Explore the vast range of titles available.

The scaled standard deviations of temperature and humidity are investigated in complex terrain. The study area is a steep Alpine valley, with six measurement sites of different slope, orientation and roughness (i-Box experimental site, Inn Valley, Austria). Examined here are several assumptions forming the basis of Monin–Obukhov similarity theory (MOST), including constant turbulence fluxes with height and the degree of self-correlation between the involved turbulence variables. Since the basic assumptions for the applicability of the MOST approach—horizontally homogeneous and flat conditions—are violated, the analysis is performed based on a local similarity hypothesis. The scaled standard deviations as a function of local stability are compared with previous studies from horizontally homogeneous and flat terrain, horizontally inhomogeneous and flat terrain, weakly inhomogeneous and flat terrain, as well as complex terrain. As a reference, similarity relations for unstable and stable conditions are evaluated using turbulence data from the weakly inhomogeneous and flat terrain of the Cabauw experimental site in the Netherlands, and assessed with the same post-processing method as the i-Box data. Significant differences from the reference curve and also among the i-Box sites are noted, especially for data derived from the i-Box sites with steep slopes. These differences concern the slope and the magnitude of the best-fit curves, illustrating the site dependence of any similarity theory.

The present paper shows that local similarity theories, proposed for the strongly-stratified boundary layers, can be derived as invariant solutions defined under the Lie-group theory. A system truncated to the mean momentum and buoyancy equations is considered for this purpose. The study further suggests how similarity functions for the mean profiles are determined from the vertical fluxes, with a potential dependence on a measure of the anisotropy of the system. A time scale that is likely to characterize the transiency of a system is also identified as a non-dimensionalization factor.

Aiming at the difficult problem of plant leaf recognition on the large-scale database, a two-stage local similarity based classification learning (LSCL) method is proposed by combining local mean-based clustering (LMC) method and local sparse representation based classification (SRC) (LWSRC). In the first stage, LMC is applied to coarsely classifying the test sample. k nearest neighbors of the test sample, as a neighbor subset, is selected from each training class, then the local geometric center of each class is calculated. S candidate neighbor subsets of the test sample are determined with the first S smallest distances between the test sample and each local geometric center. In the second stage, LWSRC is proposed to approximately represent the test sample through a linear weighted sum of all k × S samples of the S candidate neighbor subsets. Experimental results on the leaf image database demonstrate that the proposed method not only has a high accuracy and low time cost, but also can be clearly interpreted.

Depth perception capability is one of the essential requirements for various autonomous driving platforms. However, accurate depth estimation in a real-world setting is still a challenging problem due to high computational costs. In this paper, we propose a lightweight depth completion network for depth perception in real-world environments. To effectively transfer a teacher’s knowledge, useful for the depth completion, we introduce local similarity-preserving knowledge distillation (LSPKD), which allows similarities between local neighbors to be transferred during the distillation. With our LSPKD, a lightweight student network is precisely guided by a heavy teacher network, regardless of the density of the ground-truth data. Experimental results demonstrate that our method is effective to reduce computational costs during both training and inference stages while achieving superior performance over other lightweight networks.

Few-shot learning is one of the most challenging problems in computer vision due to the difficulty of sample collection in many real-world applications. It aims at classifying a sample when the number of training samples for each identity is limited. Most of the existing few-shot learning models learn a distance metric with pairwise or triplet constraints. In this paper, we make initial attempts on learning local and global similarities simultaneously to improve the few-shot classification performance in terms of accuracy. In particular, our system differs in two aspects. Firstly, we develop a neural network to learn the pairwise local relationship between each pair of samples in the union set that is composed of support set and query set, which fully utilize the supervision. Secondly, we design a global similarity function from the manifold perspective. The latent assumption is that if the neighbors of one sample are similar to those of another sample, the global similarity between them will be high. Otherwise, the global similarity of the two samples will become very low even if the local similarity between them is high. Meanwhile, we propose a new loss according to the pairwise local loss and task-specific global loss, encouraging the model toward better generalization. Extensive experiments on three popular benchmarks (Omniglot, miniImageNet and tieredImageNet) demonstrate that our simple, yet effective approach can achieve competitive accuracy compared to the state-of-the-art methods.

Three formulations of the turbulence-length scales used in numerical modelling of atmospheric flows are compared. The comparison is made using the Mellor–Yamada–Nakanishi–Niino turbulence closure model within the stable boundary layer local similarity framework. With an appropriate choice of model constants, the model predictions are barely discernible and compare well with the empirical data obtained from the SHEBA campaign.

Residual useful life (RUL) prediction is the core of prognostics and health management. Similarity-based residual life prediction (SbRLP) is vital in RUL prediction due to its independence from degradation modeling, as well as high accuracy and robustness in prediction. However, researchers typically adopt a fixed time scale and global similarity search to perform similarity measurement, leading to considerable prediction errors and prolonged prediction times. Hence, a novel SbRLP method based on a dynamic time scale and local similarity search is proposed herein. First, the monitoring variables are reduced using the variable selection method based on multilayer information overlap. Next, the health states of reference samples are divided into five states using the K-means algorithm and the health states of the operating sample are recognized using the L-KNN algorithm. Further, dynamic time scales of the operating and reference samples are determined based on their length proportions of degradation trajectory at different prediction times. The local similarity search intervals of reference samples are obtained based on their health state levels. Next, the RULs of the operating sample are predicted using the local similarity search intervals and dynamic time scales. Finally, the effectiveness and superiority of the enhanced SbRLP are demonstrated using the commercial modular aero-propulsion system simulation dataset. The results reveal that the enhanced SbRLP yields a more accurate and efficient prediction of RUL in comparison with alternative methods.

With an unprecedented decade-long time series from a temperate eutrophic lake, we analyzed bacterial and environmental co-occurrence networks to gain insight into seasonal dynamics at the community level. We found that (1) bacterial co-occurrence networks were non-random, (2) season explained the network complexity and (3) co-occurrence network complexity was negatively correlated with the underlying community diversity across different seasons. Network complexity was not related to the variance of associated environmental factors. Temperature and productivity may drive changes in diversity across seasons in temperate aquatic systems, much as they control diversity across latitude. While the implications of bacterioplankton network structure on ecosystem function are still largely unknown, network analysis, in conjunction with traditional multivariate techniques, continues to increase our understanding of bacterioplankton temporal dynamics.

Normal moveout (NMO)-based velocity analysis can provide macro velocity models for prestack data processing and seismic attribute inversion. Datasets with an increasing size require conventional velocity analysis to be transformed to a more automatic mode. The sensitivity to multiple reflections limits the wide application of automatic velocity analysis. Thus, we propose an automatic velocity analysis method for seismic data-containing multiples to overcome the limit of multiple interference. The core idea of the proposed algorithm is to utilize a multi-attribute analysis system to transform the multiple attenuation problem to a multiple identification problem. To solve the identification problem, we introduce the local similarity to attribute the predicted multiples and build a quantitative attribute called multiple similarity. Considering robustness and accuracy, we select two supplementary attributes based on velocity and amplitude difference, i.e., velocity variation with depth and amplitude level. Then we utilize the technique for order preference by similarity to ideal solution (TOPSIS) to balance weights for different attributes in automatic velocity analysis. An RGB system is adopted for multi-attributes fusion in velocity spectra for visualization and quality control. Using both synthetic and field examples to evaluate the effectiveness of the proposed method for data-containing multiples, the results demonstrate the excellent performance in the accuracy of the extracted velocity model.

Self-supervised image representation learning is centered on constructing general features with good quality from unlabeled images. However, most recent works mainly focus on high-level semantics, ignoring the lower level features, which affects the generality of representations. Moreover, many methods do not take into account the adverse information in the features, which limits the performance of learned representations. Considering these issues, this paper proposes a novel multi-level self-supervised representation learning framework (MLSRL) via triple-way attention fusion and local similarity optimization to improve the generalization and quality of the learned representations. Concretely, multi-level features obtained from self-supervised learning are fused effectively by the triple-way attention fusion module, which captures the importance of features channel-by-channel and fuses the weighted features to make the learned representations more abundant and general. Furthermore, we consider the influence of redundant or irrelevant information on feature quality, and design a novel local similarity optimization strategy based on mutual information for optimizing the fused features, and thus improving the quality of representations. Experiments on public Cifar10, Cifar100, Stl10 and Tiny ImageNet datasets demonstrate the effectiveness of the proposed method in image representation learning.