Abstract:Person re-identification (Re-ID),which involves retrieving the same person across cameras,is a key technology in the field of intelligent video surveillance.However,due to the complexity of surveillance scenarios,traditional single-modal approaches encounter limitations in extreme conditions such as low lighting and foggy days.Given the practical demands and the swift advancement in deep learning,multi-modal person Re-ID based on deep learning has received widespread attention.This article provides a review of the progress in multi-modal person Re-ID based on deep learning in recent years,elaborates on the shortcomings of traditional single-modal approaches and summarizes the common application scenarios and advantages of multi-modal person Re-ID,as well as the composition of various datasets.The article also highlights the relevant methods and classification of multi-modal person Re-ID across diverse scenarios,exploring current research hotspots and challenges.Finally,it discusses the future development trends and potential applications of multi-modal person Re-ID.

Abstract:RGB-Infrared person re-identification (Re-ID) is a challenging task which aims to match person images between visible and infrared modalities,playing a crucial role in criminal investigation and intelligent video surveillance.To address the weak feature extraction capability for fine-grained features in current cross-modal person Re-ID tasks,this paper proposes a person re-identification model based on fused attention and feature enhancement.First,automatic data augmentation techniques are employed to mitigate the differences in perspectives and scales among different cameras,and a cross-attention multi-scale Vision Transformer is proposed to generate more discriminative feature representations by processing multi-scale features.Then the channel attention and spatial attention mechanisms are introduced to learn information important for distinguishing features when fusing visible and infrared image features.Finally,a loss function is designed,which adopts the adaptive weight based hard triplet loss,to enhance the correlation between each sample and improve the capability of identifying different persons from visible and infrared images.Extensive experiments conducted on the SYSU-MM01 and RegDB datasets show that the proposed approach achieves mAP of 68.05% and 85.19%,respectively,outperforming many state-of-the-art approaches.Moreover,ablation experiments and comparative analysis validate the superiority and effectiveness of the proposed model.

Abstract:Existing person re-identification (Re-ID) methods often struggle with inaccurate feature extraction and misidentification of person features due to environmental noise.Here,we propose a multi-feature fusion branch network for person Re-ID based on dynamic convolution and attention mechanism.First,considering the uncertainties in illumination,human posture and occlusion,dynamic convolution is proposed to replace static convolution in ResNet50 to obtain a more robust Dy-ResNet50 model.Second,given the great difference in camera perspective and the likelihood of people being occluded by objects,self-attention and cross-attention mechanisms are embedded into the backbone network.Finally,the cross entropy loss function and the hard triplet loss function are used as the model's loss functions,and experiments are carried out on public datasets of DukeMTMC-ReID,Market-1501 and MSMT17.The results show that the proposed model outperforms current mainstream models in Rank-1 (first hit) and mAP (mean Average Precision) on three public datasets,indicating its high identification accuracy.

Abstract:To solve the poor image quality and subsequent low efficiency of machine vision tasks on rainy days,an image rain removal algorithm based on multi-feature interaction and dense residual is proposed.First,a multi-feature interactive convolution module is proposed to extract the semantic features of rain streaks in different spaces to enhance information utilization.Second,a multidimensional space weight attention module is constructed,and the weights of different spatial information are preliminarily integrated to enhance the characteristics of rain streaks.Then combining the advantages of dense connection and residual network,a dense residual fusion module is designed,which improves the learning ability of the network,realizes the reuse of information,and further corrects the rain information.Finally,the output image quality is improved through the linear combination of various loss functions as well as the rainy day imaging model.Experiments on several public datasets show that the subjective and objective evaluation indexes of the proposed algorithm outperform those of the classical algorithm and novel algorithms,and the detailed background information of the images can be better preserved while removing the rain streaks.

Abstract:To address the unstable detection of marine targets challenged existing artificial intelligence algorithms due to the target's complex poses and variable scales,a detection approach based on deep supervision and improved YOLOv8 is proposed.A multi-scale convolution module is designed to extract the feature information of the target's multi-receptive fields and reduce the missed detection rate.Then,a deep supervision network is added to improve the utilization ratio of deep class information and shallow location information,thus optimizing the performance of the backbone network in target feature extraction.Finally,a channel attention mechanism is introduced into the detection head to filter the irrelevant information and enhance the recognition rate of key features.Experiments on the marine target dataset show that the mAP value and the recall rate of the proposed approach reach 93.69% and 85.16%,respectively,which are 7.38 and 8.52 percentage points higher than those of the original model,and the proposed approach outperforms both classical and novel algorithms.The detection time is about 14 ms,which meets the requirements of real-time marine target detection and provides technical support for shipping management and marine accident prevention.

Abstract:Laser SLAM (Simultaneous Localization and Mapping) and visual SLAM have been fully developed and widely used in military and civil fields.However,single sensor SLAM has limitations,for instance,laser SLAM is not suitable for scenes with a large number of dynamic objects around it,while visual SLAM has poor robustness in low-texture environments.Therefore,fusion of the two technologies has great potential to compensate each other,and it can be prospected that SLAM technology combining laser and vision or even more sensors will be the mainstream direction in the future.Here,we review the development of SLAM technology,analyze the hardware information of lidar and camera,and introduce some classical open-source algorithms and datasets.Furthermore,the multi-sensor fusion schemes are detailed from perspectives of uncertainty,feature and novel deep learning.The excellent performance of multi-sensor fusion schemes in complex scenes are summarized,and the future development trend of multi-sensor fusion is prospected.

Abstract:In this article,an improved algorithm that combines jump point search and bi-directional parallel ant colony search is proposed for static grid maps to address the slow convergence and easy trapping in local optima perplexed traditional ant colony algorithms for Automated Guided Vehicle(AGV) path planning.First,the actual research environment is modeled by gridization,and the improved jump point search algorithm is used to generate the initial suboptimal path for bi-directional search,providing a reference for the initial search direction of bi-directional ant colony search.Second,an improved transition probability heuristic function is used in the bi-directional parallel ant colony search process,which considers the avoidance of collision between AGV and obstacles when determining the next transition node.Meanwhile,by designing an information sharing mechanism and combining two fusion models of improved information increment and concentration,the global information concentration is shared and updated to better explore and optimize the path and ensure the connection of bi-directional paths.Finally,experimental results are compared with those of traditional ant colony algorithms to verify the improved algorithm's global search capability,efficiency and security.

Abstract:Federated learning is an important method to address two critical challenges in machine learning:data sharing and privacy protection.However,federated learning itself faces challenges related to data heterogeneity and model heterogeneity.Existing researches often focus on addressing one of these issues while overlook the correlation between them.To address this,this paper introduces a framework named PFKD (Personalized Federated learning based on Knowledge Distillation).This framework utilizes knowledge distillation techniques to address model heterogeneity and personalized algorithms to tackle data heterogeneity,thereby achieving more personalized federated learning.Experimental analysis validates the effectiveness of the proposed framework.The experimental results demonstrate that the framework can overcome model performance bottlenecks and improve model accuracy by approximately one percentage point.Furthermore,with appropriate hyperparameter adjustment,the framework's performance is further enhanced.

Abstract:Computational offloading is an essential technology in Mobile Edge Computing (MEC).To address the shortage of computational offloading strategies in multi-user and multi-MEC server scenarios,this paper proposes a hybrid artificial bee colony approach (Artificial Reverse Sine-Cosine,ARSC).First,the opposition-based learning strategy is used to initialize the population and optimize the initial solution of the population.Then the global optimal bootstrap information of the sine-cosine algorithm is exploited to improve the local search capability in the employed bee stage.Finally,to balance the global and local search capability of the approach,the step size factor is adapted by introducing dynamic perception.Simulation results show that the proposed ARSC approach outperforms offloading strategies based on particle swarm algorithm and artificial bee colony algorithm in convergence,latency,and energy consumption.

Abstract:In the islanded microgrid,the mismatch in line impedance hinders the traditional droop control from achieving an equal distribution of reactive power among Distributed Generation (DG) sources.To address this problem and enhance the flexibility and reliability of the controller,this paper analyzes the reasons behind the failure of traditional droop control and proposes an adaptive droop coefficient that can be dynamically adjusted to meet the conditions for reactive power sharing.Additionally,a dynamic distributed observer is designed and its convergence is proven,which allows DG sources to obtain necessary information flexibly and reliably in a distributed manner.Finally,the proposed control strategy is verified through simulations in four different scenarios,and the results demonstrate its superiority and effectiveness.

Abstract:To address the internal and external disturbances in Permanent Magnet Synchronous Motor (PMSM) such as modeling errors and sudden load variation,a load estimation-based composite Active Disturbance Rejection Control (ADRC) approach is proposed for PMSM speed regulation.ADRC is adopted to replace the PI controller in the speed loop to improve the performance of control system and solve the contradiction between system rapidity and overshooting.A load torque observer is designed to correct the slow response of ESO to sudden load variation by directly estimating and compensating the load torque in real time via speed and current signals.Additionally,a semi-physical experimental platform of composite ADRC for speed regulation is constructed in Matlab/Simulink environment,and the proposed composite ADRC is compared with traditional PI control and linear ADRC.The results illustrate that the proposed approach outperforms conventional controller by reducing the speed variation by more than 30% under abrupt load variation,and has superior disturbance rejection ability and speed regulation performance.

Abstract:The electric power output of photovoltaic array exhibits multi-peak characteristics under partial shading conditions,and changes with the external environment.To achieve efficient power output,the Multi-Verse Optimization (MVO) algorithm,which has outstanding advantages in solving low dimensional and small-scale optimization problems,is exploited to carry out Maximum Power Point Tracking (MPPT),and multiple strategies are integrated to address its defects.Then Latin hypercube sampling is used to initialize the universe population,and Cauchy mutation is carried out on the universe randomly swapped according to roulette strategy,thus increasing the diversity of the universe population.Meanwhile,the Levy flight Quantum Particle Swarm Optimization (QPSO) algorithm is introduced,and the wormhole existence probability and travel distance rate are adaptively adjusted to enhance the global exploration and local development capabilities of the algorithm.Simulation on Matlab shows that the proposed approach reduces MPPT time by more than 45% and improves MPPT accuracy,indicating its better MPPT performance to improve the photovoltaic power generation efficiency.

Abstract:In this paper,we prepare m-Bi₂O₄ via a simple hydrothermal method,and analyze its structure,morphology,surface valence and piezoelectric photocatalytic performance using characterization methods such as XRD,XPS,SEM,TEM,UV-vis DRS and PFM.Furthermore,we test the piezoelectric photocatalytic activity of the material using sulfamethazine (SM) as a simulated pollutant.The results show that m-Bi₂O₄ outperforms BaTiO₃ and BiOCl in catalytic performance,indicated by its high SM degradation efficiency (96.46%) after 60 min of synergistic piezoelectric-light action,and m-Bi₂O₄ still holds high catalytic activity under weakening light energy.In addition,superoxide radicals with high oxidation activity and a small amount of hydroxyl radicals and singlet oxygen have been captured in the reaction system.We also propose a possible mechanism of piezoelectric photocatalysis.

Abstract:Ternary composites of Ni₂P/g-C₃N₄/ZnIn₂S₄ were synthesized via a hydrothermal approach,and their catalytic performance were evaluated by photoreduction of CO₂.Kinds of characterizations (XRD,SEM,TEM,XPS,UV-vis,EIS,and PL) were applied to investigate the morphology,crystal structure,surface chemical states,band structure and photoelectric property of the composites.The results showed that the heterostructure with intense contact was constructed successfully via the facet engineering.Besides,the introduction of Ni₂P and g-C₃N₄ could improve the band structure of photocatalysts,shorten the transmission distance of electrons and inhibit the recombination of photo-induced carriers effectively.Therefore,ternary composites of Ni₂P/g-C₃N₄/ZnIn₂S₄ exhibited higher catalytic activity compared with pure g-C₃N₄ and binary composites of g-C₃N₄/ZnIn₂S₄.Among Ni₂P/g-C₃N₄/ZnIn₂S₄ composites,CNZ5 (Ni₂P:g-C₃N₄:ZnIn₂S₄=1:5:7) revealed the optimal CO₂ photoreduction efficiency,in which the yields of CH₄,CH₃OH,and HCOOH were 114.72 μmol·h^-1·g^-1,17.38 μmol·h^-1·g^-1,and 20.15 μmol·h^-1·g^-1,respectively.Furthermore,the CO₂ photoreduction mechanism was obtained by in-situ DRIFTS,and the intermediates of HCO^-₃ and HCOOH were found during the reaction process.

Abstract:In this article,we investigate the optimization and coordination of the low-carbon supply chain considering triple bottom line under carbon tax policy.First,we examine the decisions of centralized supply chain considering the triple bottom line.Second,we consider four decentralized supply chain models under wholesale price contract,i.e.,Model Ⅰ,Model Ⅱ,Model Ⅲ,and Model Ⅳ.Then we study the four models under two-part tariff contracts,i.e.,Model Ⅰ-LTT,Model Ⅱ-LTT,Model Ⅲ-LTT,and Model Ⅳ-LTT.The results show that the two-part tariff contracts can perfectly coordinate the low-carbon supply chain in Model Ⅰ-LTT and Model Ⅲ-LTT,while realize Pareto improvement of the low-carbon supply chain in Model Ⅱ-LTT and Model Ⅳ-LTT.Finally,numerical analysis is conducted to assess the impact of carbon tax and Carbon Emission Reduction (CER) investment coefficient on total profit of supply chain before and after coordination.This study provides reliable theoretical basis for low-carbon enterprises in selecting appropriate CER strategies and contracts.