查看更多>>摘要：Existing high-precision object detection algorithms for UAV(unmanned aerial vehicle)aerial images often have a large number of parameters and heavy weight,which makes it difficult to be applied to mobile devices.We propose three YOLO-based lightweight object detection networks for UAVs,named YOLO-L,YOLO-S,and YOLO-M,respectively.In YOLO-L,we adopt a deconvolution approach to explore suitable upsampling rules during training to improve the detection accuracy.The convolution-batch normalization-SiLU activation function(CBS)structure is replaced with Ghost CBS to reduce the number of parameters and weight,meanwhile Maxpool max-imum pooling operation is proposed to replace the CBS structure to avoid generating parameters and weight.YOLO-S greatly reduces the weight of the network by directly introducing CSPGhostNeck residual structures,so that the parameters and weight are respectively decreased by about 15％at the expense of 2.4％mAP.And YOLO-M adopts the CSPGhostNeck residual structure and deconvolution to reduce parameters by 5.6％and weight by 5.7％,while mAP only by 1.8％.The results show that the three lightweight detection networks proposed in this paper have good performance in UAV aerial image object detection task.
查看更多>>摘要：Aiming at the low accuracy of existing binocular stereo matching and depth estimation methods,this paper proposes a multi-scale binocular stereo matching network based on semantic association.A semantic associa-tion module is designed to construct the contextual semantic association relationship among the pixels through semantic category and attention mechanism.The disparity of those regions where the disparity is easily estimated can be used to assist the disparity estimation of relatively difficult regions,so as to improve the accuracy of disparity estimation of the whole image.Simultaneously,a multi-scale cost volume computation module is proposed.Unlike the existing methods,which use a single cost volume,the proposed multi-scale cost volume computation module designs multiple cost volumes for features of different scales.The semantic association feature and multi-scale cost volume are aggregated,which fuses the high-level semantic information and the low-level local detailed information to enhance the feature representation for accurate stereo matching.We demonstrate the effectiveness of the proposed solutions on the KITTI2015 binocular stereo matching dataset,and our model achieves comparable or higher matching perfor-mance,compared to other seven classic binocular stereo matching algorithms.
查看更多>>摘要：The current intelligent auxiliary diagnosis models all follow the closed-set recognition setting.After the model is deployed online,the input data is often not completely controlled.Diagnosing an untrained disease as a known category would lead to serious medical malpractice.Therefore,realizing the open-set recognition is significant to the safe operation of the intelligent auxiliary diagnosis model.Currently,most open-set recognition models are studied for natural images,and it is very challenging to obtain clear and concise decision boundaries between known and unknown classes when applied to fine-grained medical images.We propose an open-set recognition network for medical images based on fine-grained data mixture and spatial position constraint loss(FGM-SPCL)in this work.Considering the fine graininess of medical images and the diversity of unknown samples,we propose a fine-grained data mixture(FGM)method to simulate unknown data by performing a mixing operation on known data to expand the coverage of unknown data difficulty levels.In order to obtain a concise and clear decision boundary,we propose a spatial position constraint loss(SPCL)to control the position distribution of prototypes and samples in the feature space and maximize the distance between known classes and unknown classes.We validate on a private ophthalmic OCT dataset,and extensive experiments and analyses demonstrate that FGM-SPCL outperforms state-of-the-art models.
查看更多>>摘要：Object detection is an important task in drone vision.Since the number of objects and their scales always vary greatly in the drone-captured video,small object-oriented feature becomes the bottleneck of model perform-ance,and most existing object detectors tend to underperform in drone-vision scenes.To solve these problems,we propose a novel detector named YOLO-Drone.In the proposed detector,the backbone of YOLO is firstly replaced with ConvNeXt,which is the state-of-the-art one to extract more discriminative features.Then,a novel scale-aware attention(SAA)module is designed in detection head to solve the large disparity scale problem.A scale-sensitive loss(SSL)is also introduced to put more emphasis on object scale to enhance the discriminative ability of the proposed detector.Experimental results on the latest VisDrone 2022 test-challenge dataset(detection track)show that our detector can achieve average precision(AP)of 39.43％,which is tied with the previous state-of-the-art,meanwhile,reducing 39.8％of the computational cost.
查看更多>>摘要：In modern industrial control systems(ICSs),when user retrieving the data stored in field device like smart sensor,there exists two main problems:one is lack of the verification for identification of user and field device;the other is that user and field device need exchange a key to encrypt sensitive data transmitted over the network.We propose a comprehensive authentication and key agreement framework that enables all connected devices in an ICS to mutually authenticate each other and establish a peer-to-peer session key.The framework combines two types of protocols for authentication and session key agreement:The first one is an asymmetric cryptographic key agreement protocol based on transport layer security handshake protocol used for Internet access,while the second one is a newly designed lightweight symmetric cryptographic key agreement protocol specifically for field devices.This proposed lightweight protocol imposes very light computational load and merely employs simple operations like one-way hash function and exclusive-or(XOR)operation.In comparison to other lightweight protocols,our protocol requires the field device to perform fewer computational operations during the authentication phase.The simulation results ob-tained using OpenSSL demonstrates that each authentication and key agreement process in the lightweight protocol requires only 0.005 ms.Our lightweight key agreement protocol satisfies several essential security features,including session key secrecy,identity anonymity,untraceability,integrity,forward secrecy,and mutual authentication.It is capable of resisting impersonation,man-in-the-middle,and replay attacks.We have employed the Gong-Needham-Yahalom(GNY)logic and automated validation of Internet security protocols and application tool to verify the security of our symmetric cryptographic key agreement protocol.
查看更多>>摘要：For electronic voting(e-voting)with a trusted authority,the ballots may be discarded or tampered,so it is attractive to eliminate the dependence on the trusted party.An e-voting protocol,where the final voting result can be calculated by any entity,is known as self-tallying e-voting protocol.To the best of our knowledge,addressing both abortive issue and adaptive issue simultaneously is still an open problem in self-tallying e-voting protocols.Combining Ethereum blockchain with cryptographic technologies,we present a decentralized self-tallying e-voting protocol.We solve the above problem efficiently by utilizing optimized Group Encryption Scheme and standard Ex-ponential ElGamal Cryptosystem.We use zero-knowledge proof and homomorphic encryption to protect votes'secre-cy and achieve self-tallying.All ballots can be verified by anyone and the final voting result can be calculated by any entity.In addition,using the paradigm of score voting and"1-out-of-k"proof,our e-voting system is suitable for a wide range of application scenarios.Experiments show that our protocol is more competitive and more suitable for large-scale voting.
查看更多>>摘要：Backdoor attacks pose great threats to deep neural network models.All existing backdoor attacks are designed for unstructured data(image,voice,and text),but not structured tabular data,which has wide real-world applications,e.g.,recommendation systems,fraud detection,and click-through rate prediction.To bridge this research gap,we make the first attempt to design a backdoor attack framework,named BAD-FM,for tabular data prediction models.Unlike images or voice samples composed of homogeneous pixels or signals with continuous values,tabular data samples contain well-defined heterogeneous fields that are usually sparse and discrete.Tabular data prediction models do not solely rely on deep networks but combine shallow components(e.g.,factorization machine,FM)with deep components to capture sophisticated feature interactions among fields.To tailor the backdoor attack framework to tabular data models,we carefully design field selection and trigger formation algorithms to intensify the influence of the trigger on the backdoored model.We evaluate BAD-FM with extensive experiments on four datasets,i.e.,HUAWEI,Criteo,Avazu,and KDD.The results show that BAD-FM can achieve an attack success rate as high as 100％at a poisoning ratio of 0.001％,outperforming baselines adapted from existing backdoor attacks against unstructured data models.As tabular data prediction models are widely adopted in finance and commerce,our work may raise alarms on the potential risks of these models and spur future research on defenses.
查看更多>>摘要：Joint communication-caching-computing resource allocation in wireless inland waterway communica-tions enables resource-constrained unmanned surface vehicles(USVs)to provision computation-intensive and latency-sensitive tasks forward beyond fifth-generation(B5G)and sixth-generation(6G)era.The power of such resource allo-cation cannot be fully studied unless bidirectional data computation is properly managed.A novel intelligent reflecting surface(IRS)-assisted hybrid UAV-terrestrial network architecture is proposed with bidirectional tasks.The sum of uplink and downlink bandwidth minimization problem is formulated by jointly considering link quality,task execution mode selection,UAVs trajectory,and task execution latency constraints.A heuristic algorithm is proposed to solve the formulated challenging problem.We divide the original challenging problem into two subproblems,i.e.,the joint optimization problem of USVs offloading decision,caching decision and task execution mode selection,and the joint optimization problem of UAVs trajectory and IRS phase shift-vector design.The Karush-Kuhn-Tucker conditions are utilized to solve the first subproblem and the enhanced differential evolution algorithm is proposed to solve the latter one.The results show that the proposed solution can significantly decrease bandwidth consumption in comparison with the selected advanced algorithms.The results also prove that the sum of bandwidth can be remarkably decreased by implementing a higher number of IRS elements.
查看更多>>摘要：Communication and sensing are basically required in intelligent transportation.The combination of two functions can provide a viable way in alleviating concerns about resource limitations.To achieve this,we propose an integrated sensing and communication(ISAC)system based on cellular vehicle-to-everything(C-V2X).We first analyze the feasibility of new radio(NR)waveform for ISAC system.We discuss the possibility of reusing NR waveform for sensing based on current NR-V2X standards.Ambiguity function is calculated to investigate the sensing perform-ance limitation of NR waveform.A C-V2X-based ISAC system is then designed to realize the two tasks in vehicular network simultaneously.We formulate an integrated framework of vehicular communication and automotive sensing using the already-existing NR-V2X network.Based on the proposed ISAC framework,we develop a receiver algorithm for target detection/estimation and communication with minor modifications.We evaluate the performance of the proposed ISAC system with communication throughput,detection probability,and range/velocity estimation accuracy.Simulations show that the proposed system achieves high reliability communication with 99.9999％throughput and high accuracy sensing with errors below 1 m and 1 m/s in vehicle scenarios.
万方数据