by human activities, mainly due to the combined effects of habitat degradation and fragmentation, and the introduction of exotic species. We here aim to analyze the role of environmental characteristics and spatial descriptors in supporting regional biodiversity of macroinvertebrates by considering natural ponds and artificial tanks. After the monthly variation of macroinvertebrate assemblages was assessed in three temporary and two permanent ponds in the Azorean island of Terceira during a complete inundation-desiccation annual cycle, the assemblage differences of 12 ponds (three temporary and nine permanent ponds) and 8 closely-located artificial tanks were analyzed across a range of landscape disturbances. Macroinvertebrate assemblages were found to differ according to hydroperiod and sampled months. Although the former explained the highest variance, macroinvertebrate differentiation by hydroperiod was also dependent on the study month. Our results also revealed a consistent monthly pattern of species replacement. However, the contribution of nestedness to the macroinvertebrate β-diversity was notable when temporary ponds were close to desiccation, probably indicating a deterministic loss of species due to the impoverished water conditions of the ponds facing desiccation. When the macroinvertebrate assemblages were analyzed in relation to physico-chemical variations and spatial descriptors, the artificial tanks were not clearly segregated from the natural ponds, and only differentiated by pH differences. In contrast, those natural ponds exhibiting high concentrations of total phosphorous (likely signs of anthropization) also discriminated the ordination of ponds in a distance-based redundancy analysis, and showed impoverished assemblages in comparison with well-preserved ponds. The macroinvertebrate assemblages of the natural ponds showed a significant spatial pattern, but this spatial influence was not significant when tanks and ponds were considered together. Our results suggest that tanks may act as possible reservoirs of biodiversity during the desiccation period of temporary ponds, but are unable to establish successful populations. These fishless permanent tanks can complement the conservation of a biodiversity that is largely maintained by the pristine high-altitude natural ponds. The establishment of a guideline for conservation management that also considers the artificial tanks is necessary to benefit the local and regional Azorean macroinvertebrate diversity.

Infrared small target detection via incorporating spatial structural prior into intrinsic tensor sparsity regularization

Infrared small target detection is a crucial stage in many searching and tracking applications. Many tensor decomposition-based methods have achieved well performance in the scenes with uniform backgrounds and salient targets. However, the performance is potentially prone to be degraded when encountering highly complex scenes. It is mainly because the decomposition error caused by the sparse edge structures and imprecise tensor rank measures make the recovered background deviate from the real one. To mitigate these issues, we introduce an infrared sequence tensor decomposition-based method, which combines an intrinsic tensor rank measure (ITRM) and spatial structure prior to improve background recovery. With the inter-frame background correlation, ITRM is used to regularize the low-rank component of the observed tensor, which encodes the intrinsic tensor rank insights delivered by two most typical tensor decompositions to finely rectify the estimation bias of the tensor rank. To further suppress edge structures during the decomposition, we design a spatial descriptor with edge awareness by using a nonlocal structure tensor to delineate intra-frame structural edge. Furthermore, an adaptive indicator, which fuses spatial edge information and sparsity enhancing weight, is employed to replace the inflexibly and globally sparse penalty. The solution of the proposed model is addressed by an alternative direction minimization of multipliers (ADMM). Extensive experiments on real-world infrared sequences demonstrate the outperformance of the proposed method against other state-of-the-art ones, both quantitatively and qualitatively.

Spatial modeling of biological patterns shows multiscale organization of Arabidopsis thaliana heterochromatin

The spatial organization in the cell nucleus is tightly linked to genome functions such as gene regulation. Similarly, specific spatial arrangements of biological components such as macromolecular complexes, organelles and cells are involved in many biological functions. Spatial interactions among elementary components of biological systems define their relative positioning and are key determinants of spatial patterns. However, biological variability and the lack of appropriate spatial statistical methods and models limit our current ability to analyze these interactions. Here, we developed a framework to dissect spatial interactions and organization principles by combining unbiased statistical tests, multiple spatial descriptors and new spatial models. We used plant constitutive heterochromatin as a model system to demonstrate the potential of our framework. Our results challenge the common view of a peripheral organization of chromocenters, showing that chromocenters are arranged along both radial and lateral directions in the nuclear space and obey a multiscale organization with scale-dependent antagonistic effects. The proposed generic framework will be useful to identify determinants of spatial organizations and to question their interplay with biological functions.

Detection and Localization of Mask Occluded Faces by transfer learning using Faster RCNN

Here in this work we utilized the novel transfer learning based technique, where a pretrained network model weights are used to train our Faster Region Convolutional Neural Network (FRCNN). With the help of captured mask-face dataset, we further propose a masked face detection and recognition work. It incorporate three major network training modules. Proposed network module cascade two pre-trained CNNs to extract target facial features and strong region of interest (ROI) from the input dataset image for transformed domain represention with higher spatial descriptors. By similar method, omitted facial feature cues are sufficiently reconstructed and the distortion introduced by noise pixs cues by masks are reduced. Lastly, the pre-trained CNN module is used to detect and identify masked face regions of the image.The network estimates mask detection accuracy by putting bounding boxs around all the occluded mask face for performing the classification and line regression work. Experiments conducted on proposed dataset shows good accuracy and reduced running time overfew state-of-the-arts by 11.46 %.

Improving robot’s perception of uncertain spatial descriptors in navigational instructions by evaluating influential gesture notions

Human-friendly interactive features are preferred for service robots used in emerging areas of robotic applications such as caretaking, health care, assistance, education and entertainment since they are intended to be operated by non-expert users. Humans prefer to use voice instructions, responses, and suggestions in their daily interactions. Such voice instructions and responses often include uncertain spatial descriptors such as “little” and “far”, which have no definitive quantitative meaning. Service robots involve direct interactions with human users through voice communication. Therefore, the ability to effectively quantify the meaning of such uncertain spatial descriptors is necessary for human-friendly service robots. This paper proposes a novel method to quantify the uncertain spatial descriptors in navigational instructions based on the current environmental setting and the influential notions conveyed by the pointing gestures that accompany voice instructions. The uncertain spatial descriptors are quantified by a fuzzy inference system that evaluates the spatial parameters of the current environment and the influential notions conveyed by pointing gestures, if available. According to the obtained experimental results, the proposed method is capable of improving the quantification ability of uncertain spatial descriptors by robots.

The relationship between energy use and space use of higher educational buildings in subtropical Australia

With an aim to understand the impact of a variety of space usage on the building energy consumption in university campuses, this study selected 122 university buildings located in subtropical Australia and collected their weekly electricity data and detailed space use information. Two hundred and two descriptors were used to describe the usage of each room in the studied buildings and thirty most influential descriptors were selected through factor screening. Multiple linear regression models were established to link energy use and space use. The results showed that wet laboratories use a large amount of energy because of their energy-intensive equipment and high requirement of mechanical ventilation. Health is the most energy-intensive discipline as it has more wet laboratories than others. Although research dedicated buildings have the highest energy use intensity, increasing research space alone does not lead to a significant increase of building energy consumption, while increasing teaching space has a greater energy impact. Due to the complex space use of a higher educational building, it is difficult to sub-meter its energy use; this study provides useful spatial descriptors and statistical models for the study of energy use in higher educational buildings.