Knowledge Tracing (KT) is essential for tracking students' evolving knowledge states and predicting their future performance.While current graph-based methods focus on exerciseconcept relations, they often overlook the inherent group structures among students.Similarly, emerging LLM-based approaches rely on individual histories, lacking the broader context of group references and contrastive evidence.As a result, existing individual-isolation paradigms fail to provide stable predictions and evidencebased explanations.To bridge this gap, we propose Micro-Community Knowledge Tracing (MicroC-KT), a framework that incorporates learning micro-environments to provide social-cognitive anchors for KT.MicroC-KT identifies latent learning communities via hypergraph modeling and generates dual-granular summaries to facilitate community matching and peer retrieval.By extracting contrastive group evidence, the model prompts an LLM to generate both accurate answer predictions and verifiable analysis reports.Experiments on four public datasets demonstrate that MicroC-KT significantly outperforms state-of-the-art baselines in predictive performance while providing more reliable and evidence-based explanations.

Automated anomaly detection using unmanned aerial vehicles (UAVs) is critical for wind turbine inspection, yet its reliability is often compromised by cluttered backgrounds and dynamic environmental changes. To address these challenges, we propose a unified Depth-Guided Continual Adaptation Framework that operates in two complementary modes. In the routine operation mode, a Depth-Guided Foreground Segmentation (DTF) module leverages geometric priors to suppress background interference and isolate the blade region, while an Enhanced Attention Fusion Module (EAFM) employs asymmetric cross-attention to recalibrate RGB features using depth cues for more robust anomaly representation. In the maintenance mode, we further introduce a GRPO-based memory evolution strategy for adaptation under concept drift. Specifically, the routine detector remains unsupervised, whereas maintenance adopts a safety-constrained update scheme with a small historical anchor set rather than assuming a fully label-free adaptation process. This design enables the model to incorporate informative target-domain samples while preserving historical defect sensitivity. Experiments on the Blade0 dataset and the DTU benchmark demonstrate strong perception performance (F1-score 0.880, AUROC 0.925). In cross-domain adaptation, the proposed maintenance strategy improves target-domain F1 from 0.807 to 0.920 without degrading source-domain F1, showing its effectiveness for reliable long-term wind turbine inspection.

Simultaneous Localization and Mapping (SLAM) has emerged as a cornerstone technology in intelligent transportation systems (ITS) and autonomous robots, finding widespread applications in various scenarios and multiple platforms autonomous driving tasks. However, in environments with weak textures and motion blur, achieving efficient and robust visual-inertial SLAM remains challenging. Current research utilizes line features to enhance SLAM performance in such environments, but incurs difficulties in line extraction, structured scene representation, and computational overheads involved in joint optimization. To address these challenges, this paper introduces a novel SLAM framework with both high pose estimation accuracy and high back-end computational efficiency. Firstly, an intelligent spatial line integration method is proposed to effectively reduce data redundancy in joint optimization by leveraging the remarkable stability of long line segments in structured scenes, thereby enhancing the spatial consistency of structural lines. Secondly, to combat low-light and high-speed motion environments, an optical flow tracking accuracy verification method is designed to bolster the system’s tracking performance and robustness in complex scenarios. Finally, to relieve the substantial computational overhead arising from high-dimensional optimization parameters in bundle adjustment (BA), a multi-mechanism marginalization strategy is presented to enhance the accuracy and computational efficiency of BA, while also preventing scale explosion. Comparative evaluations against state-of-the-art algorithms on both the EuRoC MAV and TUM VI benchmark datasets demonstrate that the proposed framework significantly improves localization accuracy and joint optimization efficiency.

Camera-equipped unmanned aerial vehicles (UAVs) are extensively applied in various surveillance tasks, including building inspection, surface reconstruction, and so on, which often require comprehensive and efficient three-dimensional (3-D) object visual coverage. Due to limited onboard storage, great efforts had been devoted to achieving full coverage with minimal costs in terms of the size of a viewpoint set for taking pictures and the total energy consumptions for flight. However, existing studies usually adopted greedy strategies to generate every viewpoint set without considering the requirement of subsequent path planning, and simply calculate the flight distance of a UAV as an approximate metric for its energy consumption without considering the kinematics constraints and flight motion difference of the UAV. As a result, the final solution may far deviate from the global optimal solution. To this end, this paper establishes a tightly coupled optimization framework to jointly minimize viewpoint set size and energy consumption in UAV visual coverage tasks, which comprises two sequential subtasks: generating a viewpoint set with minimal size and inherent path sequences and planning energy-efficient trajectories between viewpoints. To address them, firstly, a path-aware viewpoint set optimization strategy is developed by leveraging overlapping field of view (FOV) gains between viewpoints for path guidance and adopting a novel tree-based search algorithm that balances global exploration and local convergence. Subsequently, a high-fidelity energy optimization scheme is proposed by fusing an energy consumption model with the UAV position-posture coupled control and trajectory smoothing, and devising a customized iterative solver. Extensive simulation results demonstrated that the proposed hierarchical framework generates smaller viewpoint sets with better overlapping FOVs. In addition, energy-efficient path planning significantly reduces UAV energy consumption by up to 89.84% compared to conventional distance-optimized path planning, simultaneously decreasing task execution time.

WiFi probes are crucial tools in applications such as crowd analysis and traffic monitoring. Due to indiscriminate sniffing of WiFi packets in the environment, probes generate a large amount of irrelevant data, increasing server-side processing overhead and presenting challenges for efficient data collection and real-time information processing. Moreover, a single wired transmission mode significantly limits the application scope and deployment efficiency of probes. In response, this paper presents a lightweight WiFi probe system combining a dual transmission mode with device classification capability. First, the system hardware, composed of a single STM32 and multiple ESP8266 modules, enables both wired and wireless data transmission modes through embedded programming. Second, a lightweight logistic and decision fusion model (LLDFM) is proposed, utilizing a time window and device library dynamic management mechanism to achieve real-time classification of fixed/temporary WiFi devices, which can effectively distinguish and discard application-irrelevant fixed device data. Finally, a server web platform was developed with real-time data sniffing, storage, multi-probe monitoring, and management functions to evaluate the system's practical performance. Results from extensive real-world experiments demonstrate an average fault-free operational time of over 168 hours, a fixed/temporary WiFi device classification accuracy of 97.19 %, and the ability to filter 30.45 % of irrelevant data, reducing the corresponding data transmission volume. These findings confirm the proposed system's reliability, efficiency, and superiority.

The wind turbine blade represents a pivotal element within the context of wind power generation systems, with its performance and reliability exerting a direct influence on the overall efficiency and safety of the system. The identification of defects in wind turbine blades is of considerable economic and safety significance. The existing YOLOv8 target detector is still inadequate in terms of loss function in the process of detecting minor defects in wind turbine blades, which affects the learning efficiency and model performance. Therefore, a new loss function is proposed, which is realised by combining scale sensitivity and cascade structure. The proposed method, implemented in the YOLOv8 detector, markedly enhances both the accuracy of detection and the computational efficiency. Moreover, wind turbine blade defect detection is typically implemented in edge devices with constrained performance. To address this issue, a lightweight convolution network is introduced, reducing the number of parameters by 60% to 1.12M, thereby achieving a balance between performance and efficiency. The experimental results on the DTU and the datasets constructed by our own research team demonstrate improvements in detection accuracy by 2.28% and 1.8%, respectively.

Recent progress in crowd counting has underscored its potential across diverse real-world applications. Nevertheless, the majority of existing approaches remain constrained by reliance on unimodal data, thereby limiting both robustness and generalizability. To address these challenges, we present a novel virtual simulation framework for generating synchronized multi-modal synthetic datasets. The proposed framework supports automated data generation with high-fidelity ground-truth annotations and provides programmatic control over environmental conditions and crowd dynamics. Through comprehensive experiments employing pre-training and fine-tuning strategies, we demonstrate that the synthetic datasets produced by our framework substantially improve model performance and generalization in crowd counting tasks. The work contributes a scalable and reproducible solution to the problem of data scarcity in multi-modal crowd counting research.

Effective detection of surface defects on wind turbine blades contributes to the continuous andrgetble operation of wind power equipment. It also facilitates the management and maintenance of equipment components. Wind farms contain a large number of wind turbines and surface defects vary in type and location, making the inspection process extremely complicated. In response to these issues, this paper proposes a wind turbine blade surface defect detection model named WTBFlaw-YOLO. The model is based on an improved version of YOLOv11. First, we designed a spatial-to-depth convolutional neural network block (SPD-Conv). We replaced every convolution stride and pooling layer in the convolutional neural network with SPD-Conv. This design improves WTBFlaw-YOLO’s ability to detect blade surface defects in low-resolution wind turbine images. Secondly, we designed a high-level screening feature pyramid network (HS-FPN) in the neck section. We achieved multi-level feature fusion with HS-FPN. This fusion enables WTBFlaw-YOLO to filter targets of various sizes and enhances its ability to express features at different scales. Finally, we designed a dedicated detection head for small defect target detection (SDTDH) in the head section’s multi-head detector. We used SDTDH to improve the detection precision of WTBFlaw-YOLO for micro defects on wind turbine blade surfaces. Experimental results demonstrate that WTBFlaw-YOLO achieves 89.7% mean average precision (using mAP50 as an example) on the public dataset (wind turbine-v2). The model outperforms other models. We conducted multiple ablation experiments. The experiments confirm the effectiveness and robustness of the WTBFlaw-YOLO model.

In recent years, crowd counting techniques based on channel state information (CSI) have received increasing attention. However, many existing WiFi-based methods directly input CSI magnitude or phase into traditional models such as convolutional neural networks (CNNs), which have limited ability to extract features, and the input data are sensitive to noise interference, thus limiting the effectiveness of crowd counting in high-interference situations. To overcome these drawbacks, we introduce the interference-resistant crowd counting system VICount, which uses time-frequency analysis to reflect the movement of people as a spectrogram. To further utilize the global temporal information, we produce a sequence of spectrograms to be input into the spatio-temporal converter network VICount, in order to effectively merge the features in the time, spatial and frequency domains. Extensive offline experiments show that VICount outperforms the SOTA method with an accuracy of up to 85% on the count task of 0-10 people. In addition, it achieves the accuracy of 74% presence detection in an online public dataset1and 61% cross domain recognition without training the target domain data.

Real-time awareness of crowd counts and distribution is vital for applications like crowd management, traffic control, and urban planning. Compared to vision-based methods, passive Wi-Fi sensing-based approaches offer advantages, such as lower deployment costs, broader coverage, and greater scalability, and thus have become prevalent for fine-grained crowd analysis, e.g., estimating mobile device locations and subsequently inferring crowd density maps (CDMs). Therein, substantial localization errors are often incurred due to significant measurement noises, and even worse, the sporadic active scanning introduces sparsity in localization-based density maps (DMs), degrading the accuracy of resulting crowd analysis. To address these challenges, Wi-FDM, a passive Wi-Fi sensing-based framework for fine-grained CDM estimation, is proposed. Specifically, to mitigate the considerable localization errors, two models based on convolutional neural networks are employed to capture the relationship between received signal strength fingerprints and various locations more effectively; to alleviate the sparsity in localization-based DMs, deep compressed sensing networks are developed to reconstruct fine-grained CDMs by recover missing spatial details; to efficiently transfer the CDM reconstruction capabilities developed on semi-synthetic datasets to diverse real-world scenarios, learnable linear and convolutional residual blocks were designed to establish correlations across environments, addressing variations in scene characteristics and improving model generalization. Experimental results on real-world datasets demonstrate Wi-FDM not only surpasses state-of-the-art methods by 41.8% and 29.3% in crowd counts and distribution estimation but is also applicable to both indoor and outdoor scenes, demonstrating its potential for crowd management and analysis.

The use of Unmanned Aerial Vehicles (UAVs) to perceive surface images and other information of complex three-dimensional (3-D) structures is a key component in infrastructure inspection tasks, such as for wind turbines and bridges. As a result, UAV path planning for 3-D structure coverage has attracted significant attention from researchers. However, existing studies primarily focus on minimizing the viewpoint connection path length under the full coverage condition, neglecting UAV flight constraints, which leads to a significant discrepancy between the planned path and the actual UAV flight trajectory. This paper presents an efficient UAV path-planning method for 3-D structure coverage using an improved ant colony optimization (ACO) algorithm. A multi-constraint UAV flight-trajectory-length minimization problem is formulated to guarantee full coverage of 3-D surfaces under kinematic limits, turning-angle and attitude-rotation constraints, thereby enhancing path consistency and smoothness. A cost function based on Minimum Snap trajectory planning replaces simple waypoint-connection length to more accurately reflect actual flight effort. A beta distribution-driven ant state transition rule and an adaptive pheromone evaporation strategy mitigate the tendency of traditional ACO to stall in local optima, balancing exploration and exploitation for rapid and reliable convergence. Simulation results demonstrate that the method proposed in this paper outperforms existing typical 3-D structure coverage path-planning methods, validating its feasibility and effectiveness.

T-cell acute lymphoblastic leukemia (T-ALL) is a malignant neoplastic disease. Accurate identification of core genes helps to explore the pathogenesis of T-ALL and develop relevant targeted drugs. In this paper, we first screened the RNA-seq, CTCF ChIP-seq and DNA methylation datasets of T-ALL for intersecting differentially expressed genes (DEGs) using bioinformatics software. As such, candidate genes screening is transformed into a semi-supervised classification problem given all known T-ALL related genes. It is clear that there are no labelled negative samples but a few labelled positive samples in the dataset, motivating us to employ the PU bagging method based on Positive-unlabeled (PU) learning to discover the T-ALL related candidate genes, in which a multi-layer perceptron (MLP) classifier was employed to accomplish the classification task. On this ground, a protein-protein interaction (PPI) network was built with the candidate genes, and core genes were screened. Finally, the core genes were subjected to GO and KEGG functional enrichment analysis, search of CTD and review of relevant literature to validate the proposed method. The validation results showed that the proposed method is able to effectively predict T-ALL related core genes, and all core genes have the potential to become candidates in T-ALL biomarker studies in the future.

Multi-rotor unmanned aerial vehicles (UAVs) have been widely employed in various sensing tasks, e.g., environmental monitoring and disaster rescuing, many of which often require full coverage of terrestrial regions by UAVs. Efforts have been devoted to minimizing one of two objectives, i.e., energy consumptions and time costs of UAVs fulfilling such tasks, whereas it is still challenging to jointly optimize both objectives due to their complicated interdependent relationship. Therefore, this paper deals with the tasks of sensing terrestrial regions with multiple UAVs, and focuses on the three-dimensional (3-D) coverage problem by formulating a multi-objective optimization problem of jointly minimizing both objectives. Specifically, in order to optimize energy consumption effectively, an advanced closed-form energy consumption model for multi-rotor UAVs is developed based on a rigorous theoretical analysis by introducing the influences of torque and acceleration, which are often ignored by existing heuristic models. Moreover, considering the NP-hardness of the problem, an innovative swarm intelligence optimization framework is established by leveraging a multitasking learning pattern to exploit cross-task knowledge transfer and adopting an improved multi-objective salp swarm algorithm. Therein, two novel operators, i.e., a variable characteristic-guided hybrid solution initialization operator and a large-scale search-space-oriented multi-mechanism solution update operator, are designed to handle continuous, discrete and even high-dimensional variables involved. Real-world experiments validate the proposed energy model due to the reduction of power consumption estimation error by up to 59% compared to baselines, and besides, extensive simulations demonstrate that the proposed algorithm significantly outperforms the benchmarks in terms of both energy consumptions and time costs.

Identifying essential genes in an organism is crucial for bioinformatics and genomics research. In this paper, we propose a new computational method for gene similarity prediction, named GT-KGCN, which combines the advantages of the Graph Transformer (GT), Graph Convolutional Network (GCN), and Kolmogorov-Arnold Network (KAN). Specifically, GT-KGCN utilizes graph transformers for feature embedding and multi-head attention aggregation, GCN for hierarchical aggregation of information from neighboring nodes, and KAN for optimizing feature learning through learnable activation functions. Model inputs include Protein-Protein Interaction (PPI) networks, gene expression profiles, subcellular localization, and orthologs information. We evaluated the performance of GT-KGCN on two different PPI network databases (BioGRID and STRING) using multi-omics datasets of Escherichia coli, Saccharomyces cerevisiae, Drosophila melanogaster, and Homo sapiens. The experimental results show that GT-KGCN achieves a prediction accuracy of 99.12% for Escherichia coli, 98.47% for Saccharomyces cerevisiae, 92.22% for Drosophila melanogaster, and 92.30% for Homo sapiens. The GT-KGCN model outperforms traditional shallow machine learning methods and other deep learning methods in essential gene prediction. The ablation study verifies the positive effects of multi-omics datasets as well as different convolutional layers on the prediction performance of the model, with gene expression data showing the most significant performance improvement and demonstrating the model’s superiority.

In wireless sensor networks (WSNs), energy constraint is a long-standing problem. With the development of energy-harvesting technology, energy-harvesting wireless sensor networks (EH-WSNs) have emerged. However, energy-harvesting wireless sensor nodes can be affected by their factors or external factors, making the energy collected by the nodes unequal. When individual nodes run out of energy in advance, the data will not be forwarded to the next hop node normally, thus generating data accumulation. To address this problem, this paper proposes a packet priority-based MAC protocol (DPP-MAC) for EH-WSN, which consists of two main parts: (1) Research on the channel allocation algorithm based on the remaining energy of nodes to avoid data accumulation during the transmission of data by low- energy nodes; (2) Stacking data transmission algorithm based on packet priority to solve the problem of node data stacking and improve network performance. By comparing with the existing MAC algorithm in terms of network throughput packet loss rate, average end-to-end delay, and channel utilization, it is found that the proposed DPP-MAC improves the network performance and better solves the data buildup problem.

Bluetooth 5 employs Constant Tone Extension (CTE) signals to simulate arrival times of a common signal at multiple antennas, but incurs limited accuracy due to inevitable frequency offsets caused by Gaussian frequency shift keying (GFSK) and time offsets caused by antenna switching. To address this issue, this paper proposes SCRN, an efficient Bluetooth 5 Angle of Arrival(AoA) estimation scheme that combines spectrogram analysis with deep learning co-design. SCRN leverages neural networks to generate a super-resolution spectrogram, effectively eliminating frequency offsets and recovering the true underlying frequency. Subsequently, it utilizes a lightweight convolutional recurrent network to encode the input features into a high-dimensional latent space, so as to alleviate the effect of time offsets. Through extensive experiments conducted in practical indoor settings, the proposed scheme achieves an AoA estimation error of only 1.3 degrees, surpassing traditional algorithms by over 90%.

Angle of Arrival (AoA) estimation based on Constant Tone Extension (CTE) signals in Bluetooth 5 suffers from carrier frequency offsets (CFO) arising from the carrier frequency mismatch between the receiver and the transmitter, extra noises coming from switching frequencies when sampling CTE signals, as well as the phase folding problem due to the usage of mod2π in mapping the increasing phase values in estimation models into [0, 2π). To address these issues, a generalized phase model involving the CFO compensation and the time offset is established, and an adaptive gradient descent optimization algorithm based on the real sampling mechanism is proposed to solve the phase folding problem efficiently while obtaining high-precision AoA estimation. Experimental results show that our method reduces the average estimation error by at least 10% in outdoor scenes and 61% in indoor scenes compared to the conventional methods.

Popular vision-based crowd counting methods suffer from huge costs, limited coverage and high complexity, making it difficult to be applied for large surveillance areas, while emerging WiFi-based methods which are suitable for large surveillance areas incur limited accuracy due to the sparsity and randomness of WiFi sniffing data. Considering the fact that the variations of audio data are spatial-temporally correlated with crowd fluctuations, this paper proposes to fuse audio and WiFi sniffing data for crowd counting by developing a Cross-modal Multi-level Perception Network, termed CMPN. The CMPN can not only extract crowd features from the bimodal data to leverage the temporally continuity for compensating sparse WiFi sniffing data, but also mine the correlation of intra- and inter-modality crowd features for accurate crowd counting. Extensive experiments are conducted in a real campus with the surveillance area of about 4000m2, and demonstrate that the CMPN can achieve the mean absolute error of 5.88, resulting in a 22.12% reduction compared to the state-of-the-art WiFi-only method.

Abstract In comparison with capturing channel state information (CSI) measurements via a laptop or desktop, using a smartphone to collect CSI measurements incurs the restriction of working with a single access point and significant signal distortions, resulting in limited information for smartphone localization. Therefore, this paper intends to leverage as much available localization information as possible by ($1$) shifting the WiFi frequency from $2.4$ to $5$GHz; ($2$) calibrating the noisy CSI measurements and ($3$) fusing both amplitudes and phases of the CSI measurements, so as to enhance localization accuracy. Specifically, we first filter out distorted CSI measurements based on their distribution characteristics, then apply the advanced uniform manifold approximation and projection method to refine the mapping relations from a high-dimensional fingerprint space to a low-dimensional location space, and design a location fusion algorithm based on the continuous feature scaling model, which is able to distinguish two locations with similar fingerprints. Extensive experimental results show that the localization accuracy of the proposed approach outperforms the state-of-the-art counterparts by at least $15.5$ and $18.7\%$ using two off-the-shelf smartphones.

For Unmanned Aerial Vehicles (UAVs) with Tiny Machine Learning (TML), there is mutual exclusivity between the energy consumption for flight and the energy consumption to support their computation and processing. IoUAVs integrated with TML systems often consume substantial amounts of energy during flights, particularly when engaged in extended coverage and surveillance missions. The energy consumption of a UAV with TML performing long, wide-area coverage patrols and monitoring missions in complex areas is significant for the flight itself, and the energy required for the TML to perform calculations and processing is not guaranteed. Therefore, to better support TML computations, this study optimizes flight paths to reduce the energy consumption of UAVs while ensuring coverage. Specifically, in this study, the use of concave point elimination algorithms, enhanced convex decomposition algorithms, and determination of flight direction significantly reduced the frequency of UAV turns. The computational cost of obtaining a complete path is reduced by merging the subconvex regions and the weighted minimum traversal of the graph. This novel bidirectional forwarding path coverage path-planning (BFP-CPP) algorithm maximizes the reduction in the number of turns, reduces energy consumption, and achieves global coverage. The simulation experimental results show that compared with the existing methods without concave point elimination, the BFP-CPP algorithm can effectively reduce the number of subregions, minimize the number of drone turns, and lower energy consumption.

Crowdsourcing dramatically benefits WiFi fingerprinting localization in reducing the costs of collecting received signal strength (RSS) data during offline site survey and has gained much attention in the literature. This article proposes a deep-learning-based indoor positioning system (IPS), termed SeqIPS, to sufficiently exploit the available information in the crowdsourced sequential RSS data and inertial measurement unit (IMU) data. However, there exist the following three challenges: 1) the relatively large label noises of crowdsourced RSS data; 2) the unavailability of the labels of crowdsourced IMU data; and 3) the incorporation of the knowledge in IMU data into the localization model using RSS measurement as inputs during online localization. To this end, a co-teaching network is developed to effectively extract spatiotemporal features from sequential RSS data and meanwhile alleviate the influence of label noises. Also, a novel loss function involving IMU data is defined to impose spatial penalties, so as to further refine the localization model. Moreover, a domain adaptation module is included to effectively label the crowdsourced IMU data. Extensive experiments are conducted in a real scenario and show that SeqIPS can achieve an average localization error of 3.37 m, outperforming both traditional methods and recent deep-learning-based methods by 18.4% at least. In summary, the novelty of SeqIPS is that extra crowdsourced IMU data is exploited to refine the localization model during offline training, while only sequential RSS data is required as inputs during online localization, such that the system accuracy, simplicity, and costs are reasonably balanced.

Visual odometry (VO) is a critical component of simultaneous localization and mapping (SLAM) with extensive applications in robot navigation and beyond. However, prevalent VO methods often underperform in intricate environments with dynamic textures, insufficient lighting, and rapid rotational movements, primarily due to constrained feature selection and inadequate image structure comprehension. To address these challenges, this paper proposes a novel VO framework, termed Multi-Attention Saliency Direct Sparse Odometry (MAS-DSO). Specifically, MAS-DSO significantly bolsters performance and robustness through accurate recognition of visually salient regions and deep understanding of image structures. With regard to the problem of limited feature selection, we propose a Saliency Transformer Generative Adversarial Network (STRGAN) based on a multi-attention mechanism, narrowing the feature selection scope and enhancing its accuracy. Addressing the issue of limited understanding of image structure, we introduce a robust method for gradient computation to accurately determine the gradient values of features. Building on this, we have designed a dynamic gradient weight adjustment strategy that takes into account both the gradient magnitude and local image structure, thereby achieving precise gradient weight distribution. Comprehensive quantitative evaluations on the ICL-NUIM and TUM monoVO datasets reveal that MAS-DSO not only outperforms SalientDSO, DSO, and ORB-SLAM in performance metrics but also significantly surpasses other methods in saliency prediction performance and mapping quality. In conclusion, MAS-DSO not only augments feature selection efficiency but also enhances the processing prowess for diverse images in complex settings.

Cooperative multiple unmanned aerial vehicles (UAVs) have been widely exploited in various applications, including data collection, forest monitoring, edge computing, and so on. Due to limited onboard storage and expensive hardware costs, reducing both the energy consumption and the number of UAVs is critical for these multi-UAV applications. However, existing studies primarily revolved around energy minimization in two-dimensional (2-D) scenarios, given a sufficient but fixed number of UAVs, and most of them considered specific application scenarios, resulting in poor generality. In contrast, this paper defines a generalized application scenario, in which multiple ground nodes (GNs) are accessed by multiple UAVs in three-dimensional (3-D) scenarios, and aims to minimize the energy consumption by employing the necessary (or equivalently minimum) number of UAVs and formulating a mix-integer nonconvex problem. To this end, this paper decomposes the problem into two subproblems: energy consumption minimization for a single UAV consecutively accessing any two GNs and energy-efficient multi-UAV GN-accessing path planning employing the minimum number of UAVs. The first subproblem is solved by applying the successive convex approximation (SCA) technique and the path discretization method, while the second subproblem is addressed by designing a three-stage approximation framework based on modified particle swarm optimization (MPSO) and greedy path assignment (GPA). Comprehensive simulations demonstrate the superior performance of the proposed method in terms of optimality and efficiency compared to several other counterparts.

Regarding the passive WiFi sensing based crowd analysis, this paper first theoretically investigates its limitations, and then proposes a deep learning based scheme targeted for returning fine-grained crowd states in large surveillance areas. To this end, three key challenges are coped with: to relieve the influences of the randomness and sparsity induced by passive WiFi sensing, an attention-based deep convolutional autoencoder model is designed to recover accurate crowd density maps in a way similar to image reconstruction; to combat the anonymity caused by MAC randomization, following the identification of local high-density crowds (LHDCs) with the density clustering algorithm, i.e. DM-DBSCAN, a bidirectional convolutional LSTM based model is employed to infer LHDC speeds; to overcome the absence of passive WiFi sensing datasets for model training, three semi-synthetic datasets are produced by emulating passive WiFi sensing with practical pedestrian tracking datasets. Extensive experiments confirm that, the proposed scheme significantly outperforms existing WiFi-based methods in terms of crowd density estimation and provides superior crowd speed estimation. More importantly, the scheme can also produce consistent crowd states on a real-world dataset, revealing that it has the ability to support accurate, visualized and real-time crowd monitoring in large surveillance areas.

Crowd counting aims to estimate the number of individuals in targeted areas. However, mainstream vision-based methods suffer from limited coverage and difficulty in multi-camera collaboration, which limits their scalability, whereas emerging WiFi-based methods can only obtain coarse results due to signal randomness. To overcome the inherent limitations of unimodal approaches and effectively exploit the advantage of multi-modal approaches, this paper presents an innovative WiFi and video-fused multi-modal paradigm by leveraging a heterogeneous dual-attentional network, which jointly models the intra- and inter-modality relationships of global WiFi measurements and local videos to achieve accurate and stable large-scale crowd counting. First, a flexible hybrid sensing network is constructed to capture synchronized multi-modal measurements characterizing the same crowd at different scales and perspectives; second, differential preprocessing, heterogeneous feature extractors, and self-attention mechanisms are sequentially utilized to extract and optimize modality-independent and crowd-related features; third, the cross-attention mechanism is employed to deeply fuse and generalize the matching relationships of two modalities. Extensive real-world experiments demonstrate that our method can significantly reduce the error by 26.2%, improve the stability by 48.43%, and achieve the accuracy of about 88% in large-scale crowd counting when including the videos from two cameras, compared to the best WiFi unimodal baseline.

Proteins are among the most essential molecules in the living body and are irreplaceable in various biological processes that sustain life. Predicting their functions is essential for understanding the molecular mechanisms of cellular life activities. With the widespread use of second-generation highthroughput sequencing technologies, more and more protein sequence data are being rapidly sequenced and shared. The rapid accumulation and update of these data provide a more substantial and diverse data foundation for protein function prediction research. Traditisonal biological experimental methods are the most reliable way to determine protein function. However, relying solely on manual experimental methods to test unknown protein functions individually suffers from a high workload and long lead time. Therefore, a deep learning-based protein function prediction model called DeepLA was proposed in this paper. Firstly, the model encodes protein sequences in One-hot and uses the Position-Specific Iterative Basic Local Alignment Search Tool (PSI-BLAST) algorithm to perform a comparative search to obtain a Position-Specific Scoring Matrix (PSSM) containing protein evolutionary information. Next, the protein feature vectors were fed into a multi-channel model consisting of a convolutional neural network, a bidirectional long- and short-term memory network, and a self-attentive mechanism for feature extraction to achieve the protein function prediction task. The results showed that DeepLA exhibited good performance in the Molecular Function(MF) and Biological Process (BP) categories with values of 0.556 and 0.488 on the publicly available dataset CAFA3, respectively, which were 5.8% and 8.0% higher than other models.

In recent years, with the increasing popularity of Internet technology and the advent of the 5G communication era, the applications related to mobile IoT have exploded, and wireless Ad Hoc network has received more and more attention. In traditional Ad Hoc networks, data are often transmitted between nodes in the form of "multi-hops", and the transmission range and available energy of nodes are limited. Thus, in Ad Hoc networks, a reliable and energy-efficient data transmission protocol is needed. In this paper, we studied the existing multipath routing protocols and proposed a multiple performances based routing protocol E2LR, ( residual Energy, Energy consumption, Loop, Routing protocol ) for the shortcomings of Ad Hoc networks. This protocol has the following improvements to the existing multi-path routing protocols: (1) In the route discovery process, depending on the energy level of the nodes, the nodes make three different decisions on the received RREQ packets: discard, randomly delayed forwarding, and immediate forwarding, to filter out the nodes with lower energy and thus improve the network performance. (2) When selecting available paths, try to select nodes whose lifetime are close to each other. This can reduce the number of interruptions in the transmission path and improve the stability of the transmission path during data transmission. (3) When transmitting data, multiple paths stored in the source node are recycled to transmit data, thus balancing the node energy. Simulation results show that the E2LR routing protocol can effectively reduce the frequency of route initiation, shorten the data transmission delay, reduce the packet loss rate in the network compared with existing multipath routing protocols.

Research on Energy Harvesting Wireless Sensor Networks (EH-WSN) has received much attention in recent years. However, in practical applications, due to the forced movement of sensor nodes in the deployment environment or the randomness of deployment, the dense and sparse distribution areas of nodes are formed in the monitoring area, and there may also exist isolated nodes. In the dense region, data conflicts are easily generated between nodes, while in the sparse region, the connectivity between nodes is low, the packet loss rate becomes high, and the data of isolated nodes cannot be transmitted normally. To address the above problems, this study proposes an adaptive MAC protocol in EH-WSN in harsh environments. This paper proposes an adaptive MAC protocol for EH-WSN in harsh environments, (AHE-MAC). The main idea of AHE-MAC is that: for the uneven distribution of nodes in harsh environments, this research firstly categorizes the node deployment area into “dense area and sparse area”. Secondly, different data transmission methods are proposed for different regions; in the dense region, the nodes implement the dynamic sleep mechanism, which puts some nodes to sleep and other nodes transmit the data to the next hop nodes, thus reducing the data transmission conflicts. In the sparse region, the isolated nodes adopt the “tentative power increasing“ mechanism to transmit data packets, and the other sparse nodes adopt the improved TDMA algorithm to transmit data, thus increasing the connectivity of the nodes. The simulation results prove that the protocol ensures the effective transmission of data in different network areas and improves the network performance at the same time.

Wireless sensor network (WSN) has many applications, such as, military scenarios, habitat monitoring and home security. In recent years, with the advancement of energy harvesting (EH) technology, nodes can obtain available energy from the surrounding environment for their own use, thus extending their lifetimes. Under these conditions, research aimed at improving the WSN lifecycle has further shifted towards improving the performance of the network, albeit subject to unique energy harvesting constraints. This paper proposes an energy prediction algorithm for the devices and an Energy and Density Adaptive Clustering (EDAC) protocol to improve network throughput and transmission ratio for EH-powered WSNs. Based on the EH characteristics, we first employed Convolutional Neural Network (CNN) and Bidirectional Long-Short Term Memory (Bi-LSTM) algorithm for energy prediction, then we divide the energy of the sensor nodes into three levels: low, medium, and high energy levels. At high energy levels, nodes can be selected as cluster head nodes, while at low energy levels, nodes must sleep and charge. EDAC first uses the K-Means clustering algorithm to dynamically cluster the surviving nodes in each round and sets a threshold to partition the clustering density. On this basis, a new adaptive cluster head election formula is proposed for cluster head election based on the energy levels of nodes, the predicted energy of the next stage, and the density of clusters. In the stable communication stage of the network, we introduce a "backup cluster head" to temporarily forward the remaining data packets within the cluster when the current cluster head expires. Our simulation results show that our algorithm significantly improves throughput and data transfer rate compared to the traditional and improved clustering protocols.

The positioning of wild animals usually requires long-distance, high-precision, low-energy positioning technology, and most of the existing methods use GPS, RFID, etc., either high power consumption, high cost, or small ranging range, low accuracy. This paper comprehensively considers the needs of out-door scene positioning, and combines the characteristics of LoRa technology and drones to solve the problem that it is difficult to achieve the best cost, energy consumption and positioning error in outdoor positioning. A sequential localization algorithm for two scenarios of single / multiple LoRa terminals is proposed. Based on the Cramer-Rao lower bound theory, the Unmanned Air Vehicle(UAV) trajectory is optimized to minimize the LoRa terminal positioning error at extremely low cost and energy consumption. The simulation results show that the proposed method effectively improves the accuracy of positioning LoRa terminals, and provides theoretical guidance for the deployment of single UAV as mobile base station in positioning.

Online public transit ridership information is helpful to enhance the service quality of urban public transportation and the travel experiences of passengers. Passive WiFi sensing collects WiFi probe (request) frames sent by nearby mobile devices in a non-intrusive manner, and can thus be employed to monitor ridership. Compared with the existing non-WiFi based approaches, passive WiFi sensing based approaches demonstrate the advantages of limited interferences, large coverage, low costs and lightweight calculations. More recently, although some dedicated passive WiFi sensing based methods have been proposed in an offline mode, due to sniffing opportunistically, unknown dynamic transmission boundary, MAC randomization and the difficulty in online feature extraction, how to utilize limited sensing data to provide accurate online ridership information is still challenging. To this end, an innovative public transit ridership monitoring system built upon a customized WiFi sniffer and an online ridership estimation algorithm is developed. In the algorithm, a convolutional neural network (CNN) module and a bidirectional long short-term memory (BiLSTM) neural network module are first adopted to find correlations among inputs and capture the bidirectional time-series features, respectively; furthermore, an attention module is incorporated to determine the importance of an input sequence at different times. Real-world experiments are carried out on 8 buses corresponding to 4 bus routes in Hohhot, China. The evaluation results show that the proposed algorithm outperforms the other 4 online algorithms and the state-of-the-art offline algorithm.

Great efforts have been devoted to solving the crowd counting problem based on vision or other fine-grained measurements. Popular vision and WiFi channel state information based approaches, though are able to achieve relatively high accuracy, suffer from limited scalability. In contrast, passive WiFi sensing-based approaches are capable of supporting large surveillance areas, but often rely on certain global linear or approximately linear regression models, which cannot accurately capture the complex mapping relationship between WiFi sensing data and the corresponding crowd count, especially in a large surveillance area during a long period of time. This paper addresses the issue from the following three aspects. Firstly, in order to combat with these coarse-grained regression models, the large surveillance is partitioned into grids, such that either a local linear model or other implicit local models can be built with respect to each grid. Secondly, sequential WiFi spatial-temporal matrix (SWSTM) is defined in alignment with grids to encode the spatial-temporal information of crowds based on passive WiFi localization and a sliding time window mechanism. Thirdly, the spatial-temporal correlations among crowd features of different grids are mined to better regress such local models by using a recurrent neural network (RNN) with SWSTMs as inputs. Extensive experiments are conducted in a real campus road network with an area of about$4000m^{2}$, and demonstrate that the proposed method significantly reduces the counting error rate from 22.54% to 13.44% compared to several state-of-the-art methods.

Passive WiFi localization refers to determining the location of WiFi-enabled mobile devices by deploying dedicated WiFi access points to sniff WiFi packets transmitted by these mobile devices and measure the corresponding received signal strengths (RSSs) for the use in localization. However, most existing studies fail to consider the effect of multiple channels where WiFi packets are transmitted and sniffed. The problem is further exacerbated by device heterogeneity occurring across various mobile devices. In this article, we present a unified deep neural network (DNN)-based solution, termed DHCLoc, to address these two challenges. To be specific, a Cramer–Rao lower bound (CRLB)-based analysis reveals that utilizing multichannel information will benefit localization, motivating us to include channel information into DHCLoc. Moreover, a novel maximum likelihood estimation (MLE)-based localization framework is introduced by incorporating a new variable to characterize the RSS measurement offsets caused by device heterogeneity, inspiring us to apply adversarial training to adopt such offsets against device heterogeneity. Extensive experiments using two real-world data sets are conducted, and show that, in comparison with several existing methods, DHCLoc can improve the localization accuracy by at least 25.2% and 25.8%, respectively.

With the increasing demand on Indoor Location-Based Services (ILBS), various positioning technologies had emerged in the past decades, and WiFi-based approach is one of the most promising ones. However, the existing WiFi localization methods fail to take into account the disparate influence of packets transmitted in different channels so as to inhibit the further improvement of localization accuracy. Therefore, we present CADNN: a Channel Adaptive WiFi localization method based on Deep Neural Network (DNN). Specifically, a comprehensive analysis on signal attenuations in different channels along with error analysis based on Cramer-Rao Lower Bound (CRLB) is conducted in theory. Then, the channel set splitting scheme for practical localization to leverage multi-channel features is proposed. Finally, we design a localization framework using multi-objective regression DNN to adapt Received Signal Strength (RSS) measurements from different channel sets. The results from real-world experiments confirm the effectiveness of channel adaptive and show that CADNN can improve localization accuracy by at least 25.3% and 19.5% respectively on the two datasets and it can serve thousands users within one second.

With the convenience brought by Location‐based service (LBS), users’ requirements for indoor positioning accuracy are getting higher than ever. However, many traditional indoor WiFi positioning methods may result in limited positioning accuracy because of the limited information of Received Signal Strength (RSS) of WiFi signal. This paper proposed a context‐aware assisted WiFi positioning method (CAA‐PM), which uses context information (i.e., light and sound) to assist WiFi‐RSS for indoor positioning and uses an improved variable weight dynamic KNN fingerprint identification algorithm (VWD‐KNN). Finally, experiments are carried out by using the dataset collected in both a closed laboratory and an open long corridor, and it is shown that the proposed algorithm substantially improves the localization accuracy comparing with other three classical algorithms.

Energy Harvesting Wireless Sensor Networks (EHWSNs) have been widely studied in recent years. In solar charged EH-WSNs, the Sun illumination changes with the changes of environment, in consequence the collected energy of the sensor node is unstable, especially in rainy day, windy day or the angle of the solar panel changes. Therefore, the reasonable assignment of energy in EH-WSNs becomes critical important. In This paper, based on the solar energy charateristics, we propose a priority task scheduling algorithm that suitable for EH-WSNs, that is, the transmission method and order of collected data are determined according to task priority and the remaining energy of the node. The simulation results show that the priority task scheduling algorithm guarantees the fairness of node energy distribution, the timeliness of sending urgent tasks and the high processing rate of common tasks when the energy provided by the environment is small.

WiFi fingerprint-based localization is one of the most attractive and promising techniques targeted for indoor localization, and has attained much attention in the past decades. In addition to improving localization accuracy, various efforts have been devoted to efficiently building a radio map which is normally tedious and laborious. Therefore, this paper proposes an efficient approach for building compact radio maps based on compressed multivariate kernel density estimation (CMKDE), in the sense that only a few received signal strength (RSS) measurements are required and the resulting radio maps are far less than the sizes of traditional radio maps. Extensive experiments are carried out in a real scenario of nearly 1000 m2during several working days, and a comparison is made with two existing popular solutions including the Gaussian process regression (GPR) and another approach based on kernel density. It is shown that the proposed method outperforms its counterparts in terms of both robustness and accuracy.

With the rapid development of the Internet and combined with outsourcing, a new paradigm - crowdsourcing which shines brilliantly as a new labor mode. However, the existing pricing strategies for crowdsourcing tasks have several undesirable problems, e.g., no universal pricing model, not meeting the multiple requirements of users, pricing rely too much on decision makers, etc., which bring an unreasonable allocation of task rewards so as to make the pricing results subjective and uncontrollable. Therefore, this paper proposes a reverse pricing mechanism based on VCG and quality all-pay for collaborative crowdsourcing (VCG-QCP). The actual crowdsourcing scenario is considered with VCG mechanism, and the concept of quality all-pay is introduced to evaluate the work quality of workers who might perform the task. Then a general reverse pricing model is established by mathematical modeling, and the pricing algorithm is designed based on this model. Simulations show that the proposed method can achieve higher algorithm efficiency, higher task completion quality, a reasonable balance of benefits between employers and workers, and ensuring the truthfulness of workers' bidding.

The proliferation of mobile devices, including smartphones and tablets, has been enabling new possibilities for inferring information about the positions, behavior and activities of the users carrying these devices. For instance, by leveraging the WiFi probes sent out by mobile devices in public spaces (such as shopping malls, metro stations, etc.), even if pedestrians do not have their mobile devices to be associated with any WiFi access point (AP), it is attractive to conduct pedestrian analysis in a passive sensing approach to facilitate the efficient management of public infrastructures as well as convenient customer services. This paper considers the problem of pedestrian flow analysis by implementing a pedestrian surveillance system in the transfer channel of a metro station in Guangzhou China. Firstly, a fingerprint database is generated through a Gaussian process regression (GPR) approach. On these grounds, a pedestrian number estimation method based on linear regression is presented by making use of the fingerprint-based localization method to refine the number of mobile devices residing in the surveillance area, and a pedestrian velocity estimation method is proposed based on particle filter and the inverse distance weighted (IDW) method. According to the dataset obtained in real scenarios, the effectiveness and advantages of the proposed two methods are confirmed.

In this paper, a novel scheme is reported to adapt radio maps to environmental dynamics in an online fashion by combining crowdsourcing and gaussian process regression (GPR). Specifically, a Marginalized Particle Gaussian Process (MPGP) is adopted to recursively fuse crowdsourced fingerprints with an existing offline radio map. The advantages of the proposed scheme lie in the efficiency and scalability in comparison with the traditional approaches. Extensive experiments are carried out in a real scenario of nearly 1000 m2during five months, and a comparison is made with several existing popular solutions. It is shown that the proposed scheme outperforms its counterparts in terms of both robustness and accuracy.

Fingerprint-based localization relies on an accurate and up-to-date radio map, which is however cumbersome to obtain. In this paper, a novel scheme is proposed to online adapt radio maps to environmental dynamics by using low-cost crowdsourced received signal strength (RSS) measurements. To be specific, a coarse-grained radio map is initially established in the offline phase utilizing the standard Gaussian process regression (GPR) given a limited number of fingerprints (i.e., RSS measurements with location labels), and further can be recursively refined in the online phase given crowdsourced RSS measurements with their noisy location labels obtained through the existing radio map. Differently from existing GPR-based approaches, the proposed scheme adopts extended GPR to alleviate the model inaccuracy induced by such noisy location labels, and then presents a marginalized particle extended Gaussian process (MPEG) to recursively filter the radio map. In addition, pedestrian dead reckoning (PDR) is leveraged to calibrate such noisy location labels. Extensive experiments are carried out in a real scenario with area of nearly 1000 m2during a five-month period of time, and a thorough comparison with several existing approaches indicates that the proposed scheme gradually improves the localization accuracy on average by as much as 31.2%, while the counterparts result in fluctuant localization performance and improve the localization accuracy on average by 13.3%.

Mobile Crowd Sensing Network (MCSN) is a new sensing mode in the Internet of Things by taking advantage of individual's mobile device with multiple sensors to collect some specific data for certain applications. There are two modes of transmission mechanism in the existing systems: one is end-to-end mode via the cellular network and the other is opportunistic transmission via short-range wireless communication technology. Due to the higher cost, the former is not conducive to enlarging users'participation. This paper focuses on the mode of the latter to build a new opportunistic transmission in MCSN. Differently from most existing studies, i.e. preference-aware and energy-aware, this paper proposes an opportunistic data transmission mechanism based on SSIS model (a socialization SIS epidemic model), who can transfer the sensing data to the platform by forwarding step by step without extra cost. Specifically, firstly, SSIS model is defined based on SIS epidemic model by social information in MCSN. Additionally, SSIS model is used to analyze the social relationship of mobile sensing nodes in MCSN to obtain a social relational table. Finally, the social relational table is used to improve the Spray and Wait (SW), which is a typical opportunistic transmission mechanism, to guide the source node which performs the task of sensing to propagate the sensing data selectively to other nodes until it reaches the destination node which can send the data to the platform. Simulation results show that the performance of data transmission can be further improved by using the proposed mechanism in comparison with SW, Epidemic and Prophet.

Currently, WiFi (or IEEE 802.11) enabled infrastructures and devices have become ubiquitous, which makes it promising to provide indoor positioning services via WiFi signals. However, most existing studies on its performance are carried out based on simulations and experiments, and it is still challenging to mathematically characterize the localization error. Therefore, in this paper, the Cramer-Rao lower bound (CRLB) for the localization error by using WiFi signals is established and enables us to carry out a theoretical analysis on the fundamentals of WiFi based localization. Extensive experiments based on the well-known WiFi fingerprint-based localization are then carried out and confirm the correctness of the proposed CRLB model as well as the analysis.

Acquiring the knowledge of WiFi access point (AP) locations not only plays a vital role in various WiFi related applications, such WiFi-based indoor localization, the deployment of new WiFi APs, and so on, but also contributes to the emergence of novel applications. Most existing studies assume the well-known lognormal shadowing model, which only reflects large-scale fading in WiFi signal propagations but ignores small-scale fading induced by pervasive multipath effects. In this paper, we tackle the problem of AP localization based on the Rayleigh lognormal model which characterizes the influence of both large-scale and small-scale fading. Provided that a participant holding a smartphone is walking along a path and the smartphone automatically and continuously collects received signal strength (RSS) measurements from a target AP at known positions, particle filtering is applied to sequentially narrow the scope of possible locations of as well as the propagation parameters of the wireless signals emitted by the target AP, and the weighted mean of all candidate locations is returned as its final location estimate. Extensive experiments are carried out in typical indoor and outdoor scenarios, and reveal that the proposed method outperforms the solutions based on the lognormal model by 14.13%-70.38%.

Radio maps play a vital role in fingerprint-based indoor positioning systems (IPSs) in terms of the localization accuracy and computational overheads. Most existing studies either directly eliminate redundant APs or adopt unsupervised dimension reduction methods, say principal component analysis (PCA), to obtain a low-dimension representation of fingerprints, which consumes less storage and computational overheads. In this paper, we propose to reduce the dimensions of radio maps based on the Gaussian Process Manifold Kernel Dimension Reduction (GPMKDR) which is a supervised dimension reduction technique in comparison with the well known PCA-based method. Specifically, GPMKDR is employed to find a nonlinear and optimal embedding into the received signal strength (RSS) sample space during the offline phase, such that any RSS sample vector obtained in the online localization phase can be projected onto the optimal subspace with a lower dimension, with the result that the fingerprint-based localization can be efficiently realized based on a low-dimension radio map. Experiments show that the nonlinear GPMKDR-based method significantly improves the localization performance in comparison with the PCA-based method.

The Internet of Things aims to provide the user with deep adaptive intelligence services according to the user’s personalized characteristics. Most of the characteristics are presented in the form of high-level context. But it often lacks methods to obtain high-level context information directly in the Internet of Things. In this paper, so as to achieve the corresponding high-level context information using the specific low-level multidomain context directly obtained by different sensors in the Internet of Things, we present a machine learning method to construct a context fusion model based on the feature selection algorithm and the multiclassification algorithm. First, we propose a wrapper feature selection method based on the genetic algorithm to obtain a simpler and more important subset of the context features from the low-level multidomain context, by defining a suitable fitness function and a convergence condition. Then, we use the decision tree algorithm which is a multiclassification algorithm, based on the rules obtained by training the subset of context features, to determine which high-level context the record set of the low-level context information belongs to. Experiments confirm that the model can be used to achieve higher classification accuracy without more significant time consumption.

Most existing indoor positioning systems (IPSs) adopt the WiFi fingerprinting technique to localize WiFi enabled mobile devices. In this paper, we improve the WiFi fingerprinting based IPS by efficiently combining the universal Kriging (UK) interpolation method, K nearest neighbor (KNN) and naive Bayes classifier (NBC). Specially, the proposed IPS takes into account the comprehensive features of received signal strengths (RSSs) by adopting the UK method and area partitioning, and further mitigates the boundary effect (which deteriorates the localization accuracy when dealing with mobile devices near the boundary) by virtually augmenting the space boundary. Finally, NBC and weighted KNN (WKNN) are integrated to work with the interpolated fingerprint database. Extensive experiments are carried out in our lab, and show that the proposed IPS with 28 observation points is able to achieve the average positioning error of 1.265m, which is less by 46.6% than the counterparts of the traditional IPS with 28 observation points and is even comparable to the traditional IPS with 112 observation points.

3D city models typically consist of thousands of buildings in different types. We usually reconstruct these buildings automatically from high-resolution satellite or airborne imagery. However, for detailed roof reconstruction, 2D information offered by imagery data is not enough while DSM data is necessary. In this paper, we propose a novel level set framework for 3D building models in LOD2 with geometry structure of typical roofs. Local information is introduced towards multiphase and multichannel level set method. Its energy function is minimized when each part of roof data corresponds to the same normal vector as feature values for level set segmentation. The advantage of this method is that for complex building models, roof primitives as well as roof topology graph can be extracted from high-resolution DSM data with high accuracy, evaluated by completeness of segmentation and RMSE of 3D reconstruction. Thus, LOD2 building models can be reconstructed automatically with good performance. The very promising experimental results demonstrate the potentials of our method for large-scale building reconstruction in LOD2.