深度学习是机器学习中的一个新的研究领域。通过深度学习的方法构建深度网络来抽取特征是目
前目标和行为识别中得到关注的研究方向。为引起更多计算机视觉领域研究者对深度学习进行探索和讨论，并推
动目标和行为识别的研究，对深度学习及其在目标和行为识别中的新进展给予概述。方法首先介绍深度学习领
域研究的基本状况、主要概念和原理 然后介绍近期利用深度学习在目标和行为识别应用中的一些新进展。结
果阐述了深度学习与神经网络之间的关系，深度学习的优缺点，以及目前深度学习理论需要解决的主要问题。
结论该文对拟将深度学习应用于目标和行为识别的研究人员有所帮助。
Depth learning machine learning is a new field of research. To extract features to build depth by deep learning of the network is to head
Before the objectives and behavior recognition obtained in research attention. Computer vision is to attract more researchers to explore the depth of study and discussion, and push
Moving targets and conduct research to identify the depth of learning and new progress in the goals and behavior recognition to give an overview. The method first introduced collar depth study
Basic Situation field research, the main concepts and principles then describes recent advances in deep learning some new goals and behavior recognition applications use. Knot
Fruit describes the relationship between the depth of learning and neural networks, the advantages and disadvantages of deep learning and deep learning theory main problems to be solved.
Conclusion In this paper intends to apply to the target depth of learning and behavior recognition researchers help. (2015-11-03, PDF, 965KB, 下载21次)
NFC MIFARE FELICA S50 S70 P2P ID卡 动物标签 18092 14443 技术详细介绍
(1)——概念、分类
(2)——国际标准
(3)——能量、调制
(4)——数据编码
(5)——防冲突
(6)——通讯协议概述
(7)——ID卡
(8)——动物标签
(9)——动物标签HDX
(10)——识别号的格式变化
(11)——Mifare系列卡的共性
(12)——三次相互认证
(13)——Mifare S50与Mifare S70
(14)——Mifare S50与S70的存取控制
(15)——Mifare1的安全性及7字节序列号M1卡
(16)——Mifare?UltraLight
(17)——射频卡中数据的存储形式
(18)——Mifare Desfire
(19)——Desfire的3次握手认证和段密码生成
(20)——RC系列射频接口芯片
(21)——RC系列射频芯片的天线设计
(22)——RC系列射频芯片的寄存器操作
(23)——ISO15693的载波、调制与编码
(24)——ISO15693的防冲突与传输协议
(25)——Felica简介
(26)——Felica的文件系统
(27)——CPU卡概述
(28)——基于MF1射频卡的酒店门锁设计
(29)——射频接口芯片TRF7960
(30)——PN512的寄存器设置
NFC MIFARE FELICA S50 S70 P2P ID card animal tag 1809214443 technical details (1)- Concept, classification (2)- International Standard (3)- the energy modulation (4)- data encoding (5)- Anti-collision (6)- Communication Protocol Overview (7)- ID cards (8)- Animal tags (9)- Animal label HDX (10)- Changes format identification number (11)- Mifare series cards common (12)- three mutual authentication (13)- Mifare S50 and Mifare S70 (14)- Mifare S50 and S70 of the access control (15)- Mifare1 safety and 7-byte sequence number M1 card (16)- Mifare UltraLight (17)-? to store data in the form of RF card (18)- Mifare Desfire (19)- Desfire three-way handshake authentication and password generation segment (20)- RC Series RF Interface Chip (21)- RC Series RF chip antenna design (22)- RC Series RF chip register operations (23)- ISO15693 carrier modulation and coding (24)- ISO15693 anti-collision and transmission protocol (25)- Felica Introduction (26)- Felica file system (27)- CPU card Overview ( (2014-11-21, PDF, 27918KB, 下载85次)
PID控制包的功能块(FB)包括了用于连续控制(CONT_C)和步进控制
(CONT_S)的控制块，以及用于脉宽调制(PULSEGEN)的FB。
控制器块实现了一个纯软件控制器，它所包含的功能块可以提供控制器
的所有功能。循环计算所需的数据存储在为FB分配的数据块中。这样，
便可以按照需要频繁地调用FB。
功能块PULSEGEN和功能块CONT_C一起，可实现具有脉冲输出的控制
器，用于控制比例执行器。
PID control packet function blocks (FB) includes a control block for continuous control (CONT_C) and stepper control (CONT_S), and FB for pulse-width modulation (PULSEGEN)' s. The controller block implements a pure software controller function blocks can provide it contains all the features of the controller. Cycle to calculate the required data is stored in the FB allocated data blocks. Thus, they can call the FB frequently as needed. Function blocks and function blocks CONT_C PULSEGEN together, can achieve a pulse output controller for controlling the ratio of the actuator. (2014-07-26, PDF, 583KB, 下载4次)
摘要：新型元启发式算法例如粒子群算法，萤火虫算法，和声搜索算法已经成为现今复杂的优化问题的有效解决方法。该文基于蝙
蝠的回声定位行为提出了一种新型的元启发式算法———蝙蝠算法，同时也将现有的一些算法的优点引入到该算法中。 改文对该算
法进行了详细的公式化表述并对其执行流程的作出了说明，并且将该算法与遗传算法、粒子群优化算法等算法进行了比较。仿真结
果表明,蝙蝠算法明显优于其他算法，并对进一步的研究作出了展望。
Summary: The new meta-heuristic algorithms such as particle swarm optimization, firefly algorithm, harmony search algorithm has become an effective solution to today' s complex optimization problems. In this paper, based on bat echolocation behavior presents a new meta-heuristic algorithm--- bat, but also some of the advantages of the existing algorithms is introduced into the algorithm. Change the text of the algorithm in detail and make a note formulaic expressions its implementation process, and the algorithm and genetic algorithms, particle swarm optimization algorithm are compared. Simulation results show that the algorithm is superior to other algorithms bat, and made the prospects for further research. (2013-12-17, PDF, 615KB, 下载53次)
基于循环统计理论, 对循环平稳信号进行处理, 主要研究了信号的二阶循环统计特性, 即循环自相关函数和循环谱密度, 指出循环自相关函数不为零的循环频率对应着信号中的某些故障, 并
可以对调幅信号进行解调. 通过循环频率扫描方法提取的调制源分布在循环频率域的低频段, 其结
果可用循环频率-频率- 循环谱密度的三维图表示. 用仿真信号对该方法进行验证, 并应用于滚动轴承的内、外圈及滚动体的故障诊断, 可以有效地分离出所对应的故障特征频率.
Statistical theory based on the cycle of cyclostationary signal processing, mainly studied the statistical properties of the signal of second cycle, ie cyclic autocorrelation function and cyclic spectral density, noting that cyclic autocorrelation function is not zero cycle frequency corresponds to a certain signal These faults, and can be demodulated AM signal through cyclic frequency scanning method to extract the modulation source distributed in the frequency domain of the low frequency cycle, the results can be cyclic frequency- Frequency- loop spectral density of the three-dimensional graph. using simulated signals right This method is validated and used rolling bearing, the outer ring and rolling fault diagnosis can be effectively isolated from the corresponding fault characteristic frequency. (2013-08-15, PDF, 166KB, 下载60次)
新一代高性能无人机飞控系统的研究与设计
张小林 赵宇博 范力思
I n o r de r t o cau se t he U A V f lig ht co nt r o l sy st e m has t he f o r mida ble da t a- ha ndling ca pa cit y , t h e lo w po we r lo ss , t he st r o ng
f le x ibilit y an d a hig he r int e g r at io n r a t e, pr o po sed o ne kind of t e chn ol og y ba sed on SO P C w hic h ca n so lv e t hes e p r ob lem s o n U A V f lig ht co n
t r o l sy st e m. T r a nsf e r s m any N io s so f t pr o c ess o r . A lt e r a br in gi ng t he I P co r e a s w el l as t he per ip her y ha r dw a r e cir c uit . on e kind o f h ig h
per f o r man ce f lig ht co nt r o l sy st em h as be en de sig ned. Co mpa r e s w it h t he t r a diti on al U A V co nt r o l sy s te m, t his o ne hav e ve r y st r o ng dat a handl ing capa cit y , t h e sma ll v o lume a nd lo w po w e r lo s s. T he a ct ua l fl ig ht r esul t indic at e d : Ea ch m odu la r de sig n is r ea so nab le, t he o ve r al l
sy st em mo v e ment is st a ble. T his sy st e m ca n b e ser ve d a s t he ne x t g ene r (2013-05-02, PDF, 606KB, 下载5次)
传统Snake 模型存在的缺点是, 其初始轮廓必须靠近图像中感兴趣目标的真实边缘,否则会得到错误结
果,且由于Snake 模型的非凸性,结果不能进入感兴趣目标的深凹部分,很容易陷入局部极小点. 由此该文提出一
种基于力场分析的主动轮廓模型,详细分析了基于欧氏距离变换的距离势能力场分布,归纳出感兴趣目标上真轮
廓点与假轮廓点的判别标准. 建立了由曲线能量到最终结果的有效方法,避免了Snake 陷入局部极小点. 实验结果
表明,该模型具有较大的捕获区域,能够进入感兴趣目标的深凹部分,准确提取感兴趣目标的轮廓. 与GVFSnake
模型相比, 该模型具有很小的计算量.
The t raditional snake initial contour should be close to the t rue boundary of interested ob2
ject in an image , or else it would converge to the wrong result . Next , active contours have difficulties
progressing into boundary concavities. Moreover , the t raditional snake and it s almost kinds of im2
proved methods are easy to get into local minimum because snake models are non2convex. An active
contour model based on force field analysis , namely FFASnake model , is presented in this paper.
Based on analyzing force dist ribution rules of distance potential force field , a standard is int roduced
here to distinguish the false one f rom contour point s. (2012-12-28, PDF, 318KB, 下载21次)
CAN总线控制器设计，CAN 是Controller Area Network 的缩写（以下称为CAN），是ISO国际标准化的串行通信协议。在当前的汽车产业中，出于对安全性、舒适性、方便性、低公害、低成本的要求，各种各样的电子控制系统被开发了出来。由于这些系统之间通信所用的数据类型及对可靠性的要求不尽相同，由多条总线构成的情况很多，线束的数量也随之增加。为适应“减少线束的数量”、“通过多个LAN，进行大量数据的高速通信”的需要，1986 年德国电气商博世公司开发出面向汽车的CAN 通信协议。此后，CAN 通过ISO11898 及ISO11519 进行了标准化，现在在欧洲已是汽车网络的标准协议。
CAN bus controller design, CAN stands for Controller Area Network (hereinafter referred to as CAN) is a serial communication protocol of the ISO international standardization. The current auto industry, for safety, comfort, convenience, low pollution, low-cost requirements of a wide range of electronic control systems have been developed out. Different data types used for communication between the system and reliability requirements, multiple bus Harness the number increases. To adapt to " reduce the number of harnesses, a large number of high-speed data communication needs of the number of the LAN, in 1986 the German Electrical and Bosch developed the CAN communication protocol for automotive. Since then, CAN ISO11898 and ISO11519 standardization in the European automotive network standard protocols. (2012-08-15, PDF, 2259KB, 下载13次)
对于InSAR 系统,当波长较长时,干涉图条纹频率较低,相位解缠容易,但反演的高程精度较低
而当波长较短时,干涉图条纹频率较高,相位解缠困难,但反演的高程精度较高. 为了充分利用多波段
InSAR 干涉相位之间的互补信息,设计了利用中国余数定理的多波段InSAR 相位解缠方法,采用由DEM
仿真的多波段干涉图进行了相位解缠实验,得到了理想的解缠结果,成功解决了因地形起伏较大或基线较长而引起的干涉相位欠采样处的相位解缠难题.
For the InSAR System熏when the wavelength is longer熏the frequency of fringes will be lower熏the
phase unwrapping will be easier熏and the accuracy of derived heights will be lower. However熏when the
wavelength is shorter熏the frequency of fringes will be higher熏the phase unwrapping will be more difficult熏and
the accuracy of derived heights will be higher. In order to make the best use of compensatory information in
multi鄄band InSAR phases熏the phase unwrapping algorithm with the Chinese Remainder Theorem for multi鄄band
InSAR is designed. The multi鄄band interferograms simulated from DEM are applied to do phase unwrapping
experiments. The satisfying phase unwrapping results are obtained. The difficulty in phase unwrapping of sub鄄
sampling interferometric phases due to great hypsography or longer baseline is removed. (2012-03-22, PDF, 6622KB, 下载63次)
对于多输入多输出系统, 传统的基于训练序列的信道辨识方法为跟踪信道的变化, 必须
发送大量冗余信息, 传输效率较低 而基于子空间的纯盲方法又带有不可避免的模糊性, 辨识结果
精度不够。在子空间法的基础上, 利用少量训练符号估计出模糊矩阵, 以此矫正子空间法得到的结
果。仿真结果显示: 在保证辨识精度的前提下, 该方法减少了训练符号等冗余, 提高了传输效率。
In o rder to t rack the channel variat ion of MIMO OFDM sy stems, t radit io nal channel
identif icat io n methods based on training sequence had to send larg e amount of redundant informat ion,
w hich result s in low er eff iciency o f t ransmissio n. Meanw hile, the blind method based o n subspace may
have inevitable ambiguity and low er precision of identif icat io n. In this paper, an est imat io n method fo r the
ambiguity mat rix is proposed, which utilizes less tr aining symbo ls to rect ify the result s f rom subspace
method. T he simulation result s show that this pro posed algo rithm can reduce the amount of t raining
symbols and incr ease the ef ficiency of t ransmission. (2011-12-11, PDF, 244KB, 下载23次)
本篇論文提出的整合式解交錯(Integrated De-interlacing)的演算法，可以有效提昇移
動區域的畫面，但是當移動估計不正確時，反而會使移動補償後的畫面變得很差，為了
改善這種情況，因此結合移動可適性解交錯的優點，並將空間圖場內插(Spatial
Interpolation)的方式改成ELA(Edge Line Average)來設計，經過電腦模擬的結果發現，不僅在視覺上提高畫面的解析度，在某些影像峰值訊號雜訊比(Peak Signal Noise Ratio ,
PSNR)也比線平均解交 錯(Line Average De-interlacing)多出好幾分貝的畫質增益。
此外，在整合式解交錯演算法中也增加影片偵測(Film Detection)和影像加強(Image
Enhancement)的演算法設計，在這樣演算法的架構下，透過影片偵測的演算法，我們可
真實地還原3:2 Pull Down 的影片格式，而不會有鋸齒狀(Saw-Toothed)的畫面出現，而影
像加強的演算法，則可以在解交錯後，經過影像的調整，使輸出畫面呈現不同的效果，
達到消費者的需求。
The main theme of this thesis is an integrated de-interlacing system, which incorporates
several known and improved techniques in a nice manner to produce good de-interlaced image
quality. We first develop an accurate motion detector that classifies image regions into
stationary, low-motion, and high-motion categories. The simple field merging method is
applied to the stationary regions. The edge line average interpolation method is applied to the
slow-motion regions. Finally, the motion-compensated interpolation is applied to the
high-motion regions. In addition, hierarchical motion estimation and motion vector smoothing
techniques are employed to enhance the quality of estimated motion vectors. Our computer
simulation shows that the subjective image quality is improved by using the proposed scheme.
Also, its PSNR measures are better than the conventional spatial or temporal interpolation
schemes. (2010-10-26, PDF, 1144KB, 下载10次)
分析快速细化算法和 OPTA 细化算法不足产生的内在原因 ,提出一种新的基于重心的快速细化算法.该算
法根据被细化图像的特点 ,用密度重心快速将纹线细化到 3 个像素宽度内 ,计算 4 邻域拓扑实现彻底细化.仿真结
果表明 ,在细化效率方面 ,该算法一次遍历删除超过一半的大量冗余像素 ,是快速细化算法的 3～7 倍 在细化要求
方面 ,该算法可达到绝对单像素、 光滑无毛刺 ,并能保持端点不被吞噬 ,能够很好地满足图像细化的要求.
The int rinsic origin of deficiencies in both t he fast t hinning algorit hm ( F TA) and one2pass
t hinning algorithm (OPTA) was analyzed , and a barycenter based fast t hinning algorit hm was presented.
According to t he image’ s feat ure to be fined , the veins of t he image were fined to be of t hree pixel in widt h
by using density barycenter and t hen t he 42neighborhood topology was comp uted to make complete
t hinning. Simulation result showed t hat , in connection wit h t he t hinning efficiency , t he proposed algo2
rit hm exhibited so perfect ly that a great amount (more t han a half ) of redundant pixel s was eliminated af2
ter single overall checking , being 3 to 7 times faster t han t hat wit h fast t hinning algorit hm. As to t he re2
quirement of t hinning , this algorithm could implement absolute single2pixel t hinning wit hout blurred edges
and keep the ends un2swallowed , so that t he requirement of image t hinning was well met . (2010-05-12, PDF, 531KB, 下载55次)
该文档提出了一种应用于开关电容流水线模数转换器的CMoS预运放一锁存比较
器．该比较器采用UMC混合／射频0．18肛m 1P6M P衬底双阱CMOS工艺设计，工作电压为
1．8 V．该比较器的灵敏度为0．215 mV，最大失调电压为12 mV，差分输入动态范围为1．8
V，分辨率为8位，在40 M的工作频率下，功耗仅为24．4 ttW．基于0．18 gm工艺的仿真结
果验证了比较器设计的有效性．
A CMOS preamplifier-latch comparator used in switched··capacitor pipeline analog··to-digital con·-
verter WBS presented．The comparator WaS d髑igned under UMC Mixed．Mode／RF 0．18 btm 1P6M P．Sub Twin—
Well CMOS process and worked with 1．8V power supply．The sensitivity of the comparator was 0．215 mV，
the largest offset voltage was 12 mV，the differentiaI input range Was 1．8 V，the resolution was 8 bit and the
power dissipation Was only 24．4 gW at 40 MHz．HSPICE simulations of the comparator implemented in a 0．18
um technology demonstrate its effectiveness． (2010-03-26, PDF, 353KB, 下载16次)
Linux 价格低廉、功能强大，可以运行在X86,Alpha,Sparc,MIPS,PPC,MOTOROLA，NEC，ARM
等硬件平台上，而且开放源代码，可以定制。我们所介绍的硬件平台是基于ARM 体系结构，
由北京博创兴业科技有限公司开发的UP-NetARM3000 和UP-NetARM2410-S 实验仪器。
UP-NetARM3000 的CPU为ARM7TDMI内核的三星S3C44B0X01 芯片，由于没有MMU（内存管理单
元）只能运行uClinux，UP-NetARM2410-S的CPU为ARM920T内核的三星S3c2410 芯片，由于
有MMU可以运行标准的ARM-LINUX内核。通过这两个平台，我们可以学习嵌入式LINUX中的针
对有MMU和无MMU的不同开发过程
Linux cheap, powerful, can run on X86, Alpha, Sparc, MIPS, PPC, MOTOROLA, NEC, ARM, such as hardware platform, but also open source, can be customized. We introduce the hardware platform is based on the ARM architecture, from Beijing Fiberxon Industrial Technology Development Co., Ltd. UP-NetARM3000 and UP-NetARM2410-S experimental apparatus. UP-NetARM3000 the CPU for the ARM7TDMI core S3C44B0X01 Samsung chip, in the absence of MMU (memory management unit) can only run uClinux, UP-NetARM2410-S for the CPU Samsung ARM920T kernel S3C2410 chip, as has MMU can run standard ARM-LINUX kernel. Through these two platforms, we can learn for embedded LINUX in MMU and MMU have different development process (2008-05-18, PDF, 5938KB, 下载36次)
用单片机实现电梯控制系统
电梯控制系统的工作原理是：当某
层有要梯信号输入时，呼梯信号锁存系
统将要梯信号锁存，待单片机查询到要
梯信号后，根据要梯信号的位置(即楼
层数)和电梯所处的位置，决定电梯运
行方向，并启动电梯到要梯层停梯、开
门，待乘客进入电梯关门后，再根据乘
客要求把乘客送到目的层。
with MCU elevator control system of the elevator control system working principle is : When a staircase layer to the input signal, calling signal latch system will be staircase signal latches, SCM pending inquiries to the staircase to signal, according to the location staircase signal (that is, number of floors) and the locations of elevators, decide the direction of the lift operation, and start to the elevator to stop staircase layer Ossetia, opened the door, the passengers entered the elevator closed, According passengers require passengers to the target zone. (2007-03-24, PDF, 109KB, 下载113次)
无线应用协议（Wireless Application Protocol，WA P）是WA P论坛经过不断努力得到的成
果，它提供了一个业界技术规范，以便开发出适用于各种无线通信网络的应用和业务。WA P
论坛的工作范围就是为各种业务和应用制定一系列的技术规范。无线市场正在快速增长，新
的用户不断增多，新的业务不断涌现。为了使运营商和生产者能够从容面对先进业务、多种
类业务和快速、灵活的业务生成等诸多的挑战， WA P规定了一系列传输层、会话层和应用层
协议。
Wireless Application Protocol (Wireless Application Protocol. WA P) P Forum WA is a result of efforts the results, it provides an industry technical standards, developed to apply to all kinds of wireless communication networks and business applications. WA P scope of the work of the Forum is for various business applications and formulated a series of technical standards. Wireless market is growing rapidly, new users are increasing and new business continued to emerge. To enable operators and producers to calmly face the advanced business, and more types of business fast, flexible service creation and many other challenges, WA P provides a series of transport layer, session layer and application layer protocol. (2006-11-20, PDF, 82KB, 下载274次)