基于深度学习的无人车夜视图像语义分割

为了增强无人车对夜视图像的场景理解,在夜间模式下更快更精确地探测和识别周围环境,将深度学习应用于夜视图像的场景语义分割,提出了一种基于卷积-反卷积神经网络的无人车夜视图像语义分割方法。在传统的卷积神经网络中加入反卷积网络,构建卷积-反卷积神经网络,无需手工选取特征。通过像素到像素的学习和训练,得到图像语义分割模型,可直接用该模型预测夜视图像中每个像素所属的场景语义类别,实现无人车夜间行驶时的环境感知。实验结果表明,该方法具有较好的准确性和实时性,平均IU达到68.47。     

One- and Two-Phase Software Requirement Classification Using Ensemble Deep Learning

Recently, deep learning (DL) has been utilized successfully in different fields, achieving remarkable results. Thus, there is a noticeable focus on DL approaches to automate software engineering (SE) tasks such as maintenance, requirement extraction, and classification. An advanced utilization of DL is the ensemble approach, which aims to reduce error rates and learning time and improve performance. In this research, three ensemble approaches were applied: accuracy as a weight ensemble, mean ensemble, and accuracy per class as a weight ensemble with a combination of four different DL models—long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), a gated recurrent unit (GRU), and a convolutional neural network (CNN)—in order to classify the software requirement (SR) specification, the binary classification of SRs into functional requirement (FRs) or non-functional requirements (NFRs), and the multi-label classification of both FRs and NFRs into further experimental classes. The models were trained and tested on the PROMISE dataset. A one-phase classification system was developed to classify SRs directly into one of the 17 multi-classes of FRs and NFRs. In addition, a two-phase classification system was developed to classify SRs first into FRs or NFRs and to pass the output to the second phase of multi-class classification to 17 classes. The experimental results demonstrated that the proposed classification systems can lead to a competitive classification performance compared to the state-of-the-art methods. The two-phase classification system proved its robustness against the one-phase classification system, as it obtained a 95.7% accuracy in the binary classification phase and a 93.4% accuracy in the second phase of NFR and FR multi-class classification.     

Convexity preserving level set for left ventricle segmentation

In clinical applications of cardiac left ventricle (LV) segmentation, the segmented LV is desired to include the cavity, trabeculae, and papillary muscles, which form a convex shape. However, the intensities of trabeculae and papillary muscles are similar to myocardium. Consequently, segmentation algorithms may easily misclassify trabeculae and papillary muscles as myocardium. In this paper, we propose a level set method with a convexity preserving mechanism to ensure the convexity of the segmented LV. In the proposed level set method, the curvature of the level set contours is used to control their convexity, such that the level set contour is finally deformed as a convex shape. The experimental results and the comparison with other level set methods show the advantage of our method in terms of segmentation accuracy. Compared with the state-of-the-art methods using deep-learning, our method is able to achieve comparable segmentation accuracy without the need for training, while the deep-learning based method requires a large set of training data and high-quality manual segmentation. Therefore, our method can be conveniently used in situation where training data and their manual segmentation are not available.     

使用TensorRT进行深度学习推理

TensorRT是一个高性能的深度学习推理平台。它包括一个深度学习推理优化器和运行时为深度学习推理应用程序提供低延迟和高吞吐量。给出了一个使用TensorRT快速构建计算管道的例子,实现通过TensorRT执行智能视频分析的典型应用。该示例演示了使用片上解码器进行解码、使用片上标量进行视频缩放和GPU计算的4个并发视频流。为了演示的简单性，只有一个通道使用NVIDIA TensorRT执行对象标识，并在标识的对象周围生成包围框。该示例还使用视频转换器函数进行各种格式转换，使用EGLImage来演示缓冲区共享和图像显示。最后采用GPU卡V100对ResNet网络进行TensorRT加速性能的实际测试,结果表明TensorRT能够使吞吐量提升大约15倍。     

Atlas-based automatic mouse brain image segmentation revisited: model complexity vs. image registration

Although many atlas-based segmentation methods have been developed and validated for the human brain, limited work has been done for the mouse brain. This paper investigated roles of image registration and segmentation model complexity in the mouse brain segmentation. We employed four segmentation models [single atlas, multiatlas, simultaneous truth and performance level estimation (STAPLE) and Markov random field (MRF) via four different image registration algorithms (affine, B-spline free-form deformation (FFD), Demons and large deformation diffeomorphic metric mapping (LDDMM)] for delineating 19 structures from in vivo magnetic resonance microscopy images. We validated their accuracies against manual segmentation. Our results revealed that LDDMM outperformed Demons, FFD and affine in any of the segmentation models. Under the same registration, increasing segmentation model complexity from single atlas to multiatlas, STAPLE or MRF significantly improved the segmentation accuracy. Interestingly, the multiatlas-based segmentation using nonlinear registrations (FFD, Demons and LDDMM) had similar performance to their STAPLE counterparts, while they both outperformed their MRF counterparts. Furthermore, when the single-atlas affine segmentation was used as reference, the improvement due to nonlinear registrations (FFD, Demons and LDDMM) in the single-atlas segmentation model was greater than that due to increasing model complexity (multiatlas, STAPLE and MRF affine segmentation). Hence, we concluded that image registration plays a more crucial role in the atlas-based automatic mouse brain segmentation as compared to model complexity. Multiple atlases with LDDMM can best improve the segmentation accuracy in the mouse brain among all segmentation models tested in this study.     

基于深度学习激光熔覆层树枝晶的形貌识别

在增材制造技术中,树枝晶的表征对于分析激光熔覆层的机械性能至关重要,但目前树枝晶的标记主要由人工完成,耗时长且容易引入人为误差,而深度学习可提高目标识别准确度。基于U-net网络提出了适于识别分割树枝晶的BNC-Unet网络,将串行注意力机制和Batch Normalization层有效地部署在上采样和下采样区域,调整图像特征的权重信息。选取交并比作为分割结果的评价指标,并对比了原Unet以及不同的改进方法在该指标下的结果。在测试集中,BNCUnet网络分割树枝晶准确率指标为84.2%,比原U-net网络结果提升了8.97%。该指标表明网络能准确地从激光熔覆层金相图中识别出树枝晶形貌,且识别树枝晶的准确率显著提高,便于在激光熔覆试验后评估熔覆层性能。     

基于级联网络的年龄性别识别研究

提出了一种基于深度学习技术的年龄和性别识别方法,建立了人脸识别硬件与软件系统。运用经典的反向传播算法确定预测函数的权重矩阵和偏差。针对单个识别网络准确率不高的问题,采用多个神经网络级联的方式,对输入的目标特征进行多次判定。通过设计一种投票竞争算法,让级联网络的识别结果自动进行竞争,获胜者作为最终的预测结果。预测结果与实验数据对比表明采用级联网络可有效提高对年龄性别的识别准确率,级联后的识别准确率分别达到了88%和82.61%。     

Development of a body motion interactive system with a weight voting mechanism and computer vision technology

This study develops a body motion interactive system with computer vision technology. This application combines interactive games, art performing, and exercise training system. Multiple image processing and computer vision technologies are used in this study. The system can calculate the characteristics of an object color, and then perform color segmentation. When there is a wrong action judgment, the system will avoid the error with a weight voting mechanism, which can set the condition score and weight value for the action judgment, and choose the best action judgment from the weight voting mechanism. Finally, this study estimated the reliability of the system in order to make improvements.The results showed that, this method has good effect on accuracy and stability during operations of the human–machine interface of the sports training system.     

EBCCD的增益及信噪比研究

以背照明减薄CCD为基础的微光成像系统 ,可以得到视频信号速率的输出。这种减薄的电子轰击CCD探测系统 (EBCCD)在微光下的性能大大超过了像增强器耦合的CCD探测系统 (ICCD)。EBCCD能在微光条件下产生高对比度、高分辨力的视频信号。可以预料 ,它将在工业应用、军事应用及科学研究工作中取代通常使用的ICCD。详细讨论了影响EBCCD的增益及信噪比的各种因素。结果表明 ,若适当提高阴极和CCD之间的电压 ,减小CCD基片的厚度 ,可以进一步减小EBCCD的噪声 ,从而使信噪比得到更大的提高     

Deep Image Steganography Using Transformer and Recursive Permutation

Image steganography, which usually hides a small image (hidden image or secret image) in a large image (carrier) so that the crackers cannot feel the existence of the hidden image in the carrier, has become a hot topic in the community of image security. Recent deep-learning techniques have promoted image steganography to a new stage. To improve the performance of steganography, this paper proposes a novel scheme that uses the Transformer for feature extraction in steganography. In addition, an image encryption algorithm using recursive permutation is proposed to further enhance the security of secret images. We conduct extensive experiments to demonstrate the effectiveness of the proposed scheme. We reveal that the Transformer is superior to the compared state-of-the-art deep-learning models in feature extraction for steganography. In addition, the proposed image encryption algorithm has good attributes for image security, which further enhances the performance of the proposed scheme of steganography.     

基于U-net和可见光谱图像的黄瓜褐斑病分割

褐斑病是黄瓜主要真菌性病害之一,适宜条件下,特别是在昼夜温差大及饱和湿度条件下发病迅速,病情加重,导致黄瓜减产,造成经济损失。对黄瓜褐斑病进行病斑分割与提取,可以为后续的病害识别与诊断提供有效依据,具有重要意义。结合黄瓜褐斑病可见光谱图像,利用U-net深度学习网络构建黄瓜褐斑病语义分割模型,实现了病斑分割。首先在采集到的黄瓜褐斑病可见光谱图像中截取病斑较为突出的区域作为样本,共在40幅图像中截取到135个像素区域,区域的像素分辨率为200×200,利用Matlab的Image Labeler工具对样本进行像素标记,分别标记出感病区域和健康区域。然后构建U-net网络,该网络包含46层和48个连接,通过卷积层和线性整流层以及最大池化法来完成病斑特征提取,通过深度连接层以及上卷积层和上线性整流层完成上采样,通过跳层连接来完成复制和剪裁操作,并进行病斑特征融合。利用所构建的U-net网络进行学习训练得到语义分割模型,在135个样本中,随机选取其中96个作为训练样本,剩余的39个作为测试样本,设置迭代次数为240次,L2正则化系数为0.000 1,初始学习率为0.05,动量参数为0.9,梯度阈值为0.05,进行样本训练和测试。经过10次重复训练和测试,结果表明,基于U-net和可见光谱图像的黄瓜褐斑病语义分割模型执行时间平均为46.4 s,内存占用平均为6 665.8 MB,执行效率较高;模型准确率PA为96.23%~97.98%,MPA为97.28%~97.87%,MIoU为86.10%~91.59%,FWIoU为93.33%~96.19%,模型的稳定性较好、泛化能力较强。该研究方法利用较少的训练样本,获得了准确率较高的分割模型,为小样本机器学习提供了参考,同时为其他蔬菜的病斑分割、病害识别与诊断提供了方法依据。     

Quantum memories

We perform a review of various approaches to the implementation of quantum memories, with an emphasis on activities within the quantum memory sub-project of the EU integrated project “Qubit Applications”. We begin with a brief overview over different applications for quantum memories and different types of quantum memories. We discuss the most important criteria for assessing quantum memory performance and the most important physical requirements. Then we review the different approaches represented in “Qubit Applications” in some detail. They include solid-state atomic ensembles, NV centers, quantum dots, single atoms, atomic gases and optical phonons in diamond. We compare the different approaches using the discussed criteria.     

基于深度学习的铸件X射线图像分割研究

针对当前图像分割算法在实现工业铸件内部缺陷分割上精度低且算法不够轻量化的问题,提出一种基于改进DeepLabv3+的工业铸件内部缺陷检测算法Effi-DeepLab。该方法采用EfficientNet中的MBConv来代替原有的Xception模块进行特征提取,使特征提取网络更加高效与轻量化;针对工业铸件内部缺陷尺寸小的问题,重新设计空洞空间金字塔池化（ASPP）层中空洞卷积的扩张率,使得卷积块对小目标具有更高的鲁棒性;在解码端充分利用特征提取阶段的低阶语义信息进行多尺度特征融合,以提高小目标缺陷分割的精度。实验结果表明,在本文使用的汽车轮毂内部缺陷图像数据集中,Effi-DeepLab模型对缺陷的分割准确率和平均交并比（mIoU）分别为93.58%和89.39%,相比DeepLabv3+分别提升了2.65%和2.24%,具有更好的分割效果;此外,还通过实验验证了本文提出算法具有良好的泛化性。     

一种高精度图像采集系统的设计与特点

介绍了一种新型、高精度、实时伪彩色图像采集系统的原理、构成和特点。该系统采用１０ｂｉｔＡ／Ｄ转换器，使得高性能图像分析与处理成为可能。系统采样频率为１４．７５ＭＨｚ，对于标准ＰＡＬ制电视信号，可以保证采样后像素的横纵比为１∶１。由于采用最新ＥＰＬＤ编程技术，使得系统体积减小，可靠性显著提高。可输出标准ＰＡＬ制黑白全电视信号、ＲＧＢ分量彩色信号、Ｙ／Ｃ分量彩色信号及复合视频信号。该系统通过标准ＩＳＡ总线与计算机接口，帧存直接映射计算机内存地址空间，存取速度快。本系统特别适合于高精度的图像测量应用     

Moving object detection based on improved VIBE and graph cut optimization

Extracting foreground moving objects from video sequences is an important task and also a hot topic in computer vision and image processing. Segmentation results can be used in many object-based video applications such as object-based video coding, content-based video retrieval, intelligent video surveillance and video-based human–computer interaction. In this paper, we present a novel moving object detection method based on improved VIBE and graph cut method from monocular video sequences. Firstly, perform moving object detection for the current frame based on improved VIBE method to extract the background and foreground information; then obtain the clusters of foreground and background respectively using mean shift clustering on the background and foreground information; Third, initialize the S/T Network with corresponding image pixels as nodes (except S/T node); calculate the data and smoothness term of graph; finally, use max flow/minimum cut to segmentation S/T network to extract the motion objects. Experimental results on indoor and outdoor videos demonstrate the efficiency of our proposed method.     

Automatic Search Dense Connection Module for Super-Resolution

The development of display technology has continuously increased the requirements for image resolution. However, the imaging systems of many cameras are limited by their physical conditions, and the image resolution is often restrictive. Recently, several models based on deep convolutional neural network (CNN) have gained significant performance for image super-resolution (SR), while extensive memory consumption and computation overhead hinder practical applications. For this purpose, we present a lightweight network that automatically searches dense connection (ASDCN) for image super-resolution (SR), which effectively reduces redundancy in dense connection and focuses on more valuable features. We employ neural architecture search (NAS) to model the searching of dense connections. Qualitative and quantitative experiments on five public datasets show that our derived model achieves superior performance over the state-of-the-art models.     

深度神经网络和高光谱显微图像的二维材料纳米片识别

近年来,二维材料由于其独特的性质而受到了广泛关注。在制备二维层状晶体的各种方法中,机械剥离法获得的薄层二维材料晶体质量高,适用于基础研究及性能演示。然而用机械剥离法从衬底上获得的材料具有一定的随机性,可能包含了少许相对较厚的部分。实现对这些二维薄层材料有效、快速且智能化的表征有利于促进二维材料性能的进一步研究。提出了一种基于深度学习的表征方法,通过搭建的编解码结构的卷积神经网络语义分割算法,可以根据光学显微镜图像进行分割和快速识别二维材料纳米片。卷积神经网络作为深度学习在图像处理领域中的典型算法,能够对光学显微镜图像中的复杂信息进行特征提取。首先采用机械剥离制备MoS₂纳米片样本,通过光学显微镜采集高光谱图像并对样本进行标记,根据样本的厚度范围标记出不同的区域,对标记后的图像进一步处理,包括图像的颜色校准和剪切操作,得到用于网络训练和测试的数据集。针对光学图像中二维纳米薄片存在的低对比度、碎裂等特点,编码时加入残差结构和金字塔池化模型,有助于特征信息的提取;解码时融合编码路径中提取的浅层特征信息,以提高网络分割精度。实验中采用带权重的交叉熵损失函数解决类别数量不平衡问题和采用数据增强扩大数据集。对训练后的网络测试结果表明,模型像素精度为97.38%,平均像素精度为90.38%,均交并比为75.86%。之后通过迁移学习成功地对剥离的单层和双层石墨烯纳米片样本进行了识别,均交并比达到了81.63%,表明该方法具有普适性。通过MoS₂和石墨烯纳米片的识别演示,实现了深度学习在二维材料的光学显微镜图像中的成功应用。该方法有望在更多的二维材料上得到扩展并突破自动动态处理光学显微镜图像的问题,同时为其他纳米材料的高光谱图像处理提供参考。     

Dynamic pixel-wise weighting-based fully convolutional neural networks for left ventricle segmentation in short-axis MRI

Left ventricle (LV) segmentation in cardiac MRI is an essential procedure for quantitative diagnosis of various cardiovascular diseases. In this paper, we present a novel fully automatic left ventricle segmentation approach based on convolutional neural networks. The proposed network fully takes advantages of the hierarchical architecture and integrate the multi-scale feature together for segmenting the myocardial region of LV. Moreover, we put forward a dynamic pixel-wise weighting strategy, which can dynamically adjust the weight of each pixel according to the segmentation accuracy of upper layer and force the pixel classifier to take more attention on the misclassified ones. By this way, the LV segmentation performance of our method can be improved a lot especially for the apical and basal slices in cine MR images. The experiments on the CAP database demonstrate that our method achieves a substantial improvement compared with other well-know deep learning methods. Beside these, we discussed two major limitations in convolutional neural networks-based semantic segmentation methods for LV segmentation.     

Semi-Supervised Cross-Subject Emotion Recognition Based on Stacked Denoising Autoencoder Architecture Using a Fusion of Multi-Modal Physiological Signals

In recent decades, emotion recognition has received considerable attention. As more enthusiasm has shifted to the physiological pattern, a wide range of elaborate physiological emotion data features come up and are combined with various classifying models to detect one’s emotional states. To circumvent the labor of artificially designing features, we propose to acquire affective and robust representations automatically through the Stacked Denoising Autoencoder (SDA) architecture with unsupervised pre-training, followed by supervised fine-tuning. In this paper, we compare the performances of different features and models through three binary classification tasks based on the Valence-Arousal-Dominance (VAD) affection model. Decision fusion and feature fusion of electroencephalogram (EEG) and peripheral signals are performed on hand-engineered features; data-level fusion is performed on deep-learning methods. It turns out that the fusion data perform better than the two modalities. To take advantage of deep-learning algorithms, we augment the original data and feed it directly into our training model. We use two deep architectures and another generative stacked semi-supervised architecture as references for comparison to test the method’s practical effects. The results reveal that our scheme slightly outperforms the other three deep feature extractors and surpasses the state-of-the-art of hand-engineered features.