Emerging memories have been developed as new physical infrastructures for hosting neural networks owing to their low-power analog computing characteristics. However, accurately and efficiently programming devices in an analog-valued array is still largely limited by the intrinsic physical non-idealities of the devices, thus hampering their applications in in-situ training of neural networks. Here, we demonstrate a passive electrochemical memory (ECRAM) array with many important characteristics necessary for accurate analog programming. Different image patterns can be open-loop and serially programmed into our ECRAM array, achieving high programming accuracies without any feedback adjustments. The excellent open-loop analog programmability has led us to in-situ train a bilayer neural network and reached software-like classification accuracy of 99.4% to detect poisonous mushrooms. The training capability is further studied in simulation for large-scale neural networks such as VGG-8. Our results present a new solution for implementing learning functions in an artificial intelligence hardware using emerging memories.

自"新工科"建设等提出以来,如何强化学生解决复杂工程问题的能力培养,成为中国高等教育的重要课题.以多学科技术融合软硬件平台为基础,设计了贯穿本科4年的工程教育新模式.平台集成传感、通信、智能识别、机电一体等要素,配合企业项目制引导的课程群建设,形成了由浅入深、逐步高阶的工程能力培养机制.多年来的教学实践显示,该模式能有效提高学生的专业兴趣、创新能力和复杂工程问题的解决能力.

One of the key hot spots and difficulties in the construction of “Engineering Education Certification” and “New Engineering” is how to strengthen the cultivation of the college students’ ability to solve complex engineering problems. In this paper, a software and hardware experimental platform has been designed, which organically integrates automatic judgement and sensing communications with electromechanical integrated IoT measurement and control. And a teaching system of cross-grade supporting experimental course group based on this platform has been proposed, which is driven by the application function topics from the shallower to the deeper. It is an engineering ability training mechanism that can continuously carry out experimental teaching such as design, installation and adjustment from the lower grade to the higher grade. More than five years of teaching practice shows that this system has significantly improved the students’ professional research interest and ability to solve complex engineering problems.

在机器视觉领域,许多应用对摄像机的动态范围提出了苛刻的要求.对数曲线响应传感器的出现为图像动态范围的提升奠定了硬件基础.然而目前针对该传感器的曝光控制均需人工操作评估,自动性、实时性差.为了解决这个问题,提出了一种基于对数曲线响应传感器的高动态范围自动曝光算法,并基于现场可编程门阵列(FPGA)并行计算在硬件上实现该算法,从而能够在图像前端实现对物理场景的实时高动态曝光.通过对感兴区间和目标区间的自适应选择,在保留阴暗区域细节的基础上实现过曝区域的复原,提升图像动态范围.

In this paper we present a CCD image sensor system with high dynamic range. This feature is achieved through double analog-to-digital convertors’(ADC) architecture and field programmable gate arrays(FPGA). By doing so, the system outputs high dynamic range images in real time. The proposed scheme is a low-cost solution in the sense that it can be built on top of any traditional sensor system with minor modifications on the system-level analog-to-digital module. Through the post-layout sensor system we show the experimental images to demonstrate the effectiveness.

A totally new architecture for clock recovery is delivered in the article.Very simple circuit is designed for time-to-voltage conversion,which transforms time error into voltage error.It is very convenient to use the transformed voltage for error detection.The voltage is processed by a self-adaptive peak value capture circuit to realize voltage comparison and clock self-recovery.The circuit is designed with SMIC 0.18μm process lib.Systemic circuit simulation was carried out and experiment results indicate that the designed circuit can work within a wide band of 50MHz to 150MHz,which meets the requirement of clock recovery in a VLSI system design.

General Packet Radio Service(GPRS) provides GPRS users packet switched data service.However,in existing GPRS applications,GPRS specification does not support push access such as call termination(incoming call).For GPRS supporting terminals,the paper proposes a push access for GPRS supporting Terminals,based on Packet Data Protocol time activation and short message service,respectively.

An automatic extraction method of container identity codes based on template matching is proposed. With different kinds of noises on the image, the container code can hardly be extracted. Initially, the container image is filtered with both adaptive linear and nonlinear filters in order to reduce noise so that the candidate text lines can be properly located. Then, a series of standard templates has been brought forward according to the standard align modes of the container identification (ID) codes. Finally, the align mode of each candidate text line is obtained and then matched with those standard templates and the container ID codes can be extracted automatically. Results show that this method can segment the container ID codes with high performance.

An embedded traffic information collecting system in mu Clinux environment was developed. The system was mainly comprised of four parts--a camera used to capture the raw traffic video, a piece of DSP 6203B for traffic data processing, a compression module for the purpose of reducing the amount of data before transmitting them through network, and finally, a network module which was used for data transmission.

An embedded traffic information collecting system in μClinux environment was developed. The system was mainly comprised of four parts - a camera used to capture the raw traffic video, a piece of DSP 6203B for traffic data processing, a compression module for the purpose of reducing the amount of data before transmitting them through network, and finally, a network module which was used for data transmission.

Video-based vehicle detection is one of the most valuable techniques for the intelligentization of modern transportation system. The famous widely used video-based vehicle detection technique is the background subtraction method. The key problem of this method is how to compute and update the background effectively and efficiently. A new background generation method was proposed. The method utilized the Gaussian distribution to model each point of the background image. To deal with the sudden change of the brightness, the mean brightness of the regions with no moving objects in them was used to make compensation. Results showed that the proposed method could generate and update the background quickly and efficiently.

In this paper, some novel methods for direction map calculating, greyscale fingerprint image filtering based on direction template and binary image filtering based on mathmatical morphology method are proposed. Experiments show that the suggested approach can greatly reduce the noises and pseudo-structures, and hence is helpful for the feature extraction and fingerprint identification.

本文介绍了一种基于视频处理的交通信息采集设备的设计方案.该方案利用摄像头采集交通视频,一方面通过DSP6203B进行分析处理得到各种实时交通信息,另一方面通过ADSP 2185进行图像的小波压缩.最后,交通信息和压缩图像通过嵌入式网络传输模块传送到交通控制中心.

We present our License Plate Recognition (LPR) System. Using vehicle position sensors and image acquisition card, it captures images of the vehicles automatically and sends the images to computer, then recognizes characters of the vehicle plate. Combined with the web technique, information of certain plates and the images of the vehicles can be browsed conveniently from remote sites. The system frame and workflow of our LPR System are introduced. Some problems and two of our proposed character recognition methods, i.e. method based on PCA-LSM for limited Chinese character recognition and method based on structural feature analysis for alphabetic and digital character recognition, are addressed in full details. The field applications under practical conditions show that the whole recognition rate of our system is over 97% in daytime and 95% at night.

A novel preprocessing and localization method has been proposed for container auto recognition system. The method combines both linear filters and nonlinear filters to reduce noises on an image. Selection method of the adaptive parameters of the filters is discussed thoroughly. Tests have been made based on the method. Character lines can be properly located with a ratio of above 98%.

The paper presents the improvement and enhancement to the existing video based traffic detecting system. The works are concentrated on the acquiring and update of background image, a new solution to segmentation of vehicles from the background and adding peripheral interfaces. The simulation and experimental results are effective.

分析了在实时图像处理中采用TMS320VC5402作处理器的优缺点,设计了一种基于C5402的低端图像处理平台,适用于分辨率较低的灰度图像处理.还介绍了C5402利用IO空间和I2C总线的接口实现方式.给出了系统实例及实验结果.

本文简单介绍了PTP(Picture Transfer Protocol)协议的背景和内容，并从协议的工作过程出发，在PC上实现其在USB(USB1.1)接口上的编程，进而在此基础上成功的完成了对数码相机的控制。

通用分组无线业务(General Packet Radio Service,GPRS)作为GSM网络新的承载服务,解决了传统GSM网中数据传递慢、计费方式单一的缺点.本文重点是对GPRS移动性管理和无线资源分配做了详细分析,并对GPRS计费原则做了进一步的研究.