基于深度学习的视频跟踪研究进展综述

近年来深度学习迅猛发展，颠覆了语音识别、图像分类、文本理解等领域的算法设计思路。深度学习因其具备强大的特征提取能力，在图像识别领域的成绩尤为突出。然而深度学习与视频监控领域的结合并不多，由于深度模型具有多层网络结构，算法复杂度大，训练和更新模型时比较耗时，很难满足实时性要求。回顾了深度学习的发展史，介绍了最近10年来国内外深度学习主要模型，论述了基于深度学习的目标跟踪算法，指出了各算法的优缺点，最后对当前该领域存在的问题和发展前景进行了总结和展望。     

WavNet — Visual saliency detection using Discrete Wavelet Convolutional Neural Network

In the recent advancements in image and video analysis, the detection of salient regions in the image becomes the initial step. This plays a crucial role in deciding the performance of such algorithms. In this work, a Multi-Resolution Feature Extraction (MRFE) technique that makes use of Discrete Wavelet Convolutional Neural Network (DWCNN) for generating features is employed. An Enhanced Feature Extraction (EFE) module extracts additional features from the high level features of the DWCNN, which are used to frame both channel as well as spatial attention models for yielding contextual attention maps. A new hybrid loss function is also proposed, which is a combination of Balanced Cross Entropy (BCE) loss and Edge based Structural Similarity (ESSIM) loss that effectively identifies and segments the salient regions with clear boundaries. The method is tested exhaustively with five different benchmark datasets and is proved superior to the existing state-of-the-art methods with a minimum Mean Absolute error (MAE) of 0.03 and F-measure of 0.956.     

基于VMD和改进TCN的短期光伏发电功率预测

光伏发电功率存在波动性,且光伏出力易受各种气象特征影响,传统TCN网络容易过度强化空间特性而弱化个体特性。针对上述问题,文中提出一种基于VMD和改进TCN的短期光伏发电功率预测模型。通过VMD将原始光伏发电功率时间序列分解为若干不同频率的模态分量,将各个模态分量以及相对应的气象数据输入至改进TCN网络进行建模学习。利用中心频率法确定VMD的最优分解模态分解个数。在传统TCN预测模型的基础上,使用DropBlock正则化取代Dropout正则化以达到抑制卷积层中信息协同的效果,并引入注意力机制自主挖掘并突出关键气象输入特征的影响,量化各气象因素对光伏发电的影响,从而提高预测精度。以江苏省某光伏电站真实数据为例进行仿真实验,结果表明所提预测方法的RMSE为0.62 MW,MAPE为2.03%。     

Hybridizing adaptive large neighborhood search with kernel search: a new solution approach for the nurse routing problem with incompatible services and minimum demand

The average age of the population has grown steadily in recent decades along with the number of people suffering from chronic diseases and asking for treatments. Hospital care is expensive and often unsafe, especially for older individuals. This is particularly true during pandemics as the recent SARS-CoV-2. Hospitalization at home has become a valuable alternative to face efficiently a huge increase in treatment requests while guaranteeing a high quality of service and lower risk to fragile patients. This new model of care requires the redefinition of health services organization and the optimization of scarce resources (e.g., available nurses). In this paper, we study a Nurse Routing Problem that tries to find a good balance between hospital costs reduction and the well-being of patients, also considering realistic operational restrictions like maximum working times for the nurses and possible incompatibilities between services jointly provided to the same patient. We first propose a Mixed Integer Linear Programming formulation for the problem and use some valid inequalities to strengthen it. A simple branch-and-cut algorithm is proposed and validated to derive ground benchmarks. In addition, to efficiently solve the problem, we develop an Adaptive Large Neighborhood Search hybridized with a Kernel Search and validate its performance over a large set of different realistic working scenarios. Computational tests show how our matheuristic approach manages to find good solutions in a reasonable amount of time even in the most difficult settings. Finally, some interesting managerial insights are discussed through an economic analysis of the operating context.     

Object detection based on the multiscale location Enhancement network

For the target detection task,there are two problems in the one-stage network structure of the deep neural network model.First,whether the design of the anchor box hyperparameter is suitable or not will affect the training results of the whole network;second,a large down sampling factor will affect the positioning ability of the target.To solve these problems,this paper proposes a multi-location enhancement network.The structure of the one-stage network model is redesigned,and a better scheme for selecting the super parameters of the anchor frame is proposed.So the efficiency of the first stage network is ensured and the positioning accuracy is better than the previous one.A large number of experiments show that the multi-location enhancement network can achieve a higher positioning accuracy while ensuring real-time performance.The average accuracy of 82.5 is achieved on the public dataset (Pascal VOC 2007).     

卷积神经网络用于关节角度识别与姿势评估

为了快速准确地输出各种工作姿势风险评估结果，提出采用Kinect v2与卷积神经网络识别人体各关节角度，并输出标准姿势风险的评估得分。首先使用亚像素角点提取的棋盘标定算法标定Kinect两个摄像头，其次使用改进后的双边滤波对深度图像去噪，使用卷积神经网络识别人体关节二维位置，结合深度信息获取实际三维坐标并计算人体关节角度，最后输出姿势风险评估得分。通过两种实验分别验证了提出的Kinect角度识别与姿势评估的准确性，表明该方法关节角度识别与姿势风险评估的准确率均较高，是一种低成本、高可靠性的姿势评价方法，具有一定的科学意义和工程应用价值。     

一种基于雷达图表示的数值型数据的CNN分类方法

卷积神经网络（convolutional neural networks， CNN）是一种广泛用于分析视觉图像的分类方法.由于数值数据存在着非线性、耦合性等复杂的空间关系，因此基于CNN的数值型数据的研究较少.本文的目的是找到一种可行的方法，将CNN的应用领域扩展到数值数据.于是提出了一种基于雷达图表示的数值型数据的CNN分类方法（Radar-CNN）.该算法首先将数值数据表示成雷达图形式，然后将其输入CNN中构建分类模型.为了进一步研究特征尺度和序列对性能的影响，提出了两种改进算法Rank Radar-CNN和SFS Radar-CNN.为了验证所提算法的有效性，引入TE化工过程数据集进行实验测试并比较，实验结果表明Radar-CNN及其改进算法具有优异的性能.     

基于深度残差学习的自动驾驶道路场景理解

随着道路场景理解技术的快速发展，自主驾驶领域取得了长足的进步。在相关任务中，包括道路分割、分类和车辆检测的实时性和准确性是安全性的一个关键问题。为此，提出了一个具有编/解码器网络结构的基于深度残差学习的方法。一方面，编码器网络结构使用不同层次的残差网络来提取高维中的抽象特征，这些特征在接下来的三个任务中共享使用；另一方面，解码器网络结构采用一种子任务的并行计算机制，即道路分割、车辆检测和道路分类任务同时执行。此外，全卷积神经网络用于对提取的图像特征进行上采样以解决道路分割问题。最终，实验结果表明在保证高精度的前提下处理帧率可达到15 fps以上。     

多共振分量融合CNN的行星齿轮箱故障诊断

针对行星齿轮箱中各部件所激起的振动成分混叠、早期故障特征经常被较强的各级齿轮谐波成分以及环境噪声所湮没的问题，提出一种多共振分量融合卷积神经网络（multi-resonance component fusion based convolutional neural network，简称MRCF-CNN）的行星齿轮箱故障诊断方法。首先，对振动信号进行共振稀疏分解，得到包含齿轮谐波成分的高共振分量和可能包含轴承故障冲击成分的低共振分量；其次，构建多共振分量融合卷积神经网络，将得到的高、低共振分量和原始振动信号进行自适应的特征级融合，通过有监督的方式训练模型并进行行星齿轮箱故障诊断。对行星齿轮箱实验数据的分析结果表明，该方法能够有效分类行星齿轮箱中滚动轴承和齿轮的故障，成功对行星齿轮箱故障进行诊断，同时能够进一步增强卷积神经网络对振动信号所蕴含的故障信息的辨识能力。