Recent deep model pruning methods predominantly focus on large-scale datasets and typically require finetuning before deployment. However, in real-world applications, pruning is often necessary for scenarios with fewer classification categories, where finetuning must be avoided to preserve the model’s generalization ability. To address these challenges, we introduce a novel pruning method called Cluster-based Redundancy Elimination (CRE). Specifically, CRE represents each convolutional kernel as a point in a high-dimensional space. A distance-based strategy is then used to compute a clustering radius for each convolutional layer. Based on these radii, core point filters are selected for pruning, as they represent redundant information that can be captured by neighboring filters in the high-dimensional space. This approach eliminates the need for finetuning, thus preserving the generalization of deep models. Extensive experiments on five benchmark datasets with limited classification categories, across multiple model architectures, demonstrate the effectiveness of our method and its superiority over several state-of-the-art pruning techniques.

Deep learning has achieved significant success in stereo matching, with its training process often supervised by LiDAR measurements. However, the sparsity of real-world LiDAR data limits the ability of deep models to extract effective features from stereo images. To address this issue, a novel deep learning-based framework called sparse LiDAR point cloud supervised stereo matching (SLSM-Net) is proposed. Specifically, dense reconstruction of sparse single-frame point clouds is first designed to avoid the error introduction with the mergence of multi-frame point clouds. To effectively densify point clouds of objects in local areas, stereo images are utilized as supervision information to train the deep models. Furthermore, a coarse-to-fine structure of the deep model is designed for stereo matching. A self-supervised learning strategy, which employs a photometric consistency constraint, is second proposed along with fully supervised learning to obtain dense and precise supervision information. This stage generates coarse disparity maps from stereo images. Finally, to fully leverage the complementary characteristics of LiDAR and stereo cameras, multi-scale feature fusion of point clouds and stereo images is performed by a residual block, where the feature maps of point clouds are derived from the densification reconstruction. This stage refines the results. Experimental results indicate that SLSM-Net outperforms current state-of-the-art methods, demonstrating superior performance in stereo matching.

Current deep model pruning methods mainly focus on large datasets and often involve finetuning before deployment. However, in real-world applications, pruning is typically needed for open scenarios with few classification categories, where fine-tuning should be avoided to preserve the generalization of deep models. To address these issues, we propose a novel deep model pruning method called Cluster-based Redundancy Elimination (CRE). Specifically, CRE first represents each convolutional kernel as a point in high-dimensional space. Second, a distance-based strategy is employed to compute a clustering radius for each convolutional layer. Finally, based on these clustering radii, core point filters are selected for pruning, as they extract redundant information that can be captured by neighboring filters in the high-dimensional space. Thus, finetuning deep models to recover their performance can be removed. Comprehensive experiments on five datasets with few categories validate the effectiveness of our approach and demonstrate its superiority over several state-of-the-art pruning methods.

Iris recognition is widely regarded as one of the most reliable biometric identification technologies. Traditional methods, such as the Daugman algorithm typically normalize the annular iris region into a rectangular format during the preprocessing stage, followed by feature extraction and matching. However, these preprocessing steps often introduce distortions and struggle to adapt to multiresolution images, leading to inaccurate feature encoding. In response to these limitations, we propose a weak preprocessing algorithm for iris recognition that effectively preserves both grayscale and structural information of the iris. This approach is highly adaptable to varying image resolutions by leveraging a multiscale structural information extraction framework. It demonstrates significant improvements, achieving a matching accuracy of 96.67% on our proprietary dataset and 90% on the CASIA‐IrisV4 dataset. Compared to the Daugman and OsIris 4.0 algorithm using weak preprocessing schemes, our approach improves accuracy by 15.55% and reduces matching time by 16%. More importantly, this method presents a new idea that is different from traditional preprocessing methods with wider adaptability. It offers considerable potential for real‐world applications in security, with promising prospects for further integration with deep learning techniques.

The reconstruction of dense point clouds is an important foundation for downstream applications, such as object detection, semantic classification and surface reconstruction. Current methods focus on dense point cloud reconstruction for objects but neglect the whole scene. To address this issue, the reconstruction of dense point clouds supervised by inverse stereo matching (IS-Dense) is proposed. In detail, the Transformer model is first used to extract deep features form the point clouds. Second, point cloud features are expanded through the base upsampler. Ultimately, the point clouds would be coordinated following the feature expansion. Due to the uneven distribution of point clouds in the whole scene, some gaps and anomalies are presented in the data. Therefore, a point location refinement module supervised by inverse stereo matching is designed to solve this problem. For this module, the key is to utilize the reconstructed dense point clouds and the right image to estimate the left image. Supervised by real left images, the reconstructed dense point clouds are precise and even-distributed. The experimental results prove the superiority of the proposed method over current methods, especially for the whole scene.

The spectral super-resolution (SpeSR) from multispectral images (MSIs) to hyperspectral images (HSIs) can bring rich spectral information. The deep learning-based methods have demonstrated their powerful ability for the SpeSR task, which requires the paired HSI/MSI to train the model. However, HSIs and MSIs are always obtained at different times and under different imaging conditions, covering different areas. To address this issue, in this paper, a framework named BliEstGAN based on the generative adversarial network (GAN) is proposed to estimate the spectral resolution degradation between unpaired HSIs and MSIs that can be used for the blind SpeSR. Specifically, each MSI imaging sensor has its own unique spectral sampling process, which can be modeled as a spectral degradation from its paired HSI. Different spectral degradations can be discriminated by the deep model. Therefore, the generator of the GAN is used to estimate the spectral degradation from HSIs to MSIs, and the discriminator of the GAN is adopted to distinguish whether the estimated and real spectral degradation are similar. The large difference in spatial resolution between MSIs and HSIs makes them easy to discriminate against. Therefore, smooth hyperspectral and multispectral patches are extracted from HSIs and MSIs to eliminate this difference in spatial resolution. Furthermore, according to the imaging sensor mechanism, some special regularization terms are designed for the generator to guarantee its correct convergence. Finally, the estimated spectral resolution degradation can be adopted to generate HSI/MSI pairs for the supervised learning-based SpeSR methods. Experimental results demonstrate the effectiveness of the proposed method.

This article proposes an automated machine learning method for semantic segmentation that can be used for automated training of models in fields such as autonomous driving. This method is not specific to a particular semantic segmentation model. Users can simply upload a dataset with the semantic segmentation model they use, and then choose to use this method. This method implements end-to-end machine learning from sensing data to semantic segmentation results and model evaluation. It integrates four main components: unsupervised data reduction through feature extraction and clustering, a contrastive learning-based evaluator of semantic segmentation results, interactive reinforcement learning-based data selection, and automatic hyperparameter tuning through Bayesian optimization. We demonstrate the practicality of this method on the MRSI and Cityscapes datasets, and trained mainstream semantic segmentation models, such as BiSeNet and STDC. Our results show that this method can effectively guide semantic segmentation training and reduce the training time by more than 20$\%$.

Dense crowd counting has become an essential technology for urban security management. The traditional crowd counting methods mainly apply to the scene with a single view and obvious features but cannot solve the problem with a large area and fuzzy crowd features. Therefore, this paper proposes a crowd counting method based on high and low view information fusion (HLIF) for large and complex scenes. First, a neural network based on an attention mechanism (AMNet) is established to obtain a global density map from a high view and crowd counts from a low view. Then, the temporal correlation and spatial complementarity between cameras are used to calibrate the overlap areas of the two images. Finally, the total number of people is calculated by combining the low‐view crowd counts and the high‐view density map. Compared to single‐view crowd counting methods, HLIF is experimentally more accurate and has been successfully applied in practice.

The Internet of Things (IoT) will be widely used in all areas of life and transportation as the 5th Generation (5G) communication technology matures and becomes commercially available. Especially in the field of railway transportation, the IoT technology can alleviate the challenge caused by insufficient wireless spectrum resources and improve the railway communication performance. However, the existing IoT is made up of a large heterogeneous network. In such a super-dense heterogeneous network scenario, how to allocate the most appropriate access point (AP) according to the needs of users has become a problem demanding prompt solution, which also brings additional challenges for the intelligent transportation system (ITS) to develop green and efficient network communication technology. Therefore, focusing on the selection and access of heterogeneous networks in the Railway IoT, this article studies the spatial characteristics of the intelligent spectrum situation of the Internet of Mobile Things in the railway scenario, and establishes the opportunistic access situation of Railway IoT based on the graph convolutional neural (GCN) network. Furthermore, we utilize the GCN network to mine the spatial correlation between different APs, and propose a railway communication AP decision algorithm based on the GCN network combined with the traditional heterogeneous network multiattribute decision algorithm. Our experimental results prove that the proposed algorithm can effectively reduce transmission delay and improve the throughput of the communication system.

Cover Caption: The cover image is based on the Research Article MRSI: A multimodal proximity remote sensing data set for environment perception in rail transit by Yihao Chen et al., https://doi.org/10.1002/int.22801.

Rail transit is becoming a major mode of rapid urban and intercity passenger and freight transportation, and its safe operation is of great significance in safeguarding people's lives and properties and maintaining social stability. The current scheme of manual hazard monitoring in rail transit still remains potential safety risks. Accurate rail scene understanding is an essential step towards a smart train. Limited by the closeness of railway scenes, not much research has been conducted on the perception and understanding of rail transit. In view of the above, we propose multimodal remote sensing image (MRSI), the first multimodal proximity remote sensing data set for rail scene understanding. MRSI consists of 27k images collected from freight rail and metro following the pixel and box annotations labeled and checked manually. We used a variety of sensing devices mounted on locomotives to record track scenes under different lighting and weather conditions, including straight, curve, and fork during daytime, dusk, and nighttime, as well as under rainy days. We also include an additional infrared thermometer in the metro environment, propose a new image registration method after synchronous acquisition, and thus construct MRSI combining spatial and radiometric properties. With this data set, we can achieve segmentation of the track area and recognition of obstacles by sensing the environment in front of the train, which lead to rail scene understanding. MRSI is publicly available at https://zenodo.org/record/5732905#.YaPIpsdBwdU.

The Long Term Evolution for Railway (LTE-R) communication system is providing a reliable data link for High-Speed Railway (HSR) communication. However, when the train passes through different railway environments, the channel capacity of the base station and the number of users are always in highly dynamic changes. Therefore, accurate predicting the changing law of wireless spectrum resources can make more efficient use of wireless spectrum resources. The purpose of this paper is to use the Long Short-Term Memory network ($LST\!M$) to predict the channel occupancy changes of wireless spectrum resources. Under the premise of ensuring the safe and reliable service for primary users (PU), it provides a feasible method for the secondary user’s (SU) opportunistic access to the authorized channels, thereby improving theLTE-Rsystem Utilization rate of spectrum resources. Based on the “occupied/idle” status of the authorized channel at the previous$n$historical moments, we infer the status of the authorized channel at the current moment, build the spectrum situation of the authorized channels, and guide the SU to conduct Dynamic opportunistic Spectrum Access (DSA) to the authorized channels. The simulation results show that when SU uses the channel situation constructed by the$LST\!M$network to access the authorized channels, it has fewer handovers and lower collision rates, and can obtain higher throughput.

In order to realize the reliable and safe operation of the smart railways, and provide high quality information transmission service for passengers, the railway system needs to develop innovative communication network and advanced communication technology to meet the gradually increasing service demand of multi-dimensional comprehensive information resources. The Space-Air-Ground Integrated Network (SAGIN) can provide seamless information services for land, sea, air and space users, and is an effective solution to the challenge posed by the future smart railways to the all-time, all-domain, all-air, high-reliability and high-throughput communication. This paper aims to comprehensively discuss the technical development and application examples of High-Speed Railways (HSRs) based onSAGIN. Firstly, we analysis the development of theSAGINand the mobile communication network of theHSRs, and comprehensively discuss the single network architecture of the space-based, air-based and ground-based networks, as well as the integrated network, and discuss the application scenario and network structure of the combination of the integrated networks. At the same time, the communication services, existing problems and key technologies of the space-based, air-based and ground-based networks are discussed, and the application trend of theSAGINinHSRsis presented. Furthermore, the application scenarios of Artificial Intelligence (AI) technologies in solving the efficient resource utilization of smart railways communication and theSAGINare studied. Based on these technologies, we point out the research direction for the future development of AI technologies inSAGINinHSRscommunications.

In the evolution of railway mobile communications from Long Term Evolution for Railway (LTE-R) to the future 5th Generation Wireless System (5G), the rapid increase in the number of low-power base station nodes along the railway has brought more frequent handovers. The current handover parameter selection mechanism often relies on the on-site measured results in a limited number of discrete scenarios. It cannot deal with the continuous changing characteristics of the high-speed railway mobile communication environment, which leads to a serious lack of accuracy, adaptability and intelligence. This article hopes to construct a parameter-adaptive handover mechanism suitable for5Gin the high-speed railway dedicatedLTE-Rcommunication system. The mechanism first uses the interaction of Temporal-Difference(TD)-learning-based reinforced agents to obtain high-speed railway handover performance and network performance in different combinations of speeds and handover parameters, and continuously updates the accumulated rewards used to target optimization, obtaining a Discrete TD value cube with closely related handover performance. Further, based on the Discrete TD value cube, we use the approximation function method for the completion of “continuous” situation of handover parameter selection, and construct a continuous TD value cube and the corresponding performance cubes. Our experimental results prove that TD learning agents with function approximation can accurately estimate and predict the handover performance and network performance of state combinations with different speeds and handover parameters, and further show that the handover parameter adaptation mechanism based on the Inference ability can find the optimal handover parameters to improve the handover performance and network performance.

With the rapid development of railway transportation, railway communication is particularly important. In addition to meeting the communication needs of train control, it also guarantees the communication quality of passengers on board. In order to solve the problem of low utilization of spectrum resources in the special railway communication network, a spectrum sharing special railway network communication scheme based on the special railway communication network architecture and cognitive radio technology was put forward this paper. Furthermore, in order to improve the fairness of cognitive base station services, we proposed a secondary user queue management strategy based on Kalman Filter Prediction model to realize dynamic resource allocation. On the premise of satisfying the demand of real-time service for queuing delay, the cognitive base station adjusts the service rate of real-time service according to the predicted changes in the queue length, so as to minimize the queuing delay of NRT service. Finally, the simulation results show that the proposed method can reduce the queuing delay of NRT services, effectively improve the fairness of the cognitive base station services, and ensure the efficient utilization of the limited spectrum resources.

With the rapid development of high-speed railway, the mobile communication system for railway need to optimize the allocation of wireless communication resources according to the train motion, thus improving the quality of communication services. In this paper, we construct a chain base station (BS) network model based on LTE-R. A Gauss-Markov (GM) motion prediction model is used to simulate a period of train motion in the orbit, and a cell zooming (CZ) algorithm based on orbital motion prediction and load balance is proposed. The performance of our algorithm is verified by simulation results. When the CZ strategy is implemented or not implemented, the load rate, load balancing factor, blocking rate and throughput of the network model at different moments of the orbital motion are obtained. By comparing these performance indexes, it can be concluded that the proposed CZ algorithm improves these performance indexes.

The rapid construction of high-speed railway has brought great convenience to people's lives. However, in the case of emergencies such as natural disasters and network interruptions, it is urgent to quickly build emergency communication network to ensure the safety and reliability of railway communication. With the continuous improvement of the performance of unmanned aerial vehicle(UAV) network technology in terms of networking flexibility and communication reliability, it has been increasingly applied to the emergency communication. Based on this, we propose an emergency communication strategy to provide signal services by means of UAVs in case of communication signal failure of high-speed railway. In addition, the capacity of the relay UAV has been fully considered, so we propose an improved DSR routing protocol which chooses a better network route between the UAV providing signal services and the remote base stations. The high-speed railway emergency communication models of different scenarios are simulated by the platform of MATLAB GUI and we analyze the values of RSRP and SINR to evaluate the effectiveness and reliability of the proposed emergency communication networking scheme. The experimental results show that the proposed scheme meets the needs of emergency communication for high-speed railway.

As the high-speed railway (HSR) industry Internet-of-Things chain matures, HSR wireless communication technology has become an increasingly important research field. The efficient management of time-frequency resources for Internet-of-Things networking is the core issue of HSR wireless communication optimization. The lack of time-frequency resources in LTE-R is still severe. In this article, a new LTE-R time-frequency resource allocation optimization method based on the probabilistic graphical theory is proposed. Considering the regularity that high-speed trains always pass by the same geographical location in similar time periods, we can do some research on opportunistic spectrum accessibility in the existing LTE time-frequency resource algorithm. The probabilistic graphical theory is suitable for finding the appropriate communication access opportunity in an HSR environment. The simulation results show that our method can effectively improve the performance of various traditional LTE time-frequency resource allocation algorithms.

With the rapid development of China's high-speed railway, the traditional railway wireless communication system technology has been difficult adapting to the requirement, and is gradually being replaced by the LTE-R communication system. However, the LTE-R handover parameters selection mainly depends on historical experience, and there is no established theory or method. Therefore, research on adaptive optimization methods of handover parameters at different speeds has great significance on improving the handover performance of LTE-R systems for high-speed railway wireless communications. Combined with the environment adaptive ability of reinforcement learning, this paper proposes an adaptive optimization method based on the Q-Learning algorithm to achieve real-time estimation of the handover parameters of the LTE-R system. And based on these, we establish a performance situation map for handover parameters for different speeds, and use the generated “handover situation” to provide a basis for mobile users accessing opportunistic channels during the handover process, thereby improving handover performance. Our simulation results show that the optimized handover parameters can significantly improve the handover performance of the LTE-R system.

As the high-speed railway industry matures, higher requirements are put forward for the railway wireless communication, and the demand for spectrum resources is also increasing gradually. When the train is moving at a high speed, it will bring about the frequent handover of wireless communication networks, which will lead to the deterioration of wireless communication quality and even the dropping of calls. Therefore, based on the Cognitive Radio, we established the cognitive base station model to enable the base station to have cognitive functions in this paper. We also proposed a multiple base station cooperative reinforcement learning to achieve dynamic spectrum management. Cognitive base station can select the optimal channel for communication services by fusing interaction information between different cognitive base stations, thus reducing the failure of handover when the train crosses the cells. The experimental results showed that the proposed algorithm can improve the utilization of spectrum resources effectively.

A method based on Robust Principle Component Analysis (RPCA) technique is proposed to detect small targets in infrared images. Using the low rank characteristic of background and the sparse characteristic of target, the observed image is regarded as the sum of a low-rank background matrix and a sparse outlier matrix, and then the decomposition is solved by the RPCA. The infrared small target is extracted from the single-frame image or multi-frame sequence. In order to get more efficient algorithm, the iteration process in the augmented Lagrange multiplier method is improved. The simulation results show that the method can detect out the small target precisely and efficiently.

For large-scale iris recognition tasks, the determination of classification thresholds remains a challenging task, especially in practical applications where sample space is growing rapidly. Due to the complexity of iris samples, the classification threshold is difficult to determine with the increase of samples. The key issue to solving such threshold determination problems is to obtain iris feature vectors with more obvious discrimination. Therefore, we train deep convolutional neural networks based on a large number of iris samples to extract iris features. More importantly, an optimized center loss function referred to Tight Center (T -Center) Loss is used to solve the problem of insufficient discrimination caused by Softmax loss function. In order to evaluate the effectiveness of our proposed method, we use cosine similarity to estimate the similarity between the features on the published datasets CASIA-IrisV4 and IITD2.0. Our experiment results demonstrate that the T -Center loss can minimize intra-class variance and maximize inter-class variance, which achieve significant performance on the benchmark experiments.

Wireless communication plays a vital role in the operations of modern rail transportation. The rapid motion characteristics of the train make the wireless spectrum environment unstable and discontinuous. These uncertainties, coupled with the inherent scarcity of the spectrum, lead to inefficiencies in railroad wireless communications. The application of cognitive radio is becoming a cutting-edge research field in railway wireless communication. This paper first analyzes the physical infrastructure of the railway wireless communication network and determines base station as the key communication node in the railway environment, which can implement the cognitive radio technology. Reinforcement learning and agent theory are then used to construct a cognitive base-station model which is suitable for the railway wireless environment. Furthermore, according to the characteristics of the chain-like distribution and cascade operation of the cognitive base stations along the railway, the reinforcement base-station multi-agent system model is proposed, and the unique Dual ε - greedy mechanism is used to drive the learning of multi-agent system to avoid local optimization. Our experimental results prove that the model can significantly improve the probability of successful data transmission in the railway wireless communication network, and greatly reduce the number of wireless channel switching. In addition, the effect of Dual ε - greedy mechanism on communication performance is discussed. This reinforcement base-station multi-agent model in this paper provides a new idea for realizing the railway cognitive radio and comprehensively solves the problem of low spectrum efficiency of cognitive radio in rail transit.

With the development of intelligent driving technology, the unmanned technology of rail vehicles has also made great progress. How to determine the front limit region in the driving vehicle is a very important topic in the unmanned technology of the rail vehicle.In this paper, a method of Track-Starting-Point(TSP) identification for tramway based on integral algorithm is proposed in this paper. The method is based on the special morphological characteristics of the tramway, and takes the gray mean of column pixels as the processing basis, and uses integral operation to detect the TSP. In order to achieve accurate detection of TSP under various weather conditions, illumination conditions and terrain features, the accurate boundary extraction is finally achieved, providing a guarantee for subsequent tramway obstacle detection.

With the continuous development of the intelligent transportation industry, target tracking has become an important research direction. Under normal circumstances, due to the complex road environment and changing backgrounds, millimeter wave radar has more interference when detecting targets. In addition to the variety of targets in the road and the different scattering intensity of multiple parts, the interference of the flicker noise on the radar must be considered. The combination of these noises can affect the accuracy of radar measurement and even make the radar to lose the target for a short time. The paper constructs a target tracking model based on adaptive Sage-Husa Kalman filter algorithm to track radar signals. The algorithm can not only estimate the real-time state of the system, but also estimate and modify the parameters of the system and the statistical parameters of the noise, so that the system model is closer to the current real state of the system, thus improving the accuracy of the target tracking. Even if radar loses its target in a short time, the target tracking model can estimate the approximate value of the true value of the target. The experimental results show that this method can track the radar target accurately and estimate the position information of the lost target.

In order to solve the problem of spectrum waste in the LTE-Railway (LTE-R) system, the paper uses Cognitive Radio (CR) to improve the ability of spectrum sharing on Vehicle-to- Ground communication. By constructing a novel Cognitive Radio Network (CRN) in LTE-R system, the Cognitive LTE-R eNodeB (C-eNodeB) can work with Vehicle Gateway (VG) and allocate idle and wasted spectrum resources to the passengers communicating devices to improve spectrum utilization of LTE-R, without impacting train-ground communication. Aiming at the novel CRN architecture, a C-eNodeB Queue Management Strategy (QMS) based on Type of Service (ToS) value priority is proposed to reduce the Real-Time (RT) service delay of Secondary Users (SU) caused by FIFO QMS. The simulation results show that the proposed CRN effectively improves the spectrum utilization of LTE-R system and the C-eNodeB QMS based on the ToS value priority significantly reduce the delay of RT business of passengers.

Wireless communication is an important part of key technology in modern rail transit. The rapid trajectory of the train makes the wireless spectrum of rail transit environment instable, discontinuous and unpredictable. These uncertainties, coupled with the inherent scarcity of the spectrum, result in inefficient wireless communications for rail transit. In this paper, We first formulate spectrum management of railway cognitive radio as the distributed sequential decision problem. By using reinforcement learning and agent theory, we propose the cognitive base station model. Furthermore, a multi-cognitive-base-station cascade collaboration algorithm is described according to the characteristics of the chain distribution and cascade operation of base stations along the track. Finally, this paper evaluates the communication performance of test scenarios and proves that the system can significantly improve the probability of successful transmission and greatly reduce the number of wireless channel switching. This cognitive base station multi-agent system scheme provides a new idea for realizing the cognitive radio of the rail transit and comprehensively solving the problem of low efficiency of the wireless spectrum. The paper is also a typical case of artificial intelligence applied in the field of signal processing.