Numerous data-driven Bluetooth-based location methods have demonstrated favorable location outcomes in recent years. However, the robustness of the location model has not been thoroughly investigated. This study addresses this gap by conducting simulation tests that involve altering environment and presents a crowdsourcing-based model updating approach. We first introduced a Convolutional Neural Network (CNN) as a location basic model and constructed a spatial covariance matrix image as the network's input. We formulate the localization task as a regression problem and train the CNN to predict beacon position in multipath indoor environments. Based on this model, we introduced a Reconstruction-based Regression Network (RBRN) that updates the feature extractor parameters in the CNN localization basic model by extracting the distributional features of crowdsourced data to adapt to the dynamic fluctuations of the environment. Experimental results demonstrate a notable improvement in the positioning performance of the CNN location basic network compared to state-of-the-art Bluetooth positioning methods. Additionally, in the face of dynamic environmental changes, the crowdsourcing-based RBRN model updating method reduces the average positioning error by 15% ~ 18%, confirming the effectiveness of the proposed model updating method.

Hyperspectral images (HSIs) can finely discriminate distinct objects with a high spectral resolution, and they are widely employed in various applications. However, mixed noise severely degrades the quality of HSI and restricts the performance of subsequent tasks. As one of the critical pre-processing steps, HSI denoising has been developed rapidly, among which low-rank (LR) prior-based methods have achieved superior performance. Nevertheless, the existing approaches frequently fail to completely remove noise and reconstruct high-quality HSIs when tackling complicated mixed noise with multiple types of high-intensity stripe noise. To solve this problem, we propose an HSI denoising and destriping method based on anisotropic spatial and spectral total variation regularized double low-rank approximation (ATVDLR). The double low-rank approximation framework is devoted to separating the clean image from the mixed noise by exploiting both the global correlations of HSI tensor and the LR structure of stripe noise. Furthermore, the anisotropic spatial and spectral total variation regularization is introduced to preserve the spatial–spectral smoothness of HSI and the directional feature of stripes, thereby further suppressing high-level stripes and Gaussian noise. Finally, the alternating direction method of multipliers (ADMM) technique is designed to solve the proposed ATVDLR model. Extensive experimental results indicate that the proposed method outperforms other state-of-the-art techniques in multi-type high-intensity mixed noise reduction and image structural information protection, and has superior performance in complex mixed noise removal of real Gaofen-5 HSIs.

Hyperspectral images (HSIs) have a wealth of applications in many areas, due to their fine spectral discrimination ability. However, in the practical imaging process, HSIs are often degraded by a mixture of various types of noise, for example, Gaussian noise, impulse noise, dead pixels, dead lines, and stripe noise. Low-rank matrix decomposition theory has been widely used in HSI denoising, and has achieved competitive results by modeling the impulse noise, dead pixels, dead lines, and stripe noise as sparse components. However, the existing low-rank-based methods for HSI denoising cannot completely remove stripe noise when the stripe noise is no longer sparse. In this article, we extend the HSI observation model and propose a double low-rank (DLR) matrix decomposition method for HSI denoising and destriping. By simultaneously exploring the low-rank characteristic of the lexicographically ordered noise-free HSI and the low-rank structure of the stripe noise on each band of the HSI, the two low-rank constraints are formulated into one unified framework, to achieve separation of the noise-free HSI, stripe noise, and other mixed noise. The proposed DLR model is then solved by the augmented Lagrange multiplier (ALM) algorithm efficiently. Both simulation and real HSI data experiments were carried out to verify the superiority of the proposed DLR method.

MOTIVATION: Microbial metabolic interactions impact ecosystems, human health and biotechnology profoundly. However, their determination remains elusive, invoking an urgent need for predictive models seamlessly integrating metabolism with evolutionary principles that shape community interactions. RESULTS: Inspired by the evolutionary game theory, we formulated a bi-level optimization framework termed NECom for which any feasible solutions are Nash equilibria of microbial community metabolic models with/without an outer-level (community) objective function. Distinct from discrete matrix games, NECom models the continuous interdependent strategy space of metabolic fluxes. We showed that NECom successfully predicted several classical games in the context of metabolic interactions that were falsely or incompletely predicted by existing methods, including prisoner's dilemma, snowdrift and cooperation. The improved capability originates from the novel formulation to prevent 'forced altruism' hidden in previous static algorithms while allowing for sensing all potential metabolite exchanges to determine evolutionarily favorable interactions between members, a feature missing in dynamic methods. The results provided insights into why mutualism is favorable despite seemingly costly cross-feeding metabolites and demonstrated similarities and differences between games in the continuous metabolic flux space and matrix games. NECom was then applied to a reported algae-yeast co-culture system that shares typical cross-feeding features of lichen, a model system of mutualism. 488 growth conditions corresponding to 3221 experimental data points were simulated. Without training any parameters using the data, NECom is more predictive of species' growth rates given uptake rates compared with flux balance analysis with an overall 63.5% and 81.7% reduction in root-mean-square error for the two species respectively. AVAILABILITY AND IMPLEMENTATION: Simulation code and data are available at https://github.com/Jingyi-Cai/NECom.git. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: In a genome-scale metabolic model, the biomass produced is defined to have a molecular weight (MW) of 1 g mmol-1. This is critical for correctly predicting growth yields, contrasting multiple models and more importantly modeling microbial communities. However, the standard is rarely verified in the current practice and the chemical formulae of biomass components such as proteins, nucleic acids and lipids are often represented by undefined side groups (e.g. X, R). RESULTS: We introduced a systematic procedure for checking the biomass weight and ensuring complete mass balance of a model. We identified significant departures after examining 64 published models. The biomass weights of 34 models differed by 5-50%, while 8 models have discrepancies >50%. In total 20 models were manually curated. By maximizing the original versus corrected biomass reactions, flux balance analysis revealed >10% differences in growth yields for 12 of the curated models. Biomass MW discrepancies are accentuated in microbial community simulations as they can cause significant and systematic errors in the community composition. Microbes with underestimated biomass MWs are overpredicted in the community whereas microbes with overestimated biomass weights are underpredicted. The observed departures in community composition are disproportionately larger than the discrepancies in the biomass weight estimate. We propose the presented procedure as a standard practice for metabolic reconstructions. AVAILABILITY AND IMPLEMENTATION: The MALTAB and Python scripts are available in the Supplementary Material. CONTACT: [email protected] or [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.