Search

Recent advancements in deep learning have significantly improved performance in the multi‐view stereo (MVS) domain, yet achieving a balance between reconstruction efficiency and quality remains challenging for learning‐based MVS methods. To address this, we introduce MFE‐MVSNet, designed for more effective and precise depth estimation. Our model incorporates a pyramid feature extraction network, featuring efficient multi‐scale attention and multi‐scale feature enhancement modules. These components capture pixel‐level pairwise relationships and semantic features with long‐range contextual information, enhancing feature representation. Additionally, we propose a lightweight 3D UNet regularization network based on depthwise separable convolutions to reduce computational costs. This network employs bi‐directional skip connections, establishing a fluid relationship between encoders and decoders and enabling cyclic reuse of building blocks without adding learnable parameters. By integrating these methods, MFE‐MVSNet effectively balances reconstruction quality and efficiency. Extensive qualitative and quantitative experiments on the DTU dataset validate our model's competitiveness, demonstrating approximately 33% and 12% relative improvements in overall score compared to MVSNet and CasMVSNet, respectively. Compared to other MVS networks, our approach more effectively balances reconstruction quality with efficiency. [ABSTRACT FROM AUTHOR]

Trained deep reinforcement learning (DRL) based controllers can effectively control dynamic systems where classical controllers can be ineffective and difficult to tune. However, the lack of closed‐loop stability guarantees of systems controlled by trained DRL agents hinders their adoption in practical applications. This research study investigates the closed‐loop stability of dynamic systems controlled by trained DRL agents using Lyapunov analysis based on a linear‐quadratic polynomial approximation of the trained agent. In addition, this work develops an understanding of the system's stability margin to determine operational boundaries and critical thresholds of the system's physical parameters for effective operation. The proposed analysis is verified on a DRL‐controlled system for several simulated and experimental scenarios. The DRL agent is trained using a detailed dynamic model of a non‐linear system and then tested on the corresponding real‐world hardware platform without any fine‐tuning. Experiments are conducted on a wide range of system states and physical parameters and the results have confirmed the validity of the proposed stability analysis (https://youtu.be/QlpeD5sTlPU). [ABSTRACT FROM AUTHOR]

Image hiding is a task that hides secret images into cover images. The purposes of image hiding are to ensure the secret images are invisible to the human and the secret images can be recovered. The current state‐of‐the‐art steganography methods run the risk of secret information leakage. A safe image hiding network (SIHNet) is presented to reduce the leakage of secret information. Based on some phenomena of image hiding methods which use invertible neural network, a reversible secret image processing (SIP) module is proposed to make the secret images suitable for hiding and make the stego images leak less secret information. Besides, a reversible lost information hiding (LIH) module is used to hide the lost information into the cover images, thus the method can recover the secret images better than the method that uses random noise to replace the lost information. Experimental results show that SIHNet outperforms other state‐of‐the‐art methods on the PSNR and SSIM values of the recovered secret images and the stego images. Besides, residual images of other state‐of‐the‐art methods all contain information about secret images while residual images of SIHNet leak almost no secret information. Thus the method can prevent the listener of transmission channel from obtaining the information of the secret image through the residual image, which means SIHNet performs better in security than other state‐of‐the‐art methods. [ABSTRACT FROM AUTHOR]

Due to differences in the quantity and size of observed targets, hyperspectral images are characterized by class imbalance. The standard deep learning classification model training scheme optimizes the overall classification error, which may lead to performance imbalance between classes in hyperspectral image classification frameworks. Therefore, a novel factor annealing decoupling compositional training method is proposed in this paper. Without requiring resampling or reweighting, it implicitly modulates the training process, so standard models can sufficiently learn the representation of the minority classes and further be trained as robust classifiers. Specifically, the label‐distribution‐aware margin loss is combined with the error‐rate‐based cross‐entropy loss via combination factor, which considers both imbalanced data representation learning and classifier overall performance. Then, a factor annealing optimization training scheme is designed to adjust the combination factor, which solves the stage division problem of two‐stage decoupling learning. Experimental results on two hyperspectral image datasets demonstrate that, as compared with other competing approaches, the proposed method can continuously and stably optimize the model parameters, achieving improvements in class average metrics and difficult classes without affecting overall classification performance. [ABSTRACT FROM AUTHOR]

Synthetic images facilitate stereo matching. However, synthetic images may suffer from image distortion, domain bias, and stereo mismatch, which would significantly restrict the widespread use of stereo matching models in the real world. The first goal in this paper is to synthesize real‐looking images for minimizing the domain bias between the synthesized and real images. For this purpose, sharpened disparity maps are produced from a mono real image. Then, stereo image pairs are synthesized using these imperfect disparity maps and the single real image in the proposed pipeline. Although the synthesized images are as realistic as possible, the domain styles of the synthesized images are always very different from the real images. Thus, the second goal is to enhance the domain generalization ability of the stereo matching network. For that, the feature extraction layer is replaced with a teacher–student model. Then, a constraint of binocular contrast features is imposed on the output of the model. When tested on the KITTI, ETH3D, and Middlebury datasets, the accuracy of the method outperforms traditional methods by at least 30%. Experiments demonstrate that the approaches are general and can be conveniently embedded into existing stereo networks. [ABSTRACT FROM AUTHOR]

This paper presents a kernel extreme learning machine (KELM) integrated with the improved whale optimization algorithm (IWOA) to address the power quality disturbance (PQD) issue in microgrids. First, an adaptive variational mode decomposition method is employed to extract PQD signals in microgrids. Then, the IWOA is utilized to optimize the penalty factor and kernel function parameters for the KELM classifier model, thereby enhancing the performance of the classifier. Furthermore, the test results indicate that the proposed IWOA–KELM achieves high classification accuracy and rapid convergence for complex PQD signals.This paper presents a kernel extreme learning machine integrated with the improved whale optimization algorithm to address power quality issues in microgrids resulting from distributed power access. In this work, the adaptive variational mode decomposition method is employed to decompose the complex disturbance signals in microgrids. [ABSTRACT FROM AUTHOR]

A comparative study of two force perception skill learning approaches for robot‐assisted spinal surgery, the impedance model method and the imitation learning (IL) method, is presented. The impedance model method develops separate models for the surgeon and patient, incorporating spring‐damper and bone‐grinding models. Expert surgeons' feature parameters are collected and mapped using support vector regression and image navigation techniques. The imitation learning approach utilises long short‐term memory networks (LSTM) and addresses accurate data labelling challenges with custom models. Experimental results demonstrate skill recognition rates of 63.61%–74.62% for the impedance model approach, relying on manual feature extraction. Conversely, the imitation learning approach achieves a force perception recognition rate of 91.06%, outperforming the impedance model on curved bone surfaces. The findings demonstrate the potential of imitation learning to enhance skill acquisition in robot‐assisted spinal surgery by eliminating the laborious process of manual feature extraction. [ABSTRACT FROM AUTHOR]

Offline reinforcement learning (RL) aims to learn policies entirely from passively collected datasets, making it a data‐driven decision method. One of the main challenges in offline RL is the distribution shift problem, which causes the algorithm to visit out‐of‐distribution (OOD) samples. The distribution shift can be mitigated by constraining the divergence between the target policy and the behaviour policy. However, this method can overly constrain the target policy and impair the algorithm's performance, as it does not directly distinguish between in‐distribution and OOD samples. In addition, it is difficult to learn and represent multi‐modal behaviour policy when the datasets are collected by several different behaviour policies. To overcome these drawbacks, the authors address the distribution shift problem by implicit policy constraints with energy‐based models (EBMs) rather than explicitly modelling the behaviour policy. The EBM is powerful for representing complex multi‐modal distributions as well as the ability to distinguish in‐distribution samples and OODs. Experimental results show that their method significantly outperforms the explicit policy constraint method and other baselines. In addition, the learnt energy model can be used to indicate OOD visits and alert the possible failure. [ABSTRACT FROM AUTHOR]

Retinopathy of prematurity (ROP) mainly occurs in premature infants with low birth weight, and it is the leading cause of childhood blindness. Early and accurate ROP diagnosis is imperative for appropriate treatment. However, less research concentrates on early‐stage ROP diagnosis based on limited‐labelled images in an imbalanced dataset. To address the dilemma, this study proposed a novel self‐supervised network, MOCO‐MIM, for early ROP grading. The proposed classification network was evaluated on a total of 553 labelled fundus images from 89 preterm infants. The trained network achieved a test accuracy of 98.29% and an AUC score of 97.6% for three stages of grading. The adopted method is verified that the proposed method can be detected early stages of ROP more efficiently and grade the severity more accurately based on limited‐labelled fundus images, which is superior to the existing state‐of‐the‐art methods. [ABSTRACT FROM AUTHOR]

In order to solve the related problems of detection and tracking of shallow marine organisms, this paper designs a YOLO v5 multi‐target detection and tracking algorithm with attention mechanism. When working underwater, the authors usually encounter many difficulties. Different luminosity and complex coral background usually affect the detection of marine organisms. At the same time, the unrestricted movement of marine organisms, the ability to hide behind rocks and algae, and their mutual occlusion while swimming pose additional challenges to this task. Considering the characteristics of shallow marine organisms activity environment, the attention mechanism is added to the feature extraction network of YOLO v5 to reduce redundant information and improve the detection accuracy of shallow marine organism targets in complex environment. The improved algorithm improves the average detection accuracy of marine organisms target detection by 3.2%. Aiming at the problem of shallow marine organisms target tracking, a shallow marine organisms multi‐target tracking algorithm based on improved Deep Simple Online And Realtime Tracking (SORT) is designed. The improved YOLO v5 algorithm is used to replace Faster R‐CNN (Faster Region‐Convolutional Neural Networks) as the detector of DeepSORT tracking algorithm, and the cascade matching strategy is adopted to solve the problem that the target cannot be tracked continuously when it is occluded for a long time. The experimental results show that the shallow marine organisms multi‐target tracking algorithm based on improved DeepSORT reduces the number of id transformation of marine biological target tracking in shallow sea environment, and improves the accuracy of shallow marine organisms multi‐target tracking. [ABSTRACT FROM AUTHOR]

Accurate detection of underwater objects is a key indicator technology to effectively enhance the field of marine development and application, and is of great importance to various fields including marine military defense and seafood aquaculture. Efficient and rapid detection of underwater targets is a crucial technological challenge in this field. To meet the challenges posed by these issues, this study applies the convolutional omni‐efficient layer aggregation network (CO‐ELAN) module to the detector backbone to improve the ability of the network structure to acquire underwater objects from image information. The module improves the feature representation of gradient branching through a multi‐dimensional dynamic convolution and attention mechanism. In terms of loss calculation, the optimized normalized Wasserstein distance approach is used to predict the box distribution probabilistic modelling method to determine comparable distances to the ground box and obtain better samples of small target labels. Here, an underwater image enhancement algorithm based on white balance and underwater blur fusion is used to obtain clear images that enable improved detector performance. After the verification experiment on the URPC2018 dataset, it is found that the detector has better underwater detection ability compared with other detectors in the complex underwater environment. The proposed method achieves a 2.4% improvement over the YOLOv7 baseline model, while reducing computation costs by 5%. [ABSTRACT FROM AUTHOR]

Lesion segmentation is a fundamental task in medical image processing, often facing the challenge of subtle lesions. It is important to detect these lesions, even though they can be difficult to identify. Convolutional neural networks, an effective method in medical image processing, often ignore the relationship between lesions, leading to topological errors during training. To tackle topological errors, move is made from pixel‐level to hypergraph representations. Hypergraphs can model lesions as vertices connected by hyperedges, capturing the topology between lesions. This paper introduces a novel dynamic hypergraph learning strategy called DHLS. DHLS allows for the dynamic construction of hypergraphs contingent upon input vertex variations. A hypergraph global‐aware segmentation network, termed HGSNet, is further proposed. HGSNet can capture the key high‐order structure information, which is able to enhance global topology expression. Additionally, a composite loss function is introduced. The function emphasizes the global aspect and the boundary of segmentation regions. The experimental setup compared HGSNet with other advanced models on medical image datasets from various organs. The results demonstrate that HGSNet outperforms other models and achieves state‐of‐the‐art performance on three public datasets. [ABSTRACT FROM AUTHOR]

Modern radar systems often face various interference signals in complex and rapidly changing electronic environments. The task of suppressing this interference in the radar echo signal to extract vital information is challenging. A radar interference suppression method is proposed based on a generative adversarial network (GAN). This method effectively recovers the target signal from the echo signal, which contains interference and noise, by leveraging the powerful fitting ability of GAN. Specifically, this method was tested using coherent suppression interference, smart noise interference, and noise frequency modulation suppression interference. We compared the proposed GAN method with recurrent neural network, short‐time Fourier transform time‐varying filtering, short‐time fractional Fourier transform time‐varying filtering algorithms and RNN approach. The results show that the interference suppression algorithm based on GAN is superior to the other three algorithms. [ABSTRACT FROM AUTHOR]

The authors present spatio‐temporally resolved wideband measurements of Sub‐Terahertz (Sub‐THz) reflection coefficients in the frequency range of 92–110 GHz. A stochastic model for single reflection fixed links that is capable of modelling random scattering from small‐scale discontinuities such as those encountered in complex structures in walls and partitions of buildings is presented. The model auto‐regressively produces filter coefficients that are fed into an Infinite‐Impulse‐Response (IIR) filter which convolves them with the spatio‐temporal series in order to generate the next output sample based on previous observations. The IIR filter allows for flexible stochastic generation of samples, and its parameters can be adjusted as needed to suit different channel conditions. A total of 20 suitable start‐up filter coefficients are generated from 21.7 % of the sample size for each complex delay tap distribution, which corresponds to 50 complex instances of the channel. These coefficients are then utilised to validate the remaining measured sample set. The model is in quantitative agreement with measurement statistics and can be used to construct relatively simple modified reflection coefficients that can be used in micro‐cellular ray‐optical network planning tools. [ABSTRACT FROM AUTHOR]

The ability to predict the trajectory of an autonomous vehicle accurately is crucial for safe and efficient navigation. However, predicting diverse and multimodal futures can be challenging. Recent approaches such as attention and graph neural networks have achieved state‐of‐the‐art performance by considering agent interactions and map contexts. This study focused on multi‐agent prediction using an agent‐centric approach with transformers. This enables parallel computation and a comprehensive understanding of the environment. Two main features are introduced: an adaptive receptive field (ARF) that captures the relevant surroundings for each agent, and perception encoding, which serves as spatial context embeddings. The ARF adapts to the agent's velocity and rotation, focusing attention ahead at high speeds or to the sides when it is slower. Perception encoding divides agents or lanes into levels and encodes the information of each level. This approach enables the efficient encoding of complex spatial relationships. The proposed method combines these advances with transformer modelling for multi‐agent trajectory prediction while ensuring real‐time prediction capabilities. The approach is evaluated on the Argoverse benchmark and better performance than the state‐of‐the‐art baseline is achieved. By addressing challenges such as multimodal outputs and robustness, the study enhances the safety and efficiency of autonomous driving systems by more accurately predicting trajectories. [ABSTRACT FROM AUTHOR]

Short‐term load forecasting is critical for power system planning and operations, and ensemble forecasting methods for electricity loads have been shown to be effective in obtaining accurate forecasts. However, the weights in ensemble prediction models are usually preset based on the overall performance after training, which prevents the model from adapting in the face of different scenarios, limiting the improvement of prediction performance. In order to improve the accurateness and validity of the ensemble prediction method further, this paper proposes an ensemble deep reinforcement learning approach using Q‐learning dynamic weight assignment to consider local behaviours caused by changes in the external environment. Firstly, the variational mode decomposition is used to reduce the non‐stationarity of the original data by decomposing the load sequence. Then, the recurrent neural network (RNN), long short‐term memory (LSTM), and gated recurrent unit (GRU) are selected as the basic power load predictors. Finally, the optimal weights are ensembled for the three sub‐predictors by the optimal weights generated using the Q‐learning algorithm, and the final results are obtained by combining their respective predictions. The results show that the forecasting capability of the proposed method outperforms all sub‐models and several baseline ensemble forecasting methods. [ABSTRACT FROM AUTHOR]

This research addresses the challenge faced by most existing prediction methods in handling nonlinear data of cables. Furthermore, it proposes a novel power cable monitoring method utilizing UHF‐RFID and deep learning within an edge computing environment, specifically targeting the currently suboptimal wireless monitoring of cables. First, based on edge computing, a power cable monitoring system is designed to migrate the analysis of massive data to the edge of the network to improve the monitoring efficiency. Then, the temperature sensing chip and RFID chip were integrated to design a UHF‐RFID temperature tag, which was fixed at the cable temperature measurement point to achieve passive wireless monitoring of the cable. Finally, the parameters of the GRNN model are optimized using the beetle antennae search algorithm, and the EEMD decomposed data is input into the BAS‐GRNN model for learning to output temperature prediction results. Based on the establishment of an experimental platform, the method was demonstrated, and results showed that the maximum error between the UHF‐RFID temperature tag temperature measurement results and the thermocouple was within 0.3°C, and the average relative error of the proposed method was only 0.01, which can meet the accuracy requirements of actual monitoring of power cables. [ABSTRACT FROM AUTHOR]

Abstract Recent advancements in deep learning have significantly improved performance in the multi‐view stereo (MVS) domain, yet achieving a balance between reconstruction efficiency and quality remains challenging for learning‐based MVS methods. To address this, we introduce MFE‐MVSNet, designed for more effective and precise depth estimation. Our model incorporates a pyramid feature extraction network, featuring efficient multi‐scale attention and multi‐scale feature enhancement modules. These components capture pixel‐level pairwise relationships and semantic features with long‐range contextual information, enhancing feature representation. Additionally, we propose a lightweight 3D UNet regularization network based on depthwise separable convolutions to reduce computational costs. This network employs bi‐directional skip connections, establishing a fluid relationship between encoders and decoders and enabling cyclic reuse of building blocks without adding learnable parameters. By integrating these methods, MFE‐MVSNet effectively balances reconstruction quality and efficiency. Extensive qualitative and quantitative experiments on the DTU dataset validate our model's competitiveness, demonstrating approximately 33% and 12% relative improvements in overall score compared to MVSNet and CasMVSNet, respectively. Compared to other MVS networks, our approach more effectively balances reconstruction quality with efficiency.

Abstract Trained deep reinforcement learning (DRL) based controllers can effectively control dynamic systems where classical controllers can be ineffective and difficult to tune. However, the lack of closed‐loop stability guarantees of systems controlled by trained DRL agents hinders their adoption in practical applications. This research study investigates the closed‐loop stability of dynamic systems controlled by trained DRL agents using Lyapunov analysis based on a linear‐quadratic polynomial approximation of the trained agent. In addition, this work develops an understanding of the system's stability margin to determine operational boundaries and critical thresholds of the system's physical parameters for effective operation. The proposed analysis is verified on a DRL‐controlled system for several simulated and experimental scenarios. The DRL agent is trained using a detailed dynamic model of a non‐linear system and then tested on the corresponding real‐world hardware platform without any fine‐tuning. Experiments are conducted on a wide range of system states and physical parameters and the results have confirmed the validity of the proposed stability analysis (https://youtu.be/QlpeD5sTlPU).

Abstract Offline reinforcement learning (RL) aims to learn policies entirely from passively collected datasets, making it a data‐driven decision method. One of the main challenges in offline RL is the distribution shift problem, which causes the algorithm to visit out‐of‐distribution (OOD) samples. The distribution shift can be mitigated by constraining the divergence between the target policy and the behaviour policy. However, this method can overly constrain the target policy and impair the algorithm's performance, as it does not directly distinguish between in‐distribution and OOD samples. In addition, it is difficult to learn and represent multi‐modal behaviour policy when the datasets are collected by several different behaviour policies. To overcome these drawbacks, the authors address the distribution shift problem by implicit policy constraints with energy‐based models (EBMs) rather than explicitly modelling the behaviour policy. The EBM is powerful for representing complex multi‐modal distributions as well as the ability to distinguish in‐distribution samples and OODs. Experimental results show that their method significantly outperforms the explicit policy constraint method and other baselines. In addition, the learnt energy model can be used to indicate OOD visits and alert the possible failure.

Abstract A comparative study of two force perception skill learning approaches for robot‐assisted spinal surgery, the impedance model method and the imitation learning (IL) method, is presented. The impedance model method develops separate models for the surgeon and patient, incorporating spring‐damper and bone‐grinding models. Expert surgeons' feature parameters are collected and mapped using support vector regression and image navigation techniques. The imitation learning approach utilises long short‐term memory networks (LSTM) and addresses accurate data labelling challenges with custom models. Experimental results demonstrate skill recognition rates of 63.61%–74.62% for the impedance model approach, relying on manual feature extraction. Conversely, the imitation learning approach achieves a force perception recognition rate of 91.06%, outperforming the impedance model on curved bone surfaces. The findings demonstrate the potential of imitation learning to enhance skill acquisition in robot‐assisted spinal surgery by eliminating the laborious process of manual feature extraction.

Abstract Due to differences in the quantity and size of observed targets, hyperspectral images are characterized by class imbalance. The standard deep learning classification model training scheme optimizes the overall classification error, which may lead to performance imbalance between classes in hyperspectral image classification frameworks. Therefore, a novel factor annealing decoupling compositional training method is proposed in this paper. Without requiring resampling or reweighting, it implicitly modulates the training process, so standard models can sufficiently learn the representation of the minority classes and further be trained as robust classifiers. Specifically, the label‐distribution‐aware margin loss is combined with the error‐rate‐based cross‐entropy loss via combination factor, which considers both imbalanced data representation learning and classifier overall performance. Then, a factor annealing optimization training scheme is designed to adjust the combination factor, which solves the stage division problem of two‐stage decoupling learning. Experimental results on two hyperspectral image datasets demonstrate that, as compared with other competing approaches, the proposed method can continuously and stably optimize the model parameters, achieving improvements in class average metrics and difficult classes without affecting overall classification performance.

Abstract Image hiding is a task that hides secret images into cover images. The purposes of image hiding are to ensure the secret images are invisible to the human and the secret images can be recovered. The current state‐of‐the‐art steganography methods run the risk of secret information leakage. A safe image hiding network (SIHNet) is presented to reduce the leakage of secret information. Based on some phenomena of image hiding methods which use invertible neural network, a reversible secret image processing (SIP) module is proposed to make the secret images suitable for hiding and make the stego images leak less secret information. Besides, a reversible lost information hiding (LIH) module is used to hide the lost information into the cover images, thus the method can recover the secret images better than the method that uses random noise to replace the lost information. Experimental results show that SIHNet outperforms other state‐of‐the‐art methods on the PSNR and SSIM values of the recovered secret images and the stego images. Besides, residual images of other state‐of‐the‐art methods all contain information about secret images while residual images of SIHNet leak almost no secret information. Thus the method can prevent the listener of transmission channel from obtaining the information of the secret image through the residual image, which means SIHNet performs better in security than other state‐of‐the‐art methods.

Abstract Hyperspectral imaging has demonstrated its potential to provide correlated spatial and spectral information of a sample by a non‐contact and non‐invasive technology. In the medical field, especially in histopathology, HSI has been applied for the classification and identification of diseased tissue and for the characterization of its morphological properties. In this work, we propose a hybrid scheme to classify non‐tumor and tumor histological brain samples by hyperspectral imaging. The proposed approach is based on the identification of characteristic components in a hyperspectral image by linear unmixing, as a features engineering step, and the subsequent classification by a deep learning approach. For this last step, an ensemble of deep neural networks is evaluated by a cross‐validation scheme on an augmented dataset and a transfer learning scheme. The proposed method can classify histological brain samples with an average accuracy of 88%, and reduced variability, computational cost, and inference times, which presents an advantage over methods in the state‐of‐the‐art. Hence, the work demonstrates the potential of hybrid classification methodologies to achieve robust and reliable results by combining linear unmixing for features extraction and deep learning for classification.

Abstract The authors present spatio‐temporally resolved wideband measurements of Sub‐Terahertz (Sub‐THz) reflection coefficients in the frequency range of 92–110 GHz. A stochastic model for single reflection fixed links that is capable of modelling random scattering from small‐scale discontinuities such as those encountered in complex structures in walls and partitions of buildings is presented. The model auto‐regressively produces filter coefficients that are fed into an Infinite‐Impulse‐Response (IIR) filter which convolves them with the spatio‐temporal series in order to generate the next output sample based on previous observations. The IIR filter allows for flexible stochastic generation of samples, and its parameters can be adjusted as needed to suit different channel conditions. A total of 20 suitable start‐up filter coefficients are generated from 21.7 % of the sample size for each complex delay tap distribution, which corresponds to 50 complex instances of the channel. These coefficients are then utilised to validate the remaining measured sample set. The model is in quantitative agreement with measurement statistics and can be used to construct relatively simple modified reflection coefficients that can be used in micro‐cellular ray‐optical network planning tools.

Abstract Modern radar systems often face various interference signals in complex and rapidly changing electronic environments. The task of suppressing this interference in the radar echo signal to extract vital information is challenging. A radar interference suppression method is proposed based on a generative adversarial network (GAN). This method effectively recovers the target signal from the echo signal, which contains interference and noise, by leveraging the powerful fitting ability of GAN. Specifically, this method was tested using coherent suppression interference, smart noise interference, and noise frequency modulation suppression interference. We compared the proposed GAN method with recurrent neural network, short‐time Fourier transform time‐varying filtering, short‐time fractional Fourier transform time‐varying filtering algorithms and RNN approach. The results show that the interference suppression algorithm based on GAN is superior to the other three algorithms.

Abstract Lesion segmentation is a fundamental task in medical image processing, often facing the challenge of subtle lesions. It is important to detect these lesions, even though they can be difficult to identify. Convolutional neural networks, an effective method in medical image processing, often ignore the relationship between lesions, leading to topological errors during training. To tackle topological errors, move is made from pixel‐level to hypergraph representations. Hypergraphs can model lesions as vertices connected by hyperedges, capturing the topology between lesions. This paper introduces a novel dynamic hypergraph learning strategy called DHLS. DHLS allows for the dynamic construction of hypergraphs contingent upon input vertex variations. A hypergraph global‐aware segmentation network, termed HGSNet, is further proposed. HGSNet can capture the key high‐order structure information, which is able to enhance global topology expression. Additionally, a composite loss function is introduced. The function emphasizes the global aspect and the boundary of segmentation regions. The experimental setup compared HGSNet with other advanced models on medical image datasets from various organs. The results demonstrate that HGSNet outperforms other models and achieves state‐of‐the‐art performance on three public datasets.

Abstract In order to solve the related problems of detection and tracking of shallow marine organisms, this paper designs a YOLO v5 multi‐target detection and tracking algorithm with attention mechanism. When working underwater, the authors usually encounter many difficulties. Different luminosity and complex coral background usually affect the detection of marine organisms. At the same time, the unrestricted movement of marine organisms, the ability to hide behind rocks and algae, and their mutual occlusion while swimming pose additional challenges to this task. Considering the characteristics of shallow marine organisms activity environment, the attention mechanism is added to the feature extraction network of YOLO v5 to reduce redundant information and improve the detection accuracy of shallow marine organism targets in complex environment. The improved algorithm improves the average detection accuracy of marine organisms target detection by 3.2%. Aiming at the problem of shallow marine organisms target tracking, a shallow marine organisms multi‐target tracking algorithm based on improved Deep Simple Online And Realtime Tracking (SORT) is designed. The improved YOLO v5 algorithm is used to replace Faster R‐CNN (Faster Region‐Convolutional Neural Networks) as the detector of DeepSORT tracking algorithm, and the cascade matching strategy is adopted to solve the problem that the target cannot be tracked continuously when it is occluded for a long time. The experimental results show that the shallow marine organisms multi‐target tracking algorithm based on improved DeepSORT reduces the number of id transformation of marine biological target tracking in shallow sea environment, and improves the accuracy of shallow marine organisms multi‐target tracking.

Abstract Accurate detection of underwater objects is a key indicator technology to effectively enhance the field of marine development and application, and is of great importance to various fields including marine military defense and seafood aquaculture. Efficient and rapid detection of underwater targets is a crucial technological challenge in this field. To meet the challenges posed by these issues, this study applies the convolutional omni‐efficient layer aggregation network (CO‐ELAN) module to the detector backbone to improve the ability of the network structure to acquire underwater objects from image information. The module improves the feature representation of gradient branching through a multi‐dimensional dynamic convolution and attention mechanism. In terms of loss calculation, the optimized normalized Wasserstein distance approach is used to predict the box distribution probabilistic modelling method to determine comparable distances to the ground box and obtain better samples of small target labels. Here, an underwater image enhancement algorithm based on white balance and underwater blur fusion is used to obtain clear images that enable improved detector performance. After the verification experiment on the URPC2018 dataset, it is found that the detector has better underwater detection ability compared with other detectors in the complex underwater environment. The proposed method achieves a 2.4% improvement over the YOLOv7 baseline model, while reducing computation costs by 5%.

Abstract The ability to predict the trajectory of an autonomous vehicle accurately is crucial for safe and efficient navigation. However, predicting diverse and multimodal futures can be challenging. Recent approaches such as attention and graph neural networks have achieved state‐of‐the‐art performance by considering agent interactions and map contexts. This study focused on multi‐agent prediction using an agent‐centric approach with transformers. This enables parallel computation and a comprehensive understanding of the environment. Two main features are introduced: an adaptive receptive field (ARF) that captures the relevant surroundings for each agent, and perception encoding, which serves as spatial context embeddings. The ARF adapts to the agent's velocity and rotation, focusing attention ahead at high speeds or to the sides when it is slower. Perception encoding divides agents or lanes into levels and encodes the information of each level. This approach enables the efficient encoding of complex spatial relationships. The proposed method combines these advances with transformer modelling for multi‐agent trajectory prediction while ensuring real‐time prediction capabilities. The approach is evaluated on the Argoverse benchmark and better performance than the state‐of‐the‐art baseline is achieved. By addressing challenges such as multimodal outputs and robustness, the study enhances the safety and efficiency of autonomous driving systems by more accurately predicting trajectories.

Abstract A priority for machine learning in healthcare and other high stakes applications is to enable end‐users to easily interpret individual predictions. This opinion piece outlines recent developments in interpretable classifiers and methods to open black box models.

Abstract A novel method based on machine learning is proposed to estimate the electromagnetic radiation level at the ground plane near fifth‐generation (5G) base stations. The machine learning model was trained using data from various 5G base stations, enabling it to estimate the electric field intensity at any arbitrary radiation point when the base station provides service to different numbers of 5G terminals which are in different service modes. The inputs required for the model include the transmit power of the antenna, the antenna gain, the distance between the 5G base station and 5G terminals, terminal service modes, the number of 5G terminals and the environmental complexity around the 5G base station. Experimental results demonstrate the feasibility and effectiveness of the estimation method, with the mean absolute percentage error of the machine learning model being approximately 5.98%. This level of accuracy showcases the reliability of the approach. Moreover, the proposed method offers low costs when compared with on‐site measurements. The estimated results can be utilised to reduce test costs and provide valuable guidance for optimal site selection, thereby facilitating radio wave coverage or electromagnetic radiation regulation of 5G base stations.

Abstract Currently, the digital transformation of the power grid is underway, and the intelligent health management technology for power transformers is rapidly advancing. However, there are issues in the operation and maintenance process, such as weak information correlation and low decision‐making efficiency. Knowledge graphs have been applied in other industrial fields, such as spacecraft maintenance, to significantly improve knowledge query efficiency. However, there is a lack of literature on knowledge graph construction in the field of power transformer operation and maintenance. Additionally, there is limited publicly available data and difficulties in effectively mining operation and maintenance knowledge in this field. A method for constructing a knowledge graph for power transformer operation and maintenance based on ALBERT is proposed. Firstly, publicly available literature in the field of power transformers is collected, and a sample enhancement method using regular matching is used to enrich the semi‐structured corpora, such as power system accident investigation reports, to construct a training dataset for power transformer operation and maintenance. Then, the ALBERT‐BiLSTM‐CRF deep learning algorithm is applied to extract power transformer operation and maintenance entities from the relevant literature and accident investigation reports, and this method is compared with traditional deep learning algorithms to demonstrate its superiority. Subsequently, the ALBERT‐BiLSTM‐Attention deep learning algorithm, which incorporates ALBERT and attention mechanism, is utilised to extract relationships between power transformer operation and maintenance entities. Compared to other deep learning algorithms, this algorithm demonstrates better performance in the domain‐specific texts of power transformer operation and maintenance. Finally, the Neo4j graph database is used to visualise and present the knowledge graph, enabling decision support based on the power transformer operation and maintenance knowledge graph.

Abstract Various types of foreign fibres may be mixed in the planting, transportation, and production processes of cotton, which not only cause equipment to be out of control, but also leads to a decrease in the quality of cotton textile products and economic losses. The machine vision based detection method for cotton foreign fibres is widely used. Based on existing related research, we construct a classification dataset for cotton foreign fibres in practical scenarios, named the CF2113‐10 dataset. The authors design a basic foreign fibre classification network called CottonNet that balances performance and efficiency. The classification accuracy on the validation set reached 94.2%. In order to enhance the high‐level feature extraction ability, this paper improves the feature fusion method of residual networks and proposes CottonNet‐Res, which improves the classification accuracy to 95.1%. Finally, a classification model based on feature difference fitting, CottonNet‐Fusion, is proposed to address the classification problem of foreign fibre images sampled in complex environments. The classification accuracy of foreign fibre images sampled in ordinary scenes has improved to 97.4%, while the images sampled in complex environments maintain an accuracy of 90.3%.

Abstract The registration performance determines the widespread indoor application of 3D models acquired by depth sensors. Many advanced registration methods lack comprehensive feature aggregation and poor generalization capabilities, which improves the mismatching ratio. Here, a dual graph network is proposed by incorporating irregular shape factors to make point cloud features more expressive. At first, we transform point cloud sets into the stellar graph within the local neighbourhood of each point. The deep feature and shape factor of each point are combined in the directional‐connected irregular projection space. Subsequently, the combined features are modelled as the second graph. By the attention mechanism computation, feature information is continuously aggregated with intra‐graph and inter‐graph. Finally, a loss function is utilized to confirm point correspondence and perform the registration through singular value decomposition. Extensive experiments validate that the proposed point cloud registration method achieves state‐of‐the‐art performance.

Abstract The authors design a novel convolutional network architecture, that is, deep network with double reuses and convolutional shortcuts, in which new compressed reuse units are presented. Compressed reuse units combine the reused features from the first 3 × 3 convolutional layer and the features from the last 3 × 3 convolutional layer to produce new feature maps in the current compressed reuse unit, simultaneously reuse the feature maps from all previous compressed reuse units to generate a shortcut by an 1 × 1 convolution, and then concatenate these new maps and this shortcut as the input to next compressed reuse unit. Deep network with double reuses and convolutional shortcuts uses the feature reuse concatenation from all compressed reuse units as the final features for classification. In deep network with double reuses and convolutional shortcuts, the inner‐ and outer‐unit feature reuses and the convolutional shortcut compressed from the previous outer‐unit feature reuses can alleviate the vanishing‐gradient problem by strengthening the forward feature propagation inside and outside the units, improve the effectiveness of features and reduce calculation cost. Experimental results on CIFAR‐10, CIFAR‐100, ImageNet ILSVRC 2012, Pascal VOC2007 and MS COCO benchmark databases demonstrate the effectiveness of authors’ architecture for object recognition and detection, as compared with the state‐of‐the‐art.

Abstract Video frame interpolation (VFI) is a technique that synthesises intermediate frames between adjacent original video frames to enhance the temporal super‐resolution of the video. However, existing methods usually rely on heavy model architectures with a large number of parameters. The authors introduce an efficient VFI network based on multiple lightweight convolutional units and a Local three‐scale encoding (LTSE) structure. In particular, the authors introduce a LTSE structure with two‐level attention cascades. This design is tailored to enhance the efficient capture of details and contextual information across diverse scales in images. Secondly, the authors introduce recurrent convolutional layers (RCL) and residual operations, designing the recurrent residual convolutional unit to optimise the LTSE structure. Additionally, a lightweight convolutional unit named separable recurrent residual convolutional unit is introduced to reduce the model parameters. Finally, the authors obtain the three‐scale decoding features from the decoder and warp them for a set of three‐scale pre‐warped maps. The authors fuse them into the synthesis network to generate high‐quality interpolated frames. The experimental results indicate that the proposed approach achieves superior performance with fewer model parameters.

Abstract The medical information system is rich in datasets, but no intelligent systems can easily analyse the disease. Recently, ML (Machine Learning)‐based algorithms have acted as a handy diagnostic tool to identify whether a person is affected by thyroid or not. However, they produced classification with low accuracy and led to misclassification. Hence, the proposed system combines quantum computing with ML techniques to enhance computational power and precision. The system employs modified QPSO (Quantum Particle Swarm Optimisation) for feature selection since its searching performance is better than that of conventional PSO for selecting the optimum global position of the particle, thus selecting the relevant feature. Whereas, the QSVM (Quantum Support Vector Machine) is implemented for more accurate classification than classical SVM, as it tends to capture complex patterns in data produced due to high dimensional feature space applied by quantum kernel functions. This combination of modified QPSO and QSVM tends to increase the performance accuracy significantly. The efficiency of the proposed model is measured based on derivative parameters, such as F‐1‐score, recall, precision and accuracy, with corresponding confusion matrix and ROC. Further, the classification is compared with other traditional approaches to predict the accuracy of the proposed model with traditional methods.

Abstract The article describes an innovative methodology for the design and experimental validation of monitoring and anomaly detection algorithms, with a particular focus on the aging phenomenon, linked to the anomalous modification of the Rds(on), in devices switching in power electronic systems integrated into modern high‐performance electrified vehicles. The case study concerns an electric drive for fully electrified vehicles, in which a three‐phase axial flux synchronous motor integrated into a wheel motor (Elaphe) is used and in which a high‐efficiency three‐phase inverter, designed with SiC technology (silicon carbide). The article proposes the design and validation of the innovative aging monitoring and detection system, in four consecutive phases. The first phase involves the creation of a real‐time model of electric drive, validated through experimental data extrapolated directly during a WLTP (Worldwide Harmonized Light Vehicle Test Procedure) test. The second phase consists of the creation of a virtual dataset representative of the aging phenomenon, via an anomaly injection procedure, emulating this phenomenon with a scaling factor (depending on the value of the Rds(on)) on the current phase of the motor, relating to the inverter branch whose SiC device is affected. The third phase concerns the design of an estimator of the Rds(on), based on an ANN (Artificial Neural Network) regression model, and involves a data manipulation phase with features extraction and reduction techniques. The fourth and final phase, involves the experimental validation of the method, through PIL (Processor‐In‐the‐Loop) tests, integrating the monitoring algorithm (consisting of a real‐time model and AI‐based regression model) on the NXPs32k144 embedded platform (based on Cortex‐M4), making the algorithm interact with the electric drive model on which anomaly injection is applied.

Abstract This paper presents a novel algorithm for fault diagnosis and fault tolerant control of interleaved boost converters (IBCs) in the presence of single or double simultaneous open‐circuit faults (OCFs) in power switches. An innovative diagnosis signal (recorded by a cheap current sensor) will be introduced whose waveform mainly depends on the healthy or faulty condition of the converter. Since the diagnosis signal is periodic in the steady‐state operation of the converters, its Fourier Series coefficients, together with the duty cycle of the converter are used as distinguishing features for fault diagnosis of the OCFs in the converter. The well‐known Kalman filter is utilized for robust estimation of the Fourier‐based features. Finally, the modulation of the remaining healthy phases is rearranged in a way that the pre‐fault performance recovers. The proposed algorithm is verified in a laboratory four‐phase IBC setup in which the experimental results show its satisfactory performance. Also, the structure of the proposed method is straightforward and can be implemented in the same microcontroller which is used for voltage and current regulation of the converter.

Abstract Feature matching plays a crucial role in computer vision, with applications in visual localization, simultaneous localization and mapping (SLAM), image stitching, and more. It establishes correspondences between sets of feature points from multiple images, enabling various tasks. Over the years, feature matching has witnessed significant development, with an increasing number of methods being applied. However, different methods exhibit different degrees of applicability in different scenarios and requirements due to their different rationales. To cope with these issues, a comprehensive analysis and comparison of matching methods are essential. Existing reviews often lack coverage of deep learning models and focus more on feature detection and description, neglecting the matching process. This survey investigates feature detection, description, and matching techniques within the feature‐based image‐matching pipeline. Representative methods, their mechanisms, and application scenarios are also briefly introduced. In addition, comprehensive evaluations of classical and state‐of‐the‐art methods are conducted through extensive experiments on representative datasets. Particularly, matching‐based applications are compared to fully demonstrate the advantages of the methods. Lastly, this survey highlights current problems and development directions in matching methods, serving as a reference for researchers in the field.

Abstract Most existing multi‐view stereo (MVS) methods fail to consider global context information in the stage of feature extraction and cost aggregation. As transformers have shown remarkable performance on various vision tasks due to their ability to perceive global contextual information, this paper proposes a transformer‐based feature enhancement network (TF‐MVSNet) to facilitate feature representation learning by combining local features (both 2D and 3D) with long‐range contextual information. To reduce memory consumption of feature matching, the cross‐attention mechanism is leveraged to efficiently construct 3D cost volumes under the epipolar constraint. Additionally, a colour‐guided network is designed to refine depth maps at a coarse stage, hence reducing incorrect depth predictions at a fine stage. Extensive experiments were performed on the DTU dataset and Tanks and Temples (T&T) benchmark and results are reported.

Abstract Generalizable and accurate stereo depth estimation is vital for 3D reconstruction, especially in surgery. Supervised learning methods obtain best performance however, limited ground truth data for surgical scenes limits generalizability. Self‐supervised methods don't need ground truth, but suffer from scale ambiguity and incorrect disparity prediction due to inconsistency of photometric loss. This work proposes a two‐phase training procedure that is generalizable and retains the high performance of supervised methods. It entails: (1) performing self‐supervised representation learning of left and right views via masked image modelling (MIM) to learn generalizable semantic stereo features (2) utilizing the MIM pre‐trained model to learn robust depth representation via supervised learning for disparity estimation on synthetic data only. To improve stereo representations learnt via MIM, perceptual loss terms are introduced, which improve the model's stereo representations learnt by explicitly encouraging the learning of higher scene‐level features. Qualitative and quantitative performance evaluation on surgical and natural scenes shows that the approach achieves sub‐millimetre accuracy and lowest errors respectively, setting a new state‐of‐the‐art. Despite not training on surgical nor natural scene data for disparity estimation.

Abstract The integration of Augmented Reality (AR) into daily surgical practice is withheld by the correct registration of pre‐operative data. This includes intelligent 3D model superposition whilst simultaneously handling real and virtual occlusions caused by the AR overlay. Occlusions can negatively impact surgical safety and as such deteriorate rather than improve surgical care. Robotic surgery is particularly suited to tackle these integration challenges in a stepwise approach as the robotic console allows for different inputs to be displayed in parallel to the surgeon. Nevertheless, real‐time de‐occlusion requires extensive computational resources which further complicates clinical integration. This work tackles the problem of instrument occlusion and presents, to the authors’ best knowledge, the first‐in‐human on edge deployment of a real‐time binary segmentation pipeline during three robot‐assisted surgeries: partial nephrectomy, migrated endovascular stent removal, and liver metastasectomy. To this end, a state‐of‐the‐art real‐time segmentation and 3D model pipeline was implemented and presented to the surgeon during live surgery. The pipeline allows real‐time binary segmentation of 37 non‐organic surgical items, which are never occluded during AR. The application features real‐time manual 3D model manipulation for correct soft tissue alignment. The proposed pipeline can contribute towards surgical safety, ergonomics, and acceptance of AR in minimally invasive surgery.

Abstract With the popularity and advancement of 3D point cloud data acquisition technologies and sensors, research into 3D point clouds has made considerable strides based on deep learning. The semantic segmentation of point clouds, a crucial step in comprehending 3D scenes, has drawn much attention. The accuracy and effectiveness of fully supervised semantic segmentation tasks have greatly improved with the increase in the number of accessible datasets. However, these achievements rely on time‐consuming and expensive full labelling. In solve of these existential issues, research on weakly supervised learning has recently exploded. These methods train neural networks to tackle 3D semantic segmentation tasks with fewer point labels. In addition to providing a thorough overview of the history and current state of the art in weakly supervised semantic segmentation of 3D point clouds, a detailed description of the most widely used data acquisition sensors, a list of publicly accessible benchmark datasets, and a look ahead to potential future development directions is provided.

Abstract The AIoT, with its artificial intelligence capabilities, can further enhance Device‐to‐Device (D2D) communication. Based on Non‐Orthogonal Multiple Access (NOMA), D2D technology can effectively alleviate wireless spectrum resource pressure and improve the capacity of heterogeneous cellular networks. However, it also introduces significant system interference issues. In this paper, a resource allocation algorithm is proposed for the NOMA‐D2D heterogeneous cellular network, based on a multi‐agent deep reinforcement learning framework. Firstly, the algorithm allocates appropriate channels to D2D clusters. Then, the power allocation factors and D2D transmit power are jointly optimized to suppress the interference and improve the system performance. Simulation results show that both the channel allocation efficiency and the power control performance of the system can be significantly improved.

Abstract The vehicle platoon lane changing is significant for alleviating road congestion and diminishing transportation energy consumption. However, the lane changing strategy for a group of vehicles is still a great challenge in this field. This paper investigates the vehicle platoon lane changing problems, in which the safety and efficiency in the lane changing procedure are both taken into consideration. Since the safety of the platoon lane changing would be affected by the lane changing gap and the length of the platoon, a novel platoon lane changing strategy is proposed by using the deep Q‐network. In detail, the proposed platoon lane changing strategy contains two layers, where the first one is a decision layer and the other one is the verification layer. In the decision layer, the deep Q‐network is employed to improve the lane changing efficiency. Then, the verification layer is presented to enhance the platoon lane changing safety. In final, some typical platoon lane changing scenarios are provided in an existing ramp containing a vehicle platoon and some random vehicles. The related numerical simulations are conducted to validate the feasibility and effectiveness of the proposed approaches.

Abstract Numerous efforts have been made to address the section‐level travel speed prediction problem. However, section‐level predictions can hardly be used for fine‐grained applications, such as lane management and lane‐level navigation. The main reason for this is that significant speed heterogeneity exists among the lanes within one section. Thus, this study proposes a three‐dimensional (3D) dual attention convolution‐based deep learning model for predicting the lane‐level travel speed. 3D convolutions are designed to learn high‐dimensional spatiotemporal traffic flow features, that is, the relationships between different sections, lanes, and periods. Dual attention modules are used to focus on the traffic flow propagation patterns and to explain the model's mechanisms. To evaluate the proposed model, an indicator is introduced to assess the spatio‐temporal learning ability, based on targeting the lane‐level case. Evaluation experiments are conducted based on loop detector data in Shanghai, China. The results show that high accuracy is obtained by the proposed model, with a 2.9 km/h mean absolute error, thereby outperforming several existing methods. Finally, an in‐depth analysis is provided regarding the attention coefficients and interpretation of real‐world lane‐level traffic flow propagation patterns, so as to gain insights into the model's mechanism when capturing dynamic lane‐level traffic flow.

Currently, the digital transformation of the power grid is underway, and the intelligent health management technology for power transformers is rapidly advancing. However, there are issues in the operation and maintenance process, such as weak information correlation and low decision‐making efficiency. Knowledge graphs have been applied in other industrial fields, such as spacecraft maintenance, to significantly improve knowledge query efficiency. However, there is a lack of literature on knowledge graph construction in the field of power transformer operation and maintenance. Additionally, there is limited publicly available data and difficulties in effectively mining operation and maintenance knowledge in this field. A method for constructing a knowledge graph for power transformer operation and maintenance based on ALBERT is proposed. Firstly, publicly available literature in the field of power transformers is collected, and a sample enhancement method using regular matching is used to enrich the semi‐structured corpora, such as power system accident investigation reports, to construct a training dataset for power transformer operation and maintenance. Then, the ALBERT‐BiLSTM‐CRF deep learning algorithm is applied to extract power transformer operation and maintenance entities from the relevant literature and accident investigation reports, and this method is compared with traditional deep learning algorithms to demonstrate its superiority. Subsequently, the ALBERT‐BiLSTM‐Attention deep learning algorithm, which incorporates ALBERT and attention mechanism, is utilised to extract relationships between power transformer operation and maintenance entities. Compared to other deep learning algorithms, this algorithm demonstrates better performance in the domain‐specific texts of power transformer operation and maintenance. Finally, the Neo4j graph database is used to visualise and present the knowledge graph, enabling decision support based on the power transformer operation and maintenance knowledge graph. [ABSTRACT FROM AUTHOR]