基于DCNv2和Transformer Decoder的隧道衬砌裂缝高效检测模型研究

为解决因衬砌裂缝性状随机、分布密集、标注框分辨率低所导致的现有模型识别精度低、检测速度慢及参数量庞大等问题,以第2版可变形卷积网络(DCNv2)和端到端变换器解码器(Transformer Decoder)为基础对YOLOv8网络框架进行改进,提出了面向衬砌裂缝的检测模型DTD-YOLOv8。首先,通过引入DCNv2对YOLOv8主干卷积网络C2f进行融合以实现模型对裂缝形变特征的准确快速感知,同时采用Transformer Decoder对YOLOv8检测头进行替换以实现端到端框架内完整目标检测流程,从而消除因Anchor-free处理模式所带来的计算消耗。采用自建裂缝数据集对SSD,Faster-RCNN,RT-DETR,YOLOv3,YOLOv5,YOLOv8和DTD-YOLOv8的7种检测模型进行对比验证。结果表明:改进模型F1分数和mAP@50值分别为87.05%和89.58%;其中F1分数相较其他6种模型分别提高了14.16%,7.68%,1.55%,41.36%,8.20%和7.40%;mAP@50分别提高了28.84%,15.47%,1.33%,47.65%,10.14%和10.84%。改进模型参数量仅为RT-DETR的三分之一,检测单张图片的速度为16.01 ms,FPS为65.46帧每秒,对比其他模型检测速度得到提升。该模型在面向运营隧道裂缝检测任务需求时能够表现出高效的性能。

Parallel Implementation of the CCSDS Turbo Decoder on GPU

This paper presents a software turbo decoder on graphics processing units(GPU).Unlike previous works,the proposed decoding architecture for turbo codes mainly focuses on the Consultative Committee for Space Data Systems(CCSDS)standard.However,the information frame lengths of the CCSDS turbo codes are not suitable for flexible sub-frame parallelism design.To mitigate this issue,we propose a padding method that inserts several bits before the information frame header.To obtain low-latency performance and high resource utilization,two-level intra-frame parallelisms and an efficient data structure are considered.The presented Max-Log-Map decoder can be adopted to decode the Long Term Evolution(LTE)turbo codes with only small modifications.The proposed CCSDS turbo decoder at 10 iterations on NVIDIA RTX3070 achieves about 150 Mbps and 50Mbps throughputs for the code rates 1/6 and 1/2,respectively.

Quantized Decoders that Maximize Mutual Information for Polar Codes

In this paper,we innovatively associate the mutual information with the frame error rate(FER)performance and propose novel quantized decoders for polar codes.Based on the optimal quantizer of binary-input discrete memoryless channels(BDMCs),the proposed decoders quantize the virtual subchannels of polar codes to maximize mutual information(MMI)between source bits and quantized symbols.The nested structure of polar codes ensures that the MMI quantization can be implemented stage by stage.Simulation results show that the proposed MMI decoders with 4 quantization bits outperform the existing nonuniform quantized decoders that minimize mean-squared error(MMSE)with 4 quantization bits,and yield even better performance than uniform MMI quantized decoders with 5 quantization bits.Furthermore,the proposed 5-bit quantized MMI decoders approach the floating-point decoders with negligible performance loss.

基于encoder-decoder框架的城镇污水厂出水水质预测

由于污水厂的出水水质指标繁多、污水处理过程中反应复杂、时序非线性程度高,基于机理模型的预测方法无法取得理想效果。针对此问题,提出基于深度学习的污水厂出水水质预测方法,并以吉林省某污水厂监测水质为来源数据,利用多种结合encoder-decoder结构的神经网络预测水质。结果显示,所提结构对LSTM和GRU网络预测能力都有一定提升,对长期预测能力提升更加显著,ED-GRU模型效果最佳,短期预测中的4个出水水质指标均方根误差(RMSE)为0.7551、0.2197、0.0734、0.3146,拟合优度(R2)为0.9013、0.9332、0.9167、0.9532,可以预测出水质局部变化,而长期预测中的4个指标RMSE为1.7204、1.7689、0.4478、0.8316,R2为0.4849、0.5507、0.4502、0.7595,可以预测出水质变化趋势,与顺序结构相比,短期预测RMSE降低10%以上,R2增加2%以上,长期预测RMSE降低25%以上,R2增加15%以上。研究结果表明,基于encoder-decoder结构的神经网络可以对污水厂出水水质进行准确预测,为污水处理工艺改进提供技术支撑。

基于时空特征融合的Encoder-Decoder多步4D短期航迹预测

航迹预测在确保空中交通安全、高效运行中扮演着至关重要的角色。所预测的航迹信息是航迹优化、冲突告警等决策工具的输入,而预测准确性取决于模型对航迹序列特征的提取能力。航迹序列数据是具有丰富时空特征的多维时间序列,其中每个变量都呈现出长短期的时间变化模式,并且这些变量之间还存在着相互依赖的空间信息。为了充分提取这种时空特征,本文提出了基于融合时空特征的编码器-解码器(Spatio-Temporal EncoderDecoder,STED)航迹预测模型。在Encoder中使用门控循环单元(Gated Recurrent Unit,GRU)、卷积神经网络(Convolutional Neural Network,CNN)和注意力机制(Attention,AT)构成的双通道网络来分别提取航迹时空特征,Decoder对时空特征进行拼接融合,并利用GRU对融合特征进行学习和递归输出,实现对未来多步航迹信息的预测。利用真实的航迹数据对算法性能进行验证,实验结果表明,所提STED网络模型能够在未来10 min预测范围内进行高精度的短期航迹预测,相比于LSTM、CNN-LSTM和AT-LSTM等数据驱动航迹预测模型具有更高的精度。此外,STED网络模型预测一个航迹点平均耗时为0.002 s,具有良好的实时性。

基于Encoder-Decoder注意力网络的异常驾驶行为在线识别方法

异常驾驶行为是车辆安全运行的重大威胁,其对人员与物资的安全高效投送造成严重危害。以低成本非接触式的手机多传感器数据为基础,通过对驾驶行为特性进行数据分析,提出一种融合Encoder-Decoder深度网络与Attention机制的异常驾驶行为的在线识别方法。该方法由基于LSTM(long short-term memory)的Encoder-Decoder、Attention机制与基于SVM(support vector machine)的分类器3个模块构成。该系统识别方法包括:输入编码、注意力学习、特征解码、序列重构、残差计算与驾驶行为分类等6个步骤。该技术方法利用自然驾驶条件下所采集的手机传感器数据进行实验。实验结果表明:①手机多传感器数据融合方法对驾驶行为识别具备有效性;②异常驾驶行为必然会造成数据异常波动;③Attention机制有助于提升模型学习效果,对所提出模型的识别准确率F1-score为0.717,与经典同类模型比较,准确率得到显著提升;④对于汽车异常驾驶行为来说,SVM比Logistic与随机森林算法具有更优越的识别效果。

面向松木表面缺陷检测的改进RT-DETR模型

为提高松木表面缺陷检测精确度,保证检测速率,该研究提出一种改进RT-DETR的检测模型RIC-DETR。首先,从木材表面缺陷公开数据集中获取图片,并进行标注及数据增强,构建一个包含13642张图片的表面缺陷数据集;其次,对比VGG11、VGG13、ResNet18和VanillaNet13等网络架构,选用计算复杂度低且检测精度较高的ResNet18作为主干特征提取基准网络;然后,引入反向残差移动模块更新ResNet18中的基本块,扩展模型的感受野,改善层间的特征交互;最后,使用EfficientViT模型中的级联分组注意力机制对反向残差移动模块进行二次创新改进,降低计算资源的消耗,提升模型的表达能力。试验结果表明,RIC-DETR的精确率、召回率、平均精度均值分别为95.4%、96.0%、97.2%,均优于目前主流的YOLO系列模型,对比基准模型RT-DETR,RIC-DETR在保持高精度的情况下,参数量、浮点运算量和内存占用量大幅减少,分别降低了54%、57%、52%,同时检测速度可达63.5帧/s。RIC-DETR模型具有复杂度低、准确率高、检测速度快的特点,可为松木的表面缺陷检测提供技术支持。

利用Encoder-Decoder框架的深度学习网络实现绕射波分离及成像

利用单纯绕射波场实现地下地质异常体的识别具有坚实的理论基础,对应的实施方法得到了广泛研究,且有效地应用于实际勘探。但现有技术在微小尺度异常体成像方面收效甚微,相关研究多数以射线传播理论为基础,对于影响绕射波分离成像精度的因素分析并不完备。相较于反射波,由于存在不连续构造而产生的绕射波能量微弱并且相互干涉,同时环境干扰使得绕射波进一步湮没。因此,更高精度的波场分离及单独成像是现阶段基于绕射波超高分辨率处理、解释的重点研究方向。为此,首先针对地球物理勘探中地质异常体的准确定位,以携带高分辨率信息的绕射波为研究对象,系统分析在不同尺度、不同物性参数的异常体情况下绕射波的能量大小及形态特征,掌握绕射波与其他类型波叠加的具体形式;然后根据相应特征性质提出基于深度学习技术的绕射波分离成像方法,即利用Encoder-Decoder框架的空洞卷积网络捕获绕射波场特征,从而实现绕射波分离,基于速度连续性原则构建单纯绕射波场的偏移速度模型并完成最终成像。数据测试表明,该方法最终可满足微小地质异常体高精度识别的需求。

构建改进RT-DETR算法检测隐形眼镜环状波纹缺陷

环状波纹是隐形眼镜制造过程中水凝胶材料分布不均匀造成的、在镜体边沿向内环形收缩的表面缺陷,因在投影检测中难以发现而造成产品质量不良,环状波纹缺陷检测是隐形眼镜产品制造的技术难题之一。本文根据该缺陷特征,搭建了圆环光源照明成像系统,采集了环状波纹缺陷图像模型数据库,引入了一种基于改进RT-DETR的隐形眼镜环状波纹缺陷轻量级检测算法。首先,将RT-DETR原始ResNet18主干提取网络中的BasicBlock替换为轻量级FasterNetBlock。然后,在RT-DETR的Neck部分加入SimAM三通道注意力机制,提高模型的准确度。最后,将GIoU替换为MPDIoU损失函数加快收敛速度,提高检测精度。实验结果表明,相比于原始的RT-DETR算法,改进后的RT-DETR算法在隐形眼镜环状波纹数据库上的mAP@0.5达到了94%,提高了3.1%,Params和FLOPs相比于原始的算法分别降低了15.6%和13%。该算法极大地减小了计算量,有效提高了隐形眼镜环状波纹缺陷的均值平均精度,有望突破隐形眼镜环状波纹缺陷在线检测的技术难题。

基于RT-DETR-Faster的苹果采摘机器人实时目标检测算法

为了解决苹果采摘中目标小,实时性要求高等问题,提出了一种基于RT-DETR的采摘机器人目标检测方法,名为RT-DETR-Faster。首先,采用FasterNet部分卷积替换主干网络的传统卷积,有效提升了模型的运算速度;其次,使用改进的级联注意力编码器替换原始的编码器,使网络更专注于目标区域;最后,引入Faster_Rep融合特征模块,保留更多有效特征并减少计算量。该文在实际的果园图像上进行了实验,结果表明,该文提出的算法与原始的RT-DETR算法相比,FPS提升了34%,帧数达到了47.9,同时准确率更高,适用于苹果采摘机器人的实时目标检测任务。

Unifying Convolution and Transformer Decoder for Textile Fiber Identification

At present,convolutional neural networks(CNNs)and transformers surpass humans in many situations(such as face recognition and object classification),but do not work well in identifying fibers in textile surface images.Hence,this paper proposes an architecture named FiberCT which takes advantages of the feature extraction capability of CNNs and the long-range modeling capability of transformer decoders to adaptively extract multiple types of fiber features.Firstly,the convolution module extracts fiber features from the input textile surface images.Secondly,these features are sent into the transformer decoder module where label embeddings are compared with the features of each type of fibers through multi-head cross-attention and the desired features are pooled adaptively.Finally,an asymmetric loss further purifies the extracted fiber representations.Experiments show that FiberCT can more effectively extract the representations of various types of fibers and improve fiber identification accuracy than state-of-the-art multi-label classification approaches.

Traffic Sign Detection Model Based on Improved RT-DETR

The correct identification of traffic signs plays an important role in automatic driving technology and road safety driving.Therefore,to address the problems of misdetection and omission in traffic sign detection due to the variety of sign types,significant size differences and complex background information,an improved traffic sign detection model for RT-DETR was proposed in this study.Firstly,the HiLo attention mechanism was added to the Attention-based Intra-scale Feature Interaction,which further enhanced the feature extraction capability of the network and improved the detection efficiency on high-resolution images.Secondly,the CAFMFusion feature fusion mechanism was designed,which enabled the network to pay attention to the features in different regions in each channel.Based on this,the model could better capture the remote dependencies and neighborhood feature correlation,improving the feature fusion capability of the model.Finally,the MPDIoU was used as the loss function of the improved model to achieve faster convergence and more accurate regression results.The experimental results on the TT100k-2021 traffic sign dataset showed that the improved model achieves the performance with a precision value of 90.2%,recall value of 88.1%and mAP@0.5 value of 91.6%,which are 4.6%,5.8%,and 4.4%better than the original RT-DETR model respectively.The model effectively improves the problem of poor traffic sign detection and has greater practical value.

A Denoiser for Correlated Noise Channel Decoding: Gated-Neural Network

This letter proposes a sliced-gated-convolutional neural network with belief propagation(SGCNN-BP) architecture for decoding long codes under correlated noise. The basic idea of SGCNNBP is using Neural Networks(NN) to transform the correlated noise into white noise, setting up the optimal condition for a standard BP decoder that takes the output from the NN. A gate-controlled neuron is used to regulate information flow and an optional operation—slicing is adopted to reduce parameters and lower training complexity. Simulation results show that SGCNN-BP has much better performance(with the largest gap being 5dB improvement) than a single BP decoder and achieves a nearly 1dB improvement compared to Fully Convolutional Networks(FCN).

Fire Detection Model Based on Improved RT-DETR

Fire detection has a great impact on people’s life safety.Fire Detection-DETR(FD-DETR)is a fire detection model based on RT-DETR for early fire identification in complex fire scenes.In this study,Adown sub-sampling module was selected to improve the original convolution module,which improved the detection accuracy and reduced the number of parameter values.Using LSKA attention module on the backbone network further improved the detection accuracy.The experimental results showed that compared with the original RT-DETR model,the precision and mAP of FD-DETR flame detection are increased by 0.8%and 0.1%,respectively,which proves that the improved method proposed in this study effectively improves the feature extraction and feature fusion capabilities of the network.In the complex scene fire detection task,the performance of the improved RT-DETR algorithm is better than the original RT-DETR algorithm.

基于BO-RT-DETR的配电线路违规垂钓隐患识别方法

配电线路安全是电网安全运行的关键之一。目前,配电线路巡检主要针对已产生的故障和单一目标的风险隐患,对多目标共同构成的潜在隐患识别存在不足。针对该问题,文章提出了基于贝叶斯优化RT-DETR(Bayesian optimization-based real-time detection transformer,BO-RT-DETR)的配电线路违规垂钓隐患识别方法。通过无人机获取图像数据,基于HGNetv2提取图片的特征,利用高效混合编码器提取高维特征,IoU感知查询提高模型分类精度和IoU值,保障违规垂钓隐患的检测效果。然后,构造多目标隐患识别策略,将贝叶斯优化算法引入隐患识别模型,生成最优参数的隐患识别模型,提高模型的性能。实验结果表明,所提方法的F1分数为97.5%、mPA50为0.972、漏报率为4.7%,比RT-DETR分别提高了0.8%和0.01,漏报率降低了1.5%。

A highly reliable encoding and decoding communication framework based on semantic information

Increasing research has focused on semantic communication,the goal of which is to convey accurately the meaning instead of transmitting symbols from the sender to the receiver.In this paper,we design a novel encoding and decoding semantic communication framework,which adopts the semantic information and the contextual correlations between items to optimize the performance of a communication system over various channels.On the sender side,the average semantic loss caused by the wrong detection is defined,and a semantic source encoding strategy is developed to minimize the average semantic loss.To further improve communication reliability,a decoding strategy that utilizes the semantic and the context information to recover messages is proposed in the receiver.Extensive simulation results validate the superior performance of our strategies over state-of-the-art semantic coding and decoding policies on different communication channels.

Improved Segmented Belief Propagation List Decoding for Polar Codes with Bit-Flipping

Belief propagation list(BPL) decoding for polar codes has attracted more attention due to its inherent parallel nature. However, a large gap still exists with CRC-aided SCL(CA-SCL) decoding.In this work, an improved segmented belief propagation list decoding based on bit flipping(SBPL-BF) is proposed. On the one hand, the proposed algorithm makes use of the cooperative characteristic in BPL decoding such that the codeword is decoded in different BP decoders. Based on this characteristic, the unreliable bits for flipping could be split into multiple subblocks and could be flipped in different decoders simultaneously. On the other hand, a more flexible and effective processing strategy for the priori information of the unfrozen bits that do not need to be flipped is designed to improve the decoding convergence. In addition, this is the first proposal in BPL decoding which jointly optimizes the bit flipping of the information bits and the code bits. In particular, for bit flipping of the code bits, a H-matrix aided bit-flipping algorithm is designed to enhance the accuracy in identifying erroneous code bits. The simulation results show that the proposed algorithm significantly improves the errorcorrection performance of BPL decoding for medium and long codes. It is more than 0.25 d B better than the state-of-the-art BPL decoding at a block error rate(BLER) of 10^(-5), and outperforms CA-SCL decoding in the low signal-to-noise(SNR) region for(1024, 0.5)polar codes.

基于GRU Encoder-decoder和注意力机制的RUL预测方法

深度学习模型可直接建立机械设备的状态与剩余使用寿命(RUL)之间的映射关系,从而避免人工提取特征和建立健康指标的过程。文中基于深度学习理论,提出一种基于注意力机制和时序编码解码器(Encoder-decoder)相结合的RUL预测方法。首先,基于门控循环神经网络(GRU)构建一个时序编码解码器以实现输入序列的重构,其中GRU-Encoder对输入的多元时间序列进行编码;再引入注意力机制对GRU-Encoder在每个时刻的输出向量进行加权融合,以融合后的向量作为编码结果,并将其输入到GRU-Decoder中实现输入序列的重构,同时将编码结果映射为输入样本的RUL。采用CMAPSS数据集对所提方法的有效性进行验证,结果表明,该方法预测精度较高,可行且有效。

The"Decoding Zhonghua"International Conference on Dialogue among Civilisations Held in Beijing

The"Decoding Zhonghua"International Conference on Dialogue among Civilisations,hosted by China International Public Relations Association,China Ethnic News and Academy of Contemporary China and World Studies was held in Beijing on January 17th.With the theme"Pursing Harmonious Coexistence of Civilisations through Dialogue".

基于RT-DETR改进的皮带运输机异物识别方法

凭借兼顾检测精度与速度的特点,YOLO近年来已成为煤炭等工业领域目标检测模型的佼佼者。然而,YOLO的检测性能受到置信度阈值和非极大值抑制阈值等超参数设定的影响。因此,本研究提出了一种基于改进的RT-D ETR带式输送机非煤异物检测模型。该模型无需置信度过滤和非极大值抑制,从而提升了检测精度。此外,针对RT-DETR参数量较大、难以在计算资源有限的边缘设备上部署的问题,我们设计了一种EMA-Faster Net骨干网络,并将颈部网络的AIFI模块替换为LPE-AIFI模块。最后,我们采用TensorRT进行加速,并将模型部署到Jetson Orin Nano边缘计算设备上。实验结果表明,改进后的RT-DETR模型与具有相似参数量的YOLOv8s相比,其召回率高出5.6%,平均类别精度高出4.3%;经TensorRT加速后,模型帧率可达26.4 FPS,满足了实时监测的要求。