Study on the Submarine Slope Stability of the Deep Channel in the Caofeidian Harbor

Objective The greatest advantage of the Caofeidian Harbor is its deep channel facing the Bohai Bay. The deep channel is a natural port hub for shipping of the Caofeidian Habor. The construction of the Caofeidian Harbor has impacted the hydrodynamic environment and the sediments movement, which has attracted much attention about the geomorphic evolution, slope stability and the evolution trend after submarine slope destruction. Insight from this study might be significant for the future development of the Caofeidian Habor, including planning, operation and maintenance.

Seismic stratigraphy of the Qiongdongnan deep sea channel system,northwest South China Sea

Based on more than 4000 km 2D seismic data and seismic stratigraphic analysis, we discussed the extent and formation mechanism of the Qiongdongnan deep sea channel. The Qiongdongnan deep sea channel is a large incised channel which extends from the east boundary of the Yinggehai Basin, through the whole Qiongdongnan and the Xisha trough, and terminates in the western part of the northwest subbasin of South China Sea. It is more than 570 km long and 4–8 km wide. The chaotic (or continuous) middle (or high) amplitude, middle (or high) continuity seismic facies of the channel reflect the different lithological distribution of the channel. The channel formed as a complex result of global sea level drop during early Pliocene, large scale of sediment supply to the Yinggehai Basin, inversion event of the Red River strike-slip fault, and tilted direction of the Qiongdongnan Basin. The large scale of sediment supply from Red River caused the shelf break of the Yinggehai Basin to move torwards the S and SE direction and developed large scale of prograding wedge from the Miocene, and the inversion of the Red River strike-slip fault induced the sediment slump which formed the Qiongdongnan deep sea channel.

Sedimentary characteristics comparison and genesis analysis of the deepwater channel in the hydrate enrichment zones on the north slope of the South China Sea

Natural gas hydrate(NGH) is one of the important clean energy at present and even in the future. The study of its sedimentary environment and minerogenetic condition has long been a hot issue that has received much concern from geologists all over the world. China has successfully obtained the samples of NGH in Shenhu and Dongsha sea areas in 2007, 2013 and 2015, respectively. From this, the continental slope north of the South China Sea becomes an important test site for the study of NGH sedimentary genesis and minerogenetic condition. NGH has been found in Shenhu, Dongsha and Qiongdongnan areas within the continental slope north of South China Sea,at different depths of water, with different sedimentary characteristics, gas genesis, and minerogenetic conditions.Using a seismic sedimentology theory, combining seismic facies results of each facies, sedimentary facies and evolution of each area are documented in turn establishing a sedimentary model by considering palaeogeomorphology, sea level change and tectonic movement. The channel system and MTD(Mass Transport Deposition) system among these three areas were compared focusing on the developing position, appearance and controlling factors. Relative location among three areas is firstly defined that Dongsha area in a nearprovenance steep upper slope, Shenhu area in a normal gentle slope and Qiongdongnan area in an awayprovenance flat plain. Besides, their channel systems are classified into erosional, erosional-aggradational and aggradational channel, and MTD systems into headwall domain, translational domain and toe domain.

基于改进DeepLabV3+的囊型肝包虫病超声图像分割算法

目的:将基于改进DeepLabV3+的图像语义分割算法应用到囊型肝包虫病超声图像处理中,实现肝包虫病病灶的自动分割与检测,提高临床诊断效率。方法:本研究采用了DeepLabV3+图像语义分割网络为基础方法,并对其进行了以下改进。首先,为解决DeepLabV3+图像语义分割方法计算复杂度高,内存消耗大,难以在计算能力有限的嵌入式平台上部署,在提取图像特征信息时难以充分利用多尺度信息等问题,以MobileNetV2替换模型的原主干网络Xception,获得轻量级的模型框架。其次,将高效通道注意力应用于底层特征,降低计算复杂度,提高目标边界的清晰度。最后,将Dice Loss引入模型中,缓解模型更关注背景区域,而忽略了包含目标的前景区域等问题。结果:在自建囊型肝包虫病VOC2007数据集5种病灶类型上进行验证,实验结果表明,改进模型的平均交并比和平均像素精度分别达到73.8%和83.5%,能够预测更精细的语义分割结果,有效地优化模型复杂度和分割精度。

Channel Estimation with an Interpolation Trained Deep Neural Network

This paper proposes a deep learning-based channel estimation method for orthogonal frequency-division multiplexing (OFDM) systems. The existing OFDM receiver has low estimation accuracy when estimating channel state information (CSI) with fewer pilots. To tackle the problem, in this paper, a deep learning model is first trained by the interpolated channel frequency responses (CFRs) and then used to denoise the CFR estimated by least square (LS) estimation. The proposed deep neural network (DNN) can also be trained in a short time because it only learns the CFR and the network structure is simple. According to the simulation results, the performance of the DNN estimator can be compared with the minimum mean-square error (MMSE) estimator. Furthermore, the DNN approach is more robust than conventional methods when fewer pilots are used. In summary, deep learning is a promising tool for channel estimation in wireless communications.

Geologic-Geophysical Indicators of the Deep Structure of Zones of Geothermal Anomalies for Allocation of Channels of the Deep Heat and Mass Transfer

On the basis of the analysis of field thermogeochemical data along abnormal zones of a thermal stream in the Bukhara-Khiva, oil-and-gas region of the Turan (Tegermen, Chagakul, Shimoly Alat, Beshtepa) was succeeded to obtain important data on a deep structure of sites. Data of gas-chemical and geothermal observations show about confinedness of abnormal concentration of methane to zones of the increased values of the temperature field the measured values of temperatures (Tegermen Square and others). On geoelectric section mines 2-D of inversion of the MT-field depth of 4000 m are lower, among very high-resistance the chemogenic and carbonate deposits of the Paleozoic is traced the subvertical carrying-out abnormal zone. This zone is identified as the channel of a deep heat and mass transfer with which hydrocarbon (HC) deposits are connected. It is shown that electro-investigation when using a geophysical complex can and has to become “advancing” at exploration by oil and gas.

基于多通道特征融合的人体动作识别方法

现阶段,深度学习已在基于WiFi的人体动作识别领域得到广泛应用且取得显著成果。然而,在利用多输入、多输出(MIMO)系统强大的空间分集特性进行动作识别时,受多径效应影响,获得信道状态信息(CSI)存在对相同动作的特征描述存在差异、不同动作的特征描述存在类似、特征提取不完整和动作分类复杂的问题。为解决上述问题,本文提出一种基于双重注意力机制和多通道、多尺度的时间卷积网络的动作识别方法。首先,根据MIMO系统的空间分集特性,构建多通道信息提取模型,从各个天线接收到的信道中提取出有关动作的特性信息。然后,设计多尺度的统合机制,强化同一动作在不同通道接收数据的表征,通过整合不同尺度的动作特征,增强对动作的表征能力。再次,采用特征图融合注意力机制和特征通道注意力机制对各通道的动作特征进行聚合。注意力机制能有效地找出对最终动作识别有重要贡献的特征,使模型可以更好地进行特征聚焦。与此同时,将时间卷积网络应用于特征处理过程,使不同时间步的动作特征间的长期依赖关系得以维持,增加对复杂和连续动作的识别能力。最终,利用全局平均池化层(GAP)将各通道的特征图与动作分类器进行连接,以便多通道的动作特性能有效聚合在一起,进一步提高动作识别的精度。本文提出的模型在公共数据集7种动作测试中,实现98.72%的平均准确率。同时在自行搭建的实验室、教室和走廊等真实环境下进行测试时,10种不同的动作分别获得97.94%、97.28%和95.66%的识别准确率。实验结果充分证明了本文所提出的基于WiFi的人体动作识别模型在不同环境的有效性和优越性。

结合通道剪枝和通道注意力的轻量型车辆点云补全网络

针对现有的点云补全网络多关注于补全的精度而忽视补全效率问题,提出了一种轻量型点云补全网络来准确、高效地修复自动驾驶中的不完整车辆点云。为了提高网络推理效率,采用一种高效的一次性通道剪枝技术提高网络的补全效率;在特征提取阶段,网络加入通道注意力模块,将加权特征与全局特征拼接,通过两层多维特征信息提取,得到最终的特征向量;将特征向量传入双解码器结构中,分别通过全连接层和多层感知机生成稠密的粗糙点云和输入点云偏差值;将粗糙点云与输入点云偏差值相加得到最终的精细化完整点云。在PCN数据集和KITTI数据集上进行实验,实验结果表明在补全缺失车辆信息的实时性上有着显著的提升,并且在补全精度上也有不错的表现。

基于深度学习的水声信道估计技术

在水声通信中,传统的正交频分复用(Orthogonal Frequency Division Multiplexing, OFDM)系统需要大量导频数量和循环前缀以维持系统性能,严重浪费有限的频域资源。因此,利用深度学习辅助OFDM恢复失真的传输数据。具体地讲,利用改进的残差块提取和双向记忆接收信号。将自归一化网络(Self-Normalizing Network, SNN)与注意力机制结合,有效分配信道权重,以便系统更有效地利用信道资源,最小化信号失真。使用深度神经网络(Deep Neural Network, DNN)实现接收信号的分类,以准确地恢复接收信号。提出的ICANet(Improved Residual Network, Convolutional Neural Network and Attention Mechanism Self-Normalizing Network)模型可应用于由Bellhop软件生成的水声环境。仿真结果证明,与传统技术中的最小二乘法(Least Square, LS)以及现有的深度学习模型相比,所提出的模型在循环前缀受限的情况下可达到更低的误码率。

面向OTFS-ISAC系统的智能信道估计现状、挑战与展望

OTFS和ISAC技术均是6G移动通信的关键候选技术,OTFS和ISAC的融合系统OTFS-ISAC是当前移动通信研究的前沿。信道估计是接收机的关键处理,对系统性能起着重要的作用。同时,人工智能技术日益成为重要的通信系统信号处理手段。为此,对OTFS-ISAC系统的智能信道估计进行了综述。首先描述OTFS-ISAC的系统模型,包括调制、解调以及雷达和通信模型;其次,详细阐述了三种智能信道估计方法:基于贝叶斯学习的稀疏估计、基于具有自适应阈值的深度卷积残差网络的信道估计和基于迭代深度学习网络的信道估计,这些方法利用了人工智能技术为信道估计问题提供了新的解决途径;然后,探讨了OTFS-ISAC系统中信道估计面临的技术挑战,包括信道特性的复杂性、参数估计的不一致性与复杂性、资源分配和开销问题以及技术融合与兼容性问题;最后,展望了技术创新与突破、标准化与规范化、应用场景的拓展以及跨领域合作与融合的未来发展方向。

基于通道注意力的多任务卡具缺损检测算法

为及时地发现铁路通信漏缆卡具缺损现象,保障铁路运营安全,该文提出一种基于通道注意力的多任务卡具缺损检测算法。在YOLOv7-tiny的基础上,该算法通过在检测头增加通道注意力机制的方式,平衡不同通道之间的重要程度,提高了模型检测能力。为了实现对卡具的状态检测,算法在类别损失函数中额外增加一项状态损失,并在其中嵌入Focal loss来克服样本不足的问题。为了验证算法的有效性,该文分别对实际线路中的四种类型卡具进行实验。实验结果表明,该文提出的算法效果好、速度快,能满足实际现场需求。

面向无人机协同定位的机载深度计算编译优化

随着无人机技术快速发展,在定位信号缺失的情况下进行无人机定位成为一个研究难题。而近几年图神经网络的出现与发展,为解决这一难题提供了一种新的解决思路。然而在资源受限的无人机端侧部署图神经网络面临着无人机算储资源受限及实时性难以满足等挑战。提出面向无人机协同定位的机载深度计算编译优化方法。采用了一种轻量化的时间图卷积神经网络模型,该时间图卷积网络由图卷积网络和门控递归单元组成,将无人机群的空间依赖性和无人机位置变化的时间依赖性同时加以考虑,对无人机群位置进行精确的预测;针对该模型在时间图卷积网络上的冗余特性,提出了基于逆向Cuthill-McKee图重排和基于双深度确定性策略梯度的全局自适应剪枝算法。在保证无人机群坐标精确预测的同时,不仅能提高数据在主存的空间局部性,加速模型的运算速度,而且能够对模型进行自适应的非结构化剪枝,降低模型的存储复杂度。实验结果表明,相对于已有的时间图卷积神经网络模型,编译优化后的轻量化时间图卷积神经网络模型在保留78.8%准确率的同时,模型计算时间降低37.9%,模型的平均剪枝率达到90.3%。

多种地震属性预测致密砂岩河道——以西加盆地Deep basin地区为例

为了准确描述西加盆地致密砂岩储层的河道展布,以西加盆地Deep basin油田的致密砂岩储层为例,通过对晚白垩世Dunvegan致密砂岩储层在地震上的异常特征分析,总结出了致密砂岩河道的展布特征。利用地震相干分析技术确定了河道边界为北东向展布,宽度范围为2～3km;利用频谱分解技术预测出了河道主体的砂体发育,结合测井相、沉积相资料确定相对较好储层为河道上倾方向,即工区中东部地区,钻井验证东北部河道区获初产油256bbl/d。以上成果认识为钻探部署、经济开发提供了地质依据。

基于EAST与SVTR的芯片表面字符识别方法

为提高芯片表面字符识别的实时性和准确率,提出一种基于EAST与SVTR的字符识别算法。针对EAST文本检测算法,将主干特征提取网络替换为轻量化的深度神经网络FasterNet-T0,减少网络的计算量;添加通道注意力机制自适应学习不同通道的权重分配,加强对重要特征的筛选。改进获得文本区域得分的损失函数,采用Dice损失缓解因图像背景面积过大导致误检的问题。文本方向校正算法对图像中任意方向的文本进行水平校正。由单一视觉模型的文本识别算法SVTR完成对字符的识别。实验结果表明,改进后文本检测算法的精确率、召回率较原算法分别提升了2.43%和4.66%,单帧图片的检测速度提升了0.005 s;添加文本方向校正算法后,识别准确率提升了1.92%。与现有方法对比,验证了该算法的有效性。

基于逐尺度卷积的OFDM信道估计算法

针对快衰落信道环境下的无线通信正交频分复用系统中信道估计算法估计难度大、复杂度高等问题,提出一种轻量级的逐尺度卷积信道估计网络(scale-wise convolution channel estimation network,SCCENet)。对输入的信道状态信息矩阵下采样,构建多尺度特征金字塔,通过构建的逐尺度卷积部分以网络权重共享的方式对信道状态信息矩阵构成的多尺度特征金字塔进行不同尺度的特征融合,在不增加网络参数情况下增大感受野。仿真结果表明,该算法估计性能优于最小二乘算法,能以较低的估计复杂度取得相较于其它深度学习信道估计算法更低的估计误差。

融合累积分布函数和通道注意力机制的DeepLabV3+图像分割算法

为了解决DeepLabV3+在语义分割时未充分利用主干的低级特征,以及大倍数上采样造成有效特征缺失的问题,提出一种累积分布通道注意力DeepLabV3+(CDCA-DLV3+)模型。首先,基于累积分布函数和通道注意力提出了累积分布通道注意力(CDCA);然后,利用CDCA获取主干部分有效的低级特征;最后,采用特征金字塔网络(FPN)进行特征融合和逐步上采样,从而避免大倍数上采样所造成的特征损失。CDCA-DLV3+模型在Pascal VOC2012验证集与Cityscapes数据集上的平均交并比(mIoU)分别为80.09%和80.11%,相较于DeepLabV3+模型分别提升1.24和1.02个百分点。实验结果表明,所提模型分割结果更加精确。

SENet优化的Deeplabv3+淡水鱼体语义分割

淡水鱼头、腹、鳍的各部分快速识别与精准定位是机器人实现淡水鱼快速抓取,精确切割、提升作业效率关键技术的前提。针对深度学习的淡水鱼体语义分割算法在编码特征提取阶段产生大量无效的特征通道,以及网络不断下采样和池化操作使得鱼体某些细节信息被丢失,网络性能下降、边缘分割效果不佳的问题,提出了一种基于SENet优化后的Deeplabv3+淡水鱼头、腹、鳍的语义分割算法。利用空洞/带孔卷积(dilated/atrous convolutions)实现扩展感受野,克服细节信息丢失,达到准确定位的目的,同时SENet的优化使得Deeplabv3+通过学习的方式提升淡水鱼有用的特征并抑制对当前任务用处不大的特征,最终淡水鱼各部分的语义分割平均交并比(MIoU)在自建的淡水鱼数据集上达到了93%左右,性能得到了明显提升并达到了目前先进分割水平。

A SURVEY OF DEEP SPACE COMMUNICATIONS

Deep space communications has played an important role in deep space exploration. Compared with common satellite and terrestrial communications, deep space communications faces more challenging environment. The paper investigated the unique features of deep space communica-tions in detail, discussed the key technologies and its development trends for deep space communica-tions.

Robust and Efficient Data Transmission over Noisy Communication Channels Using Stacked and Denoising Autoencoders

We study the effects of quantization and additive white Gaussian noise(AWGN) in transmitting latent representations of images over a noisy communication channel. The latent representations are obtained using autoencoders(AEs). We analyze image reconstruction and classification performance for different channel noise powers, latent vector sizes, and number of quantization bits used for the latent variables as well as AEs’ parameters. The results show that the digital transmission of latent representations using conventional AEs alone is extremely vulnerable to channel noise and quantization effects. We then propose a combination of basic AE and a denoising autoencoder(DAE) to denoise the corrupted latent vectors at the receiver. This approach demonstrates robustness against channel noise and quantization effects and enables a significant improvement in image reconstruction and classification performance particularly in adverse scenarios with high noise powers and significant quantization effects.

SENet优化的Deeplabv3+淡水鱼体语义分割

淡水鱼头、腹、鳍的各部分快速识别与精准定位是机器人实现淡水鱼快速抓取,精确切割、提升作业效率关键技术的前提。针对深度学习的淡水鱼体语义分割算法在编码特征提取阶段产生大量无效的特征通道,以及网络不断下采样和池化操作使得鱼体某些细节信息被丢失,网络性能下降、边缘分割效果不佳的问题,提出了一种基于SENet优化后的Deeplabv3+淡水鱼头、腹、鳍的语义分割算法。利用空洞/带孔卷积(dilated/atrous convolutions)实现扩展感受野,克服细节信息丢失,达到准确定位的目的,同时SENet的优化使得Deeplabv3+通过学习的方式提升淡水鱼有用的特征并抑制对当前任务用处不大的特征,最终淡水鱼各部分的语义分割平均交并比(MIoU)在自建的淡水鱼数据集上达到了93%左右,性能得到了明显提升并达到了目前先进分割水平。