Search

In recent years, the single-stage detectors have been developed rapidly; however, compared with the multi-stage detectors, their detection precision is still relatively low. Single-stage detectors and multi-stage detectors are analyzes and compared in detail in this paper, which reveals that single-stage detectors suffer from some problems, including feature loss and inaccurate feature extraction. Therefore, this paper proposes a novel detection model, dubbed Optimized Network (OptNet), to alleviate these deficiencies. OptNet consists of three modules: pyramid of attention features, feature alignment and consistency supervision (CS). The pyramid of attention features, based on feature pyramid networks (FPNs), introduces a novel branch named attention FPN (AtFPN), which aggregates the multi-layer features of the backbone network and optimizes the object features by using lightweight attention modules. AtFPN alleviates the loss of the feature pyramid information and the blocking of feature transmission between adjacent layers. Meanwhile, it provides global information for the model. The feature alignment module aligns the anchor box to the feature by using the object location information to guide the network to extract precise object features. Finally, CS accelerates network optimization and reduces semantic differences between the features on different layers. In the detection stage, OptNet optimizes the prediction of the model with the first detection result to improve the accuracy. Experiments on the MS COCO 2017 dataset demonstrate that OptNet yields significant improvement in the detection precision.

Data structure is one of the important courses of basic education for computer students. In this study, aiming at the disadvantages of traditional data structure teaching, we try to apply scaffolding theory to teaching to improve the whole teaching mode and effect to a great extent.

In unmanned collaborative cognition scenarios, traditional IoT systems are facing with data security and node privacy problems. However, blockchain together with data encryption technique enables the trustless network due to the tamper-proof, anonymity and unbreakable properties. Nevertheless, there are still some shortages when combining blockchain with IoT systems, for example the storage consumption. Generally speaking, validation nodes in blockchain network need to store a complete copy of the ledger so as to verify transactions and reach the whole network consensus. However, in unmanned collaborative cognition scenarios, many devices' storage capacity are limited. Therefore, in order to solve the storage problem, we propose LS4BUCC, a low overhead storage architecture regarding to the new features of unmanned collaborative cognition systems. It selectively stores data according to the value density. Based on three storage mechanisms and two supportive methods, LS4BUCC can reduce the data storage and maintain the system security and efficiency at the same time.

Palmprint has several advantages in authentication on mobile devices; however, some serious challenges severely restrict the development of palmprint verification in mobile environments, such as various locations and gestures of hand, complex backgrounds, diverse illuminations, limited hardware resources. In order to overcome the aforementioned problems, we designed and implemented a novel palmprint verification system with an assistance technique, namely dual-line-single-point (DLSP), on Android mobile devices. The environment setup and variable configuration on Android operating system were carried out according to the demand analysis. In software subsystem, the main interface and image capture interface were designed to meet the system requirements. All kinds of classes and threads were programmed in Java language. Finally the software and hardware were jointly debugged. The sufficient test results confirm the effectiveness and practicability of the developed palmprint verification system.

JointCloud is a new generation of cloud computing model which facilitates developers to customize cloud services. JCLedger is a blockchain based distributed ledger for JointCloud computing which can make cloud resources exchange more reliable and convenient, and it is the combination of JointCloud and BlockChain. One of the most important elements for creating JCLedger is the consensus algorithm. PoW (Proof of Work) is the consensus algorithm for Bitcoin, which is proved to be quite safe but needs much computing power. The original PoW is not suitable for JCLedger because the identities of participants are not equal in computing power, which may lead to accounting monopoly, and the throughput cannot satisfy the requirement of the massive and high-frequency transactions in JointCloud. In this paper, we propose a PoW based consensus algorithm called Proof of Participation and Fees (PoPF), which can save much computing power and handled transactions more efficiently for JCLedger. In our design, only the candidates have the opportunities for mining and the candidates are chosen according to the ranking which is determined by two factors: the times of the participant to be the accountant and the fees the participant has paid. The difficulty for candidates of solving the PoW hash puzzle is different (the higher ranking means easier for mining). The simulation experiment shows that the distribution of accountants is well-balanced, that is to say, the unequal computing power of participants in JointCloud is shielded, and all the users who have enough contribution in JCLedger will have the opportunities to be accountants.

Scene classification is more challenging than object classification due to higher ambiguity in scene labels. In this work, we propose to use the filter weights at the last stage of a CNN model trained by the Places dataset, which is also known as the "scene anchor vector (SAV)", to explain the source of confusions. An SAV points to a cluster of images. If two anchor vectors have a smaller angle, we see overlapping image clusters, leading to a set of confusing classes. To overcome it, we propose to merge images associated with confusing anchor vectors into a confusion set and split the set in an unsupervised fashion to create multiple subsets. It is called the "automatic subset clustering (ASC)" process. Each of these subsets contains scene images of strong visual similarity. After the ASC process, we train a random forest (RF) classifier for each confusion subset to allow better scene classification. The ASC/RF scheme can be added on top of any existing scene-classification CNN as a post-processing module with little extra training effort. It is shown by extensive experimental results that, for a given baseline CNN, the ASC/RF scheme can offer a significant performance gain.

We target at solving the problem of automatic image segmentation. Although 1D contour and 2D surface cues have been widely utilized in existing work, 3D depth information of an image, a necessary cue according to human visual perception, is however overlooked in automatic image segmentation. In this paper, we study how to fully utilize 1D contour, 2D surface, and 3D depth cues for image segmentation. First, three elementary segmentation modules are developed for these cues respectively. The proposed 3D depth cue is able to segment different textured regions even with similar color, and also merge similar textured areas, which cannot be achieved using state-of-the-art approaches. Then, a content-dependent spectral (CDS) graph is proposed for layered affinity models to produce the final segmentation. CDS is designed to build a more reliable relationship between neighboring surface nodes based on the three elementary cues in the spectral graph. Extensive experiments not only show the superior performance of the proposed algorithm over state-of-the-art approaches, but also verify the necessities of these three cues in image segmentation.

Nowadays, cloud computing is hard to effectively sustain the implementation of the commercial model of Internet Service globalization. There is a growing trend to build an environment of cloud service, with the capacity to serve anytime and anywhere, by mutual cooperation between cloud service providers around the world. However, this tendency will raise a key issue which is how to provide a benign environment, that allows self-collaboration and fair competition, for different cloud service providers with diverse stakeholder. Inspired by the concept and structure of Service-Oriented Architecture (SOA) service, this paper proposes a structure named JointCloud Corporation Environment (JCCE), which offers a mutual benefit and win-win JointCloud environment for global cloud service providers. JCCE contains three core services, which are Distributed Cloud Transaction, Distributed Cloud Community and Distributed Cloud Supervision. Also, facing with different cloud service participants, JCCE offers three main service modes for their consumption, supply and coordination. This study plays a significant role in supporting the sharing and self-collaboration of multiple cloud entities, and promoting the development of cloud service market healthy and orderly.

The Root Mean Square current (Irms) has been explored on both N doped un-silicided poly and silicided poly at 40nm technology node. It is found that poly resistance-current (RI) curves show an initially high resistance due to Schottky Rectifying contact. However, poly Irms can be estimated similar to metal Irms. Parameters for poly Irms calculation is provided on the basis of 5°C Joule heating.

Radar usually employs several signal processing methods successively for detection. Each processing step influences the radar detection performance. Thus, analysis of the radar detection performance considering the complete signal processing flow provides significant reference for radar detector design. In this paper, with the assumption that matched filtering, square-law detection, non-coherent integration, cell averaging constant false alarm rate detection and double-threshold detection are employed sequentially for radar detection, radar detection performance in K distributed sea clutter plus Rayleigh distributed thermal noise is analyzed. Mathematical expressions of radar false alarm probability and detection probability are derived. Radar detection performances for four types of Swerling fluctuating targets in various circumstances are presented by simulation. And comprehensive effect of K distributed sea clutter plus Rayleigh distributed thermal noise on radar detection performance is analyzed.

The contour-guided color palette (CCP) is proposed for robust image segmentation. It efficiently integrates contour and color cues of an image. To find representative colors of an image, color samples along long contours between regions, similar in spirit to machine learning methodology that focus on samples near decision boundaries, are collected followed by the mean-shift (MS) algorithm in the sampled color space to achieve an image-dependent color palette. This color palette provides a preliminary segmentation in the spatial domain, which is further fine-tuned by post-processing techniques such as leakage avoidance, fake boundary removal, and small region mergence. Segmentation performances of CCP and MS are compared and analyzed. While CCP offers an acceptable standalone segmentation result, it can be further integrated into the framework of layered spectral segmentation to produce a more robust segmentation. The superior performance of CCP-based segmentation algorithm is demonstrated by experiments on the Berkeley Segmentation Dataset.

Nowadays, while we are enjoying the convenience brought by such a huge repository of online web information, we may come across difficulties in finding the web pages we want related to particular information we are searching for. Hence, it is essential to classify web documents to facilitate the search and retrieval of pages. Existing algorithms work well with a small quantity of web pages, whereas, they become slow and even non-effective while dealing with a large scale of web pages. Recently, some of these algorithms were adapted to distributed platforms which boosted their classification speeds effectively. However, due to high dimensions of web page features, the parallel classifiers were still trained with limited capacity training sets. In addition, these methods didn't improve the classification itself, merely boosted by high computing performance of distributed platforms. So oriented to large-scale web page classification, we propose to integrate anchor graph hashing with K-Nearest Neighbour(KNN) classifier to reduce the pages' original feature dimensions. The hash value of each page is used for training and classification instead of the original vectors. Experimental comparison with the original KNN on a large dataset demonstrates the efficacy of our proposed method.

Establishing correspondences between minutiae sets is a fundamental issue in fingerprint recognition. This paper proposes a graph matching strategy. It addresses this problem using two extended clique models: the local clique model (LCM) and the global quasi-clique model (GQCM). First, minutia matching is formulated as maximal clique or quasi-clique detection. Similarities and compatibilities between minutia pairs are integrated and encoded into the correspondence graph. Correct matched minutia pairs then correspond to those strongly connected subgraphs, like the maximal clique or the maximal quasi-clique. Second, a two-level matching algorithm is proposed. In the local level, a novel similarity calculation of minutia topologic structures is proposed. It calculates similarities of minutia structures based on LCM and can be solved by maximal clique detection. In the global level, a novel similarity calculation of entire minutia sets is proposed. It calculates entire fingerprint similarity based on GQCM and can be solved by spectral correspondence techniques. Proposed method has stronger description ability and better robustness to noise and nonlinearity. Experiments conducted on FVC 2002 and 2004 databases demonstrate the effectiveness and the efficiency.

In this paper, a new non-linear distortion robust fingerprint feature named Minutiae Phase Difference Feature (MPDF) is proposed, and used to measure the similarity of fingerprint local structures. Then a novel minutiae expanding method based on Moving Least Squares (MLS) transformation is introduced to regain the lost matched minutiae pairs after obtaining a few valid pairs. Finally a score evaluation strategy based on convex hull of transformed minutiae is adopted, which is more reasonable for partial overlapping fingerprint. Experiments of proposed algorithm conducted on FVC2004 database, comparing with other matching methods, show that the MPDF feature and proposed algorithm have promising performances.

Faster than at-speed testing provides an effective way to detect small delay defects (SDDs). It requires test patterns to be delicately classified into groups according to the delay of sensitized paths. Each group of patterns is applied at certain frequency. In this paper, we propose to generate tests for faster than at-speed testing using path delay fault (PDF) model and single path sensitization criterion. An effective path selection and grouping method is introduced, which could quickly and accurately identify paths whose delay falls into a given delay span. Several techniques are used to improve the efficiency of the testable path selection procedure. Experimental results on ISCAS’89 benchmark circuits show that the proposed method could achieve high transition fault coverage and high test quality of SDDs with low CPU time.

Do students really welcome failure, the mother of success? This paper reports the implementation of a trial-and-failure learning strategy, in both entry and senior level computer science classes. The idea is simple: given a sophisticated course project, let students try project submissions as many times as they want before the project deadline. For each submission, a thorough inspection is performed by an automated grading system called APOGEE. A student project has to accomplish not only the functional requirements but also many desired quality attributes such as robustness and security. In 2009, the trial-and-failure strategy has been adopted by four universities in five class sections. We report some interesting observations from student survey results, e.g., one can find out if factors like students' positive experiences of programming, choice of programming language, years of working experience, and instructors are predictive variables for positive attitudes toward the trial-and-failure learning experience as a whole.

Sophisticated course projects in senior computer science classes have raised great challenges to educators. Students now have stronger needs for instant and individualized guidance when exposed to the complexity of software development. In this paper, we describe the application of Bayesian Network, a probabilistic graphic model, to automated project failure analysis. Students can make multiple project submissions to an automated grader before the project deadline. Each submission will be graded immediately, by running a collection of test cases prepared by instructors in advance. Then failure behaviors of test cases are piped to a Bayesian Network (BN) inference engine. The BN engine generates a report on the most probable failure causes, and helps students in removing software bugs. The analysis of the BN engine can be improved by parameter learning. A case study is presented to illustrate the effectiveness of the approach.

While major research universities enjoy a large pool of cutting-edge research projects for technology transfer, small liberal art institutions have to adopt a different model for promoting entrepreneurship due to resource limits. We propose a sandbox model for teaching entrepreneurship in a software engineering class. Student activities are closely monitored and coordinated by the instructor. Introduction of entrepreneurial thinking is tightly integrated with the milestone projects in the class. Students are allowed to fail, but they can always resume from the next milestone. We admit that the sandbox approach is highly debatable. Some faculty members believe that a fixed project topic may limit the creativity of students. On the other hand, we argue for the effectiveness of the approach. A good coverage of entrepreneurship topics can be achieved without requiring extra room in the curriculum. We report our preliminary experiences of implementing the sandbox model at Hofstra University.

Creativity and innovativeness are among the most essential attributes of engineering graduates and also of successful entrepreneurs. Entrepreneurship, or the process of starting a new venture, is one of the main roads to new technological innovations. This paper presents two novel models of entrepreneurship education integrated in computer science and computer engineering curricula and geared towards computing students with entrepreneurial intentions. To expose all computing students to entrepreneurial ideas and to spark their entrepreneurial spirit, we also developed several entrepreneurship add-on modules for existing CS and CE disciplines. All these programs have been developed and implemented at Hofstra University, with modules implemented also at Qatar University. Preliminary evaluation results are presented and discussed.

This thesis first analyses the causes for producing swing angle, and then proposes to reduce this angle by using swing pilot platform. The mechanical structure of the platform is introduced, and a new Magnetic Heading Measurement System based on the magnetoresistive sensor and the MEMS accelerometer is designed. According to the features of main components, the error compensation algorithm is used to make sure the accuracy of direction angle. The test results indicate that the swing pilot platform can keep the direction angle stable even under crosswinds condition. The accuracy of aerial photographic swing angle is within three degrees.

This paper presents the design of BAUT, a tutoring system that explores statistical approach for providing instant project failure analysis. Driven by a Bayesian Network (BN) inference engine, BAUT analyzes the test cases failed by a student project submission and provides instant diagnosis that guides students in identifying and removing software bugs. Using a parameter learning process, BAUT is able to improve the quality of its analysis. The initial case study with the prototype demonstrates the potential of the system.

DLT (Direct Linear Transformation) is one of the most popular methods of data simplification for camera calibration. An improved DLT method is proposed, the projective pseudo-inverse matrix is used to describe the relations between 2D coordinates and 3D coordinates, the improved DLT method can calculate the projective pseudo-inverse matrix based on only 4 known pixel points. The experiment results show that the proposed method is more efficient and more lightly limited by the prerequisites than traditional one, so it can be widely used in practice.

Analyzing interactions among peers that interact via messages is a crucial problem due to increasingly distributed nature of current software systems, especially the ones built using the service oriented computing paradigm. In service oriented computing, interactions among peers participating to a composite service involve message exchanges across organizational boundaries in a distributed computing environment. In order to build such systems in a reliable manner, it is necessary to develop techniques for analysis and verification of interactions among services. Collaboration diagrams provide a convenient visual model for modeling service interactions. In this paper, we present a tool that 1) checks the realizability of interactions specified by the given collaboration diagram, 2) verifies the LTL properties of the interactions specified by the given collaboration diagram by automatically converting it to a state machine model, and 3) synthesizes peer state machines that realize the set of interactions specified by the given collaboration diagram.

Straight line is an important descriptor in image analysis. The drawbacks of traditional straight line parameters extracting methods are anal sized, and a novel multi-line parameters extracting method is proposed. Firstly, the image is divided into blocks, and RHT (random Hough transform) is used to detect straight line parameters in each sub-image. Then, global parameters are extracted based on the relationship of parameters between the sub-images and the global image. Comparisons with traditional scheme are presented and the results demonstrating the performance of the proposed method.

Feature matching method is the key technology of image retrieval. A matching method based on SVD is proposed in this paper. Firstly, the feature matrix is constructed and the eigenvalue matrix is calculated based on SVD transformation?Then, the eigenvalue matrix is mapped into vector space?and similarity is measured by comparing the angle between vectors. Finally, the matching method is applied to image retrieval. Experimental results show that the proposed method is feasible and superior to the traditional distance matching method.

Video shot boundary detection is an important early stage of video analysis. In this paper, a new method for shot boundary detection using local color features (LCF) is presented. Firstly, interest points are detected for each frame. Then, local images are obtained based on the interest points. Finally, shot boundaries are detected based on the histogram difference of local images. LCF can overcome the deficiency of traditional histogram-based method that different contents frames have similar histograms, and can distinguish between gradual changes and camera motions effectively. Experimental results successfully validate the new method and show that it can effectively detect both abrupt transitions and gradual transitions.

This paper presents an automated testing and tutoring system called AWAT for teaching dynamic Web applications. Based on Watir, a retrogression testing tool,the system simulates actions of human being and automatically interprets and executes tests cases specified in Excel spreadsheets. AWAT is able to demo to students,step by step, how a submitted project fails a functional or non-functional requirement. Further textual hint (prepared by faculty) can be shown to students. AWAT is unique in its ability to interact with Web GUI applications developed in any languages and deployed on any platforms. AWAT provides instant feedback to students,which enriches their learning experiences significantly.

In order to overcome the disadvantages of traditional centralized video monitoring system or distributed systems based on wired network, we propose a framework for distributed video surveillance in heterogeneous environment. Video flows are compressed with the scalable video encoding standard MPEG-4 and transmitted over Internet or wireless network. Video surveillance can be performed wherever there is Internet or mobile telephone signal. The feasibility of this framework has been demonstrated with a prototype implementation. The system is cheaper and easier to achieve with simple equipments, so it can be widely used in practice.

This paper presents a course model for teaching component-based software development (CBSD) in the software engineering undergraduate curriculum to deepen and broaden student comprehension of component software. All hands-on lab practices are designed with open-source software tools. The course endows students the precious experiences of making design decisions via comparative study, risk analysis, and synthesis of a vast number of open-source components.