基于GFU和分层LSTM的组群行为识别研究方法

提出一种以"关键人物"为核心，使用门控融合单元（GFU，Gated Fusion Unit）进行特征融合的组群行为识别框架，旨在解决两个问题：①组群行为信息冗余，重点关注关键人物行为特征，忽略无关人员对组群行为的影响；②组群内部交互行为复杂，使用GFU有效融合以关键人物为核心的交互特征，再通过LSTM时序建模成为表征能力更强的组群特征.最终，通过softmax分类器进行组群行为类别分类.该算法在排球数据集上取得了86.7%的平均识别率.     

基于高斯混合隐马尔科夫模型与人工神经网络的紧急换道行为预测方法

为了有效降低因驾驶员紧急换道行为而诱发的交通事故，提高道路交通事故链阻断效率，提出一种基于高斯混合隐马尔科夫模型（GMM-HMM）和人工神经网络（ANN）的紧急换道行为预测方法。首先利用GMM-HMM对车辆行驶状态以及驾驶行为连续观察序列进行换道意图辨识，采用ANN预测下一时段的驾驶行为，再预测换道过程中的横向加速度变化率，从而判断紧急换道的危险程度。驾驶员在环仿真实验及实车实验结果表明，该方法预测避险成功率达92.83%，实验避险成功率达90.32%。该方法能有效地对紧急换道行为进行提前警告与干预。     

Engineering of Lipid Nanoparticles by the Multifunctionalization of the Surface with Amino Acid Derivatives for the Neutralization of a Target Toxic Peptide

Protein affinity reagents (e.g., antibodies) are often used for basic research, diagnostics, separations, and disease therapy. Although a lot of “synthetic” protein affinity reagents have been developed as a cost-effective alternative to antibodies, their low biocompatibility is a considerable problem for clinical application. Lipid nanoparticles (LNP) represent a highly biocompatible drug delivery agent. However, little has been reported that LNP itself works as a protein affinity reagent in living animals. Here, LNP is engineered for binding to and neutralizing a target toxic peptide in living animals by multifunctionalization with amino acid derivatives. Multifunctionalized LNP (MF-LNP) is prepared using amino acid derivative-conjugated lipids. Optimized MF-LNP exhibits nanomolar affinity to the target toxic peptide and inhibits toxic peptide-dependent hemolysis and cytotoxicity. In addition, MF-LNP captures and neutralizes the toxic peptide after intravenous injection in the bloodstream; in addition, MF-LNP does not release the toxic peptide in the accumulated organ. These results reveal the potential of using LNP as a highly biocompatible protein affinity reagent such as an antidote.     

基于卷积神经网络的表情识别

在传统静态表情识别研究基础上,提出一种简单的人脸裁剪方法,再用浅层卷积神经网络进一步提取特征并进行表情识别。以CK+和JAFFE为实验数据集,进行预处理效果对比实验、数据增强实验、单种表情识别实验和跨数据集六分类实验。结果表明,针对数据量较少的情况,提出的表情识别方法效果明显且鲁棒性更优。     

Study of traffic characteristics based on internet web real-time traffic conditions by image identification technology

With the rapid development of Internet, it is increasingly convenient to obtain real-time traffic condition information, which has greatly stimulated the improvement of urban traffic guidance. Traffic conditions are generally divided into four grades in the existing network platform, which are expressed in different colours. The understanding of traffic condition is still at the level of abstract senses. Therefore, it is difficult to grasp the characteristics of urban traffic. To this end, a new idea is proposed in this paper, and the new idea is to study the urban traffic characteristics based on real-time traffic condition information extraction with image identification technology. With this method, we can not only quantify the abstract traffic condition information, but also solve the loss of traffic condition information. In addition, an instance is analysed in this paper, it shows that it can provide references for urban traffic organization management very well.     

一种广义不可分支持向量机算法

针对标准的C-SVM(C-support vector machine)算法在处理很多实际分类问题时，对识别错误代价损失差异很大的极端情况表现出的局限性，提出一种通用的 广义支持向量机算法。根据识别错误后所付出的代价，可以把最优分类面向代价损失低的一方进行推移，留给代价损失高的一方更大的空间，提高其识别率，从而减小识别错误后带来的代价损失。该方法进一步提高了标准C SVM的适用性以及样本的正确识别率，将新算法应用到高分辨雷达距离像的识别中，实验证明,广义C-SVM能取得比传统C-SVM更好的识别效果。     

Rapidly finding CAD features using database optimization

Automatic feature recognition aids downstream processes such as engineering analysis and manufacturing planning. Not all features can be defined in advance; a declarative approach allows engineers to specify new features without having to design algorithms to find them. Naive translation of declarations leads to executable algorithms with high time complexity. Database queries are also expressed declaratively; there is a large literature on optimizing query plans for efficient execution of database queries. Our earlier work investigated applying such technology to feature recognition, using a testbed interfacing a database system (SQLite) to a CAD modeler (CADfix). Feature declarations were translated into SQL queries which are then executed.The current paper extends this approach, using the PostgreSQL database, and provides several new insights: (i) query optimization works quite differently in these two databases, (ii) with care, an approach to query translation can be devised that works well for both databases, and (iii) when finding various simple common features, linear time performance can be achieved with respect to model size, with acceptable times for real industrial models. Further results also show how (i) lazy evaluation can be used to reduce the work performed by the CAD modeler, and (ii) estimating the time taken to compute various geometric operations can further improve the query plan. Experimental results are presented to validate our main conclusions.     

Human action recognition in H.264/AVC compressed domain using meta-cognitive radial basis function network

In this paper, we propose a H.264/AVC compressed domain human action recognition system with projection based metacognitive learning classifier (PBL-McRBFN). The features are extracted from the quantization parameters and the motion vectors of the compressed video stream for a time window and used as input to the classifier. Since compressed domain analysis is done with noisy, sparse compression parameters, it is a huge challenge to achieve performance comparable to pixel domain analysis. On the positive side, compressed domain allows rapid analysis of videos compared to pixel level analysis. The classification results are analyzed for different values of Group of Pictures (GOP) parameter, time window including full videos. The functional relationship between the features and action labels are established using PBL-McRBFN with a cognitive and meta-cognitive component. The cognitive component is a radial basis function, while the meta-cognitive component employs self-regulation to achieve better performance in subject independent action recognition task. The proposed approach is faster and shows comparable performance with respect to the state-of-the-art pixel domain counterparts. It employs partial decoding, which rules out the complexity of full decoding, and minimizes computational load and memory usage. This results in reduced hardware utilization and increased speed of classification. The results are compared with two benchmark datasets and show more than 90% accuracy using the PBL-McRBFN. The performance for various GOP parameters and group of frames are obtained with twenty random trials and compared with other well-known classifiers in machine learning literature.