A Lightweight Network with Dual Encoder and Cross Feature Fusion for Cement Pavement Crack Detection

Automatic crack detection of cement pavement chiefly benefits from the rapid development of deep learning,with convolutional neural networks(CNN)playing an important role in this field.However,as the performance of crack detection in cement pavement improves,the depth and width of the network structure are significantly increased,which necessitates more computing power and storage space.This limitation hampers the practical implementation of crack detection models on various platforms,particularly portable devices like small mobile devices.To solve these problems,we propose a dual-encoder-based network architecture that focuses on extracting more comprehensive fracture feature information and combines cross-fusion modules and coordinated attention mechanisms formore efficient feature fusion.Firstly,we use small channel convolution to construct shallow feature extractionmodule(SFEM)to extract low-level feature information of cracks in cement pavement images,in order to obtainmore information about cracks in the shallowfeatures of images.In addition,we construct large kernel atrous convolution(LKAC)to enhance crack information,which incorporates coordination attention mechanism for non-crack information filtering,and large kernel atrous convolution with different cores,using different receptive fields to extract more detailed edge and context information.Finally,the three-stage feature map outputs from the shallow feature extraction module is cross-fused with the two-stage feature map outputs from the large kernel atrous convolution module,and the shallow feature and detailed edge feature are fully fused to obtain the final crack prediction map.We evaluate our method on three public crack datasets:DeepCrack,CFD,and Crack500.Experimental results on theDeepCrack dataset demonstrate the effectiveness of our proposed method compared to state-of-the-art crack detection methods,which achieves Precision(P)87.2%,Recall(R)87.7%,and F-score(F1)87.4%.Thanks to our lightweight crack detectionmodel,the parameter count of the model in real-world detection scenarios has been significantly reduced to less than 2M.This advancement also facilitates technical support for portable scene detection.

Remaining Useful Life Prediction of Rail Based on Improved Pulse Separable Convolution Enhanced Transformer Encoder

In order to prevent possible casualties and economic loss, it is critical to accurate prediction of the Remaining Useful Life (RUL) in rail prognostics health management. However, the traditional neural networks is difficult to capture the long-term dependency relationship of the time series in the modeling of the long time series of rail damage, due to the coupling relationship of multi-channel data from multiple sensors. Here, in this paper, a novel RUL prediction model with an enhanced pulse separable convolution is used to solve this issue. Firstly, a coding module based on the improved pulse separable convolutional network is established to effectively model the relationship between the data. To enhance the network, an alternate gradient back propagation method is implemented. And an efficient channel attention (ECA) mechanism is developed for better emphasizing the useful pulse characteristics. Secondly, an optimized Transformer encoder was designed to serve as the backbone of the model. It has the ability to efficiently understand relationship between the data itself and each other at each time step of long time series with a full life cycle. More importantly, the Transformer encoder is improved by integrating pulse maximum pooling to retain more pulse timing characteristics. Finally, based on the characteristics of the front layer, the final predicted RUL value was provided and served as the end-to-end solution. The empirical findings validate the efficacy of the suggested approach in forecasting the rail RUL, surpassing various existing data-driven prognostication techniques. Meanwhile, the proposed method also shows good generalization performance on PHM2012 bearing data set.

ASP源代码加密程序Script Encoder算法研究

微软为了保护脚本代码的安全性 ,以 COM组件的形式提供了一种对脚本代码进行编码加密的技术 .但其安全性到底如何呢 ?本文将通过“反编译”的方式对其加密。

一种融合AutoEncoder与CNN的混合算法用于图像特征提取

深度学习方法在图像的特征提取方面具有优势。针对传统特征提取方法需要先验知识的不足,提出一种自动编码器(Auto Encoder)与卷积神经网络(convolutional neural network,CNN)相结合的深度学习特征提取方法。该方法给Auto Encoder加入快速稀疏性控制,据此对图像训练出基本构件,并初始化CNN的卷积核;同时,给CNN加入了滤波机制,使输出特征保持稀疏性。实验结果表明,在Minist手写数字库和Yale人脸库的识别效果上,提出的特征提取方法均取得了较好的结果,实验进一步通过交叉验证T检验指出,引入滤波机制的特征提取模型优于没有采用滤波机制的模型。

基于Windows Media Encoder实现流媒体同步控制

基于Windows Media技术的流媒体同步控制需要ASF的支持。现有方法要求直接修改ASF文件的头对象而不易编程实现。通过对比研究,提出了一种基于Windows Media Encoder编码器后处理过程间接修改ASF文件的方法,并探讨了该方法的基本原理。实际系统中的成功应用证明了该方法能有效降低编程难度并易于集成,对实现流媒体同步控制具有一定的参考价值。

基于ENCODER_ATT机制的远程监督关系抽取

在信息抽取中,关系抽取是一项准确识别自然语言中实体间关系的关键技术。针对关系抽取模型中容易丢失关键语义特征问题及远程监督的基本假设容易引入噪声数据的问题,本文提出一种基于远程监督的ENCODER_ATT关系抽取模型。基于循环神经网络构造的ENCODER模型在以词级别进行特征记忆提取,并在句子层面进行语义特征信息整合,保证不遗失关键语义特征的同时去除冗余特征。然后在句子层面引入了注意力机制来降低噪声数据对实验结果的影响。在真实的数据集上进行实验,并绘制准确率-召回率曲线,实验结果表明ENCODER_ATT模型对比同类型的关系抽取方法有明显的提升。

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Auto Encoder的智能监控指纹识别系统

针对目前已有的嵌入式指纹识别系统往往采用手工提取,不能自动学习并提取识别所需的特征及识别正确率仍然不高的缺点,提出一种基于自动编码器(Auto Encoder)和LSSVM的指纹识别系统。首先,提出采用FPS200作为指纹传感器采集指纹数据,然后将采集的数据经过滤波和二值化等预处理,通过比较差异算法获得Auto Encoder中的权值和偏置等参数,从而得到训练好的Auto Encoder用于指纹图像特征提取。最后,将自动提取的特征进行训练和分类,将投票最多的分类作为指纹识别的结果。通过测试表明,系统能较精确地实现指纹识别,具有收敛速度快、正确识别率高和匹配时间短的优点。

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。

在Flash中基于Adobe Media Encoder组件的视频导入与应用方法

在flash cs6的默认情况下,Flash cs6只支持flv和f4v格式的视频.如果不是这种格式的视频,我们可以使用Flash cs6自带的视频转换组件Adobe Media Encoder将其他视频格式转换成FLV和F4V格式.本文主要讲解如何使用flash自带的Adobe Media Encoder组件进行视频文件的转换,导入和使用.

Blind recognition of k/n rate convolutional encoders from noisy observation

Blind recognition of convolutional codes is not only essential for cognitive radio, but also for non-cooperative context. This paper is dedicated to the blind identification of rate k/n convolutional encoders in a noisy context based on Walsh-Hadamard transformation and block matrix (WHT-BM). The proposed algorithm constructs a system of noisy linear equations and utilizes all its coefficients to recover parity check matrix. It is able to make use of fault-tolerant feature of WHT, thus providing more accurate results and achieving better error performance in high raw bit error rate (BER) regions. Moreover, it is more computationally efficient with the use of the block matrix (BM) method. © 2017 Beijing Institute of Aerospace Information.

基于AutoEncoder的油气管道控制系统异常状态监测方法

压缩机控制电路的健康状态管理在管道运输中至关重要。通常油气管道压缩机系统部署地点远离城市,环境恶劣,且负荷高、工作时间长,因此故障频发。构建可靠的健康状态检测模型通常需要大量的故障样本,然而在实际数据中,故障样本相对稀缺。采用一种基于自编码器(auto encoder,AE)的单分类方法对油气管道控制系统的异常状态进行辨识。该模型仅需对系统的正常工作状态进行学习,通过编码器可实现特征的自适应提取,从而对数据进行抽象表示,并获得较好的非线性映射能力;当数据分布异常时,系统可区分其与正常信号间的差异,并进行预警。实验部分采用西部输油管道控制系统中实地获取的通信解码信号以及电源信号进行验证,并以单分类支持向量机方法作对比实验,表明了所提出方法的有效性。

基于Encoder-CNN的土壤氮含量光谱预测模型研究

基于光谱的土壤氮含量预测模型泛化能力弱是制约其推广应用的瓶颈。鉴于特征提取及非线性表达能力方面的优势,深度学习模型具有较强的泛化能力。提出一种融合自动编码器和卷积神经网络(Encoder-CNN)的土壤氮含量光谱预测模型,探索模型结构和参数对模型性能的影响。根据以往研究成果和相关性分析,获得180个与氮含量强相关的波长,将其作为Encoder-CNN模型输入,而将土壤氮含量作为模型输出。Encoder-CNN模型利用自动编码器的编码部分进行光谱数据降维,然后输入到卷积神经网络进行土壤氮含量预测。设计2种网络结构,每种网络结构包含2种不同参数设置,共4个模型,用以探索Encoder-CNN土壤氮含量光谱预测模型结构和参数对模型性能的影响。利用公开数据集LUCAS对模型进行训练。按3σ原则对公开数据集LUCAS进行异常值检测与处理,获得20791个数据,其中18711个样本作为训练集,2080个样本作为测试集,对Encoder-CNN模型进行训练。结果表明:对于自动编码器,在相同隐含层数下,最后的隐含层神经元个数为30时,复现效果最优。增加隐含层数,会提升复现效果。增加卷积核数量,特别是尺寸为1×1卷积核,能够提高模型的预测性能与可靠性。增加池化层的网络结构,模型预测精度提升至0.90以上。增加全连接层神经元数量也会提升模型性能。利用自采集的黑龙江黑土实时光谱数据集进行模型迁移,观察模型泛化能力。当模型迭代100次后,在黑龙江数据集上的预测精度即可达到0.90以上;当迭代次数为900时,模型在训练集和测试集上的预测精度可以达到0.98。结果表明,所构建的Encoder-CNN土壤氮含量光谱预测模型具有较好的泛化能力。

A correction method of encoder bias in satellite laser ranging system

In a satellite laser ranging telescope system, well-aligned encoders of the elevation and azimuth axes are essential for tracking objects. However, it is very difficult and time-consuming to correct the bias between the absolute-position indices of the encoders and the astronomical coordinates, especially in the absence of a finder scope for our system. To solve this problem, a method is presented based on the phenomenon that all stars move anti-clockwise around Polaris in the northern hemisphere. Tests of the proposed adjustment procedure in a satellite laser ranging （SLR）system demonstrated the effectiveness and the time saved by using the approach, which greatly facilitates the optimization of a trackin~ svstem.

Determination of optimal period of absolute encoders with single track cyclic gray code

Low cost and miniaturized rotary encoders are important in automatic and precise production. Presented here is a code called Single Track Cyclic Gray Code (STCGC) that is an image etched on a single circular track of a rotary encoder disk read by a group of even spread reading heads to provide a unique codeword for every angular position and features such that every two adjacent words differ in exactly one component, thus avoiding coarse error. The existing construction or combination methods are helpful but not sufficient in determining the period of the STCGC of large word length and the theoretical approach needs further development to extend the word length. Three principles, such as the seed combination, short code removal and ergodicity examination were put forward that suffice determination of the optimal period for such absolute rotary encoders using STCGC with even spread heads. The optimal periods of STCGC in 3 through 29 bit length were determined and listed.

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。

Application of Instantaneous Rotational Speed to Detect Gearbox Faults Based on Double Encoders

Considerable studies have been carried out on fault diagnosis of gears, with most of them concentrated on conventional vibration analysis. However, besides the complexity of gear dynamics, the diagnosis results in terms of vibration signal are easily misjudged owing to the interference of sensor position or other components. In this paper, an alternative gearbox fault detection method based on the instantaneous rotational speed is proposed because of its advantages over vibration analysis. Depending on the timer/counter-based method for the pulse signal of the optical encoder, the varying rotational speed can be obtained e ectively. Owing to the coupling and meshing of gears in transmission, the excitations are the same for the instantaneous rotational speed of the input and output shafts. Thus, the di erential signal of instantaneous rotational speeds can be adopted to eliminate the e ect of the interference excitations and extract the associated feature of the localized fault e ectively. With the experiments on multistage gearbox test system, the di erential signal of instantaneous speeds is compared with other signals. It is proved that localized faults in the gearbox generate small angular speed fluctuations, which are measurable with an optical encoder. Using the di erential signal of instantaneous speeds, the fault characteristics are extracted in the spectrum where the deterministic frequency component and its harmonics corresponding to crack fault characteristics are displayed clearly.

Absolute multi-pole encoder with a simple structure based on an improved gray code to enhance the resolution

We developed a novel absolute multi-pole encoder structure to improve the resolution of the multi-pole encoder, realize absolute output and reduce the manufacturing cost of the encoder. The structure includes two ring alnicos defined as index track and sub-division track, respectively. The index track is magnetized based on the improved gray code, with linear halls placed around the track evenly. The outputs of linear halls show the region the rotor belongs to. The sub-division track is magnetized to N-S-N-S (north-south-north-south), and the number of N-S pole pairs is determined by the index track. Three linear hall sensors with an air-gap of 2 mm are used to translate the magnetic filed to voltage signals. The relative offset in a single N-S is obtained through look-up. The magnetic encoder is calibrated using a higher-resolution incremental optical encoder. The pulse output from the optical encoder and hall signals from the magnetic encoder are sampled at the same time and transmitted to a computer, and the relation between them is calculated, and stored in the FLASH of MCU (micro controller unit) for look-up. In the working state, the absolute angle is derived by looking-up with hall signals. The structure is simple and the manufacturing cost is very low and suitable for mass production.

基于改进Encoder-Decoder模型的新闻摘要生成方法

针对通过Extractive方式实现自动文摘而存在文本连贯性欠缺和出现未登录词问题,提出一种基于改进Encoder-Decoder模型的新闻摘要生成方法。首先,在数据预处理的过程中融入额外的语言特征,如词语的词性和TF-IDF,使词语具有多维度的含义;其次,采用Decoder/Pointer机制在摘要中指向原文本中的位置对低频词进行处理;最后,采用注意力机制来协助模型记忆输入数据并确定其注意程度。在News2016zh数据集上进行实验,结果表明基于改进Encoder-Decoder模型与基线Encoder-Decoder相比,ROUGE-1、ROUGE-2和ROUGE-L值分别提高了32.1%、30.5%和32.5%,在摘要连贯性方面也得到了较好提升。

Design and implementation of LDPC encoder based on FPGA

A low density parity check(LDPC)encoder with the codes of(8176,7154)and encoding rate of 7/8 under CCSDS standard for near space communication is designed.Based on LDPC encoding theory,the FPGA-based coding algorithm is designed.Based on the characteristics of LDPC generating matrix,the cyclic shift register is introduced as the core of the encoding circuit,and the shift-register-Adder-Accumulator(SRAA)structure is adopted to realize the fast calculation of matrix multiplication,so as to construct the encoding module with partial parallel encoding circuit as the core.In addition,the serial port input and output module,RAM storage module and control module are also designed,which together constitute the encoder system.The design scheme is implemented by FPGA hardware and verified by simulation and experiment.The results show that the test results of the designed LDPC encoder are consistent with the theoretical results.Therefore,the coding system is practical,and the design method is simple and efficient.