基于Depth-wise卷积和视觉Transformer的图像分类模型

图像分类作为一种常见的视觉识别任务,有着广阔的应用场景。在处理图像分类问题时,传统的方法通常使用卷积神经网络,然而,卷积网络的感受野有限,难以建模图像的全局关系表示,导致分类精度低,难以处理复杂多样的图像数据。为了对全局关系进行建模,一些研究者将Transformer应用于图像分类任务,但为了满足Transformer的序列化和并行化要求,需要将图像分割成大小相等、互不重叠的图像块,破坏了相邻图像数据块之间的局部信息。此外,由于Transformer具有较少的先验知识,模型往往需要在大规模数据集上进行预训练,因此计算复杂度较高。为了同时建模图像相邻块之间的局部信息并充分利用图像的全局信息,提出了一种基于Depth-wise卷积的视觉Transformer(Efficient Pyramid Vision Transformer,EPVT)模型。EPVT模型可以实现以较低的计算成本提取相邻图像块之间的局部和全局信息。EPVT模型主要包含3个关键组件:局部感知模块(Local Perceptron Module,LPM)、空间信息融合模块(Spatial Information Fusion,SIF)和“+卷积前馈神经网络(Convolution Feed-forward Network,CFFN)。LPM模块用于捕获图像的局部相关性;SIF模块用于融合相邻图像块之间的局部信息,并利用不同图像块之间的远距离依赖关系,提升模型的特征表达能力,使模型学习到输出特征在不同维度下的语义信息;CFFN模块用于编码位置信息和重塑张量。在图像分类数据集ImageNet-1K上,所提模型优于现有的同等规模的视觉Transformer分类模型,取得了82.6%的分类准确度,证明了该模型在大规模数据集上具有竞争力。

基于改进FeatDepth的足球运动场景无监督单目图像深度预测

为了在降低成本的同时提高图像深度信息预测的精确度,并将深度估计应用于足球运动场景,提出一种基于改进FeatDepth的足球运动场景无监督单目图像深度预测方法。首先,对原FeatDepth引入注意力机制,使模型更加关注有效的特征信息;其次,将FeatDepth中的PoseNet网络和DepthNet网络分别嵌入GAM全局注意力机制模块,为网络添加额外的上下文信息,在基本不增加计算成本的情况下提升FeatDepth模型深度预测性能;再次,为在低纹理区域和细节上获得更好的深度预测效果,由单视图重构损失与交叉视图重构损失组合而成最终的损失函数。选取KITTI数据集中Person场景较多的部分进行数据集制作并进行仿真实验,结果表明,改进后的FeatDepth模型不仅在精确度上有所提升,且在低纹理区域及细节处拥有更好的深度预测效果。最后,对比模型在足球场景下的推理效果后得出,改进后的模型在低纹理区域(足球、球门等)及细节处(肢体等)有更好的深度预测效果,实现了将基于无监督的单目深度估计模型应用于足球运动场景的目的。

Three-dimensional crystal defect imaging by STEM depth sectioning

One of the major innovations awaiting in electron microscopy is full three-dimensional imaging at atomic resolution.Despite the success of aberration correction to deep sub-angstrom lateral resolution,spatial resolution in depth is still far from atomic resolution.In scanning transmission electron microscopy(STEM),this poor depth resolution is due to the limitation of the illumination angle.To overcome this physical limitation,it is essential to implement a next-generation aberration corrector in STEM that can significantly improve the depth resolution.This review discusses the capability of depth sectioning for three-dimensional imaging combined with large-angle illumination STEM.Furthermore,the statistical analysis approach remarkably improves the depth resolution,making it possible to achieve three-dimensional atomic resolution imaging at oxide surfaces.We will also discuss the future prospects of three-dimensional imaging at atomic resolution by STEM depth sectioning.

Investigation of Groundwater Quality with Borehole Depth in the Basin Granitoids of the Ashanti Region of Ghana

The dependence of groundwater quality on borehole depth is usually debatable in groundwater studies, especially in complex geological formations where aquifer characteristics vary spatially with depth. This study therefore seeks to investigate the relationship between borehole depth and groundwater quality across the granitoid aquifers within the Birimian Supergroup in the Ashanti Region. Physicochemical analysis records of groundwater quality data were collected from 23 boreholes of public and private institutions in the Ashanti Region of Ghana, and the parametric values of iron, fluoride, total hardness, pH, nitrate, and nitrite were used to study the groundwater quality-depth relationship. The results showed that the depth-to-groundwater quality indicated a marginal increase in water quality in the range of 30 to 50 m, which is mathematically represented by the low-value correlation coefficient (r2 = 0.026). A relatively significant increase occurs in the depth range of 50 to 80 m, which is given by a correlation coefficient of r2 = 0.298. The mean percent parameter compatibility was 74%, 82%, 89%, and 97% at 50, 60, 70, and 80 m depths, respectively. The variations in groundwater quality per depth ratio ranged from 1.48, 1.37, 1.27, and 1.21 for 50, 60, 70, and 80 m depth, respectively. The recommended minimum borehole depth for excellent groundwater quality is suggested with a compatibility per meter depth ratio of 1.37. This results in a range between 50 and 70 m as the most desirable drilling depth for excellent groundwater quality within the granitoids of the Birimian Supergroup of the Ashanti Region in Ghana.

Perpendicular-Cutdepth:Perpendicular Direction Depth Cutting Data Augmentation Method

Depth estimation is an important task in computer vision.Collecting data at scale for monocular depth estimation is challenging,as this task requires simultaneously capturing RGB images and depth information.Therefore,data augmentation is crucial for this task.Existing data augmentationmethods often employ pixel-wise transformations,whichmay inadvertently disrupt edge features.In this paper,we propose a data augmentationmethod formonocular depth estimation,which we refer to as the Perpendicular-Cutdepth method.This method involves cutting realworld depth maps along perpendicular directions and pasting them onto input images,thereby diversifying the data without compromising edge features.To validate the effectiveness of the algorithm,we compared it with existing convolutional neural network(CNN)against the current mainstream data augmentation algorithms.Additionally,to verify the algorithm’s applicability to Transformer networks,we designed an encoder-decoder network structure based on Transformer to assess the generalization of our proposed algorithm.Experimental results demonstrate that,in the field of monocular depth estimation,our proposed method,Perpendicular-Cutdepth,outperforms traditional data augmentationmethods.On the indoor dataset NYU,our method increases accuracy from0.900 to 0.907 and reduces the error rate from0.357 to 0.351.On the outdoor dataset KITTI,our method improves accuracy from 0.9638 to 0.9642 and decreases the error rate from 0.060 to 0.0598.

Extraction of Acoustic Normal Mode Depth Functions Using Range-Difference Method with Vertical Linear Array Data

Data-derived normal mode extraction is an effective method for extracting normal mode depth functions in the absence of marine environmental data.However,when the corresponding singular vectors become nonunique when two or more singular values obtained from the cross-spectral density matrix diagonalization are nearly equal,this results in unsatisfactory extraction outcomes for the normal mode depth functions.To address this issue,we introduced in this paper a range-difference singular value decomposition method for the extraction of normal mode depth functions.We performed the mode extraction by conducting singular value decomposition on the individual frequency components of the signal's cross-spectral density matrix.This was achieved by using pressure and its range-difference matrices constructed from vertical line array data.The proposed method was validated using simulated data.In addition,modes were successfully extracted from ambient noise.

The penetration depth of atomic radicals in tubes with catalytic surface properties

Catalysis of molecular radicals is often performed in interesting experimental configurations.One possible configuration is tubular geometry.The radicals are introduced into the tubes on one side,and stable molecules are exhausted on the other side.The penetration depth of radicals depends on numerous parameters,so it is not always feasible to calculate it.This article presents systematic measurements of the penetration depth of oxygen atoms along tubes made from nickel,cobalt,and copper.The source of O atoms was a surfatron-type microwave plasma.The initial density of O atoms depended on the gas flow and was 0.7×10^(21)m^(-3),2.4×10^(21)m^(-3),and 4.2×10^(21)m^(-3)at the flow rates of 50,300,and 600 sccm,and pressures of 10,35,and 60 Pa,respectively.The gas temperature remained at room temperature throughout the experiments.The dissociation fraction decreased exponentially along the length of the tubes in all cases.The penetration depths for well-oxidized nickel were 1.2,1.7,and 2.4 cm,respectively.For cobalt,they were slightly lower at 1.0,1.3,and 1.6 cm,respectively,while for copper,they were 1.1,1.3,and 1.7 cm,respectively.The results were explained by gas dynamics and heterogeneous surface association.These data are useful in any attempt to estimate the loss of molecular fragments along tubes,which serve as catalysts for the association of various radicals to stable molecules.

HVSR analysis of pumice sands for sediment depth characterization:A case study for Guadalajara,Mexico

The horizontal to vertical spectral ratio(HVSR)methodology is used here to characterize pumice soils and to image the three-dimensional surface geometry of Guadalajara,Mexico.Similar to other Latin American cities,Guadalajara is exposed to high seismic risk,with the particularity of being the largest urban settlement in Latin America built on pumice soils.Methodology has not yet been tested to characterize subsoil depths in pumice sands.Due to the questionable use of traditional geotechnical tests for the analysis of pumice soils,HVSR provides an alternative for its characterization without altering its fragile and porous structure.In this work,resonance frequency(F0)and peak amplitude(A0)are used to constrain the depth of the major impedance contrast that represents the interface between bedrock and pumice soil.Results were compared with borehole depths and other available geotechnical and geophysical data and show good agreement.One of the profiles estimated on the riverbanks that cross the city,reveals different subsoil thickness that could have an impact on different site responses on riverine areas to an eventual earthquake.Government and academic efforts are combined in this work to characterize depth sediments,an important parameter that impacts the regulations for construction in the city.

Effect of burial depth of a new tunnel on the seismic response of an existing tunnel

Burial depth is a crucial factor affecting the forces and deformation of tunnels during earthquakes.One key issue is a lack of understanding of the effect of a change in the buried depth of a single-side tunnel on the seismic response of a double-tunnel system.In this study,shaking table tests were designed and performed based on a tunnel under construction in Dalian,China.Numerical models were established using the equivalent linear method combined with ABAQUS finite element software to analyze the seismic response of the interacting system.The results showed that the amplification coefficient of the soil acceleration did not change evidently with the burial depth of the new tunnel but decreased as the seismic amplitude increased.In addition,the existing tunnel acceleration,earth pressure,and internal force were hardly affected by the change in the burial depth;for the new tunnel,the acceleration and internal force decreased as the burial depth increased,while the earth pressure increased.This shows that the earth pressure distribution in a double-tunnel system is relatively complex and mainly concentrated on the arch spandrel and arch springing of the relative area.Overall,when the horizontal clearance between the center of the two tunnels was more than twice the sum of the radius of the outer edges of the two tunnels,the change in the burial depth of the new tunnel had little effect on the existing one,and the tunnel structure was deemed safe.These results provide a preliminary understanding and reference for the seismic performance of a double-tunnel system.

Quantitative prediction model for the depth limit of oil accumulation in the deep carbonate rocks:A case study of Lower Ordovician in Tazhong area of Tarim Basin

With continuous hydrocarbon exploration extending to deeper basins,the deepest industrial oil accumulation was discovered below 8,200 m,revealing a new exploration field.Hence,the extent to which oil exploration can be extended,and the prediction of the depth limit of oil accumulation(DLOA),are issues that have attracted significant attention in petroleum geology.Since it is difficult to characterize the evolution of the physical properties of the marine carbonate reservoir with burial depth,and the deepest drilling still cannot reach the DLOA.Hence,the DLOA cannot be predicted by directly establishing the relationship between the ratio of drilling to the dry layer and the depth.In this study,by establishing the relationships between the porosity and the depth and dry layer ratio of the carbonate reservoir,the relationships between the depth and dry layer ratio were obtained collectively.The depth corresponding to a dry layer ratio of 100%is the DLOA.Based on this,a quantitative prediction model for the DLOA was finally built.The results indicate that the porosity of the carbonate reservoir,Lower Ordovician in Tazhong area of Tarim Basin,tends to decrease with burial depth,and manifests as an overall low porosity reservoir in deep layer.The critical porosity of the DLOA was 1.8%,which is the critical geological condition corresponding to a 100%dry layer ratio encountered in the reservoir.The depth of the DLOA was 9,000 m.This study provides a new method for DLOA prediction that is beneficial for a deeper understanding of oil accumulation,and is of great importance for scientific guidance on deep oil drilling.

Three-dimensional constrained gravity inversion of Moho depth and crustal structural characteristics at Mozambique continental margin

Mozambique's continental margin in East Africa was formed during the break-off stage of the east and west Gondwana lands. Studying the geological structure and division of continent-ocean boundary(COB) in Mozambique's continental margin is considered of great significance to rebuild Gondwana land and understand its movement mode. Along these lines, in this work, the initial Moho was fit using the known Moho depth from reflection seismic profiles, and a 3D multi-point constrained gravity inversion was carried out. Thus, highaccuracy Moho depth and crustal thickness in the study area were acquired. According to the crustal structure distribution based on the inversion results, the continental crust at the narrowest position of the Mozambique Channel was detected. According to the analysis of the crustal thickness, the Mozambique ridge is generally oceanic crust and the COB of the whole Mozambique continental margin is divided.

What induced the trend shift of mixed-layer depths in the Antarctic Circumpolar Current region in the mid-1980s?

An obvious trend shift in the annual mean and winter mixed layer depth(MLD)in the Antarctic Circumpolar Current(ACC)region was detected during the 1960–2021 period.Shallowing trends stopped in mid-1980s,followed by a period of weak trends.The MLD deepening trend difference between the two periods were mainly distributed in the western areas in the Drake Passage,the areas north to Victoria Land and Wilkes Land,and the central parts of the South Indian sector.The newly formed ocean current shear due to the meridional shift of the ACC flow axis between the two periods is the dominant driver for the MLD trends shift distributed in the western areas in the Drake Passage and the central parts of the South Indian sector.The saltier trends in the regions north to Victoria Land and Wilkes Land could be responsible for the strengthening mixing processes in this region.

DGConv: A Novel Convolutional Neural Network Approach for Weld Seam Depth Image Detection

We propose a novel image segmentation algorithm to tackle the challenge of limited recognition and segmentation performance in identifying welding seam images during robotic intelligent operations.Initially,to enhance the capability of deep neural networks in extracting geometric attributes from depth images,we developed a novel deep geometric convolution operator(DGConv).DGConv is utilized to construct a deep local geometric feature extraction module,facilitating a more comprehensive exploration of the intrinsic geometric information within depth images.Secondly,we integrate the newly proposed deep geometric feature module with the Fully Convolutional Network(FCN8)to establish a high-performance deep neural network algorithm tailored for depth image segmentation.Concurrently,we enhance the FCN8 detection head by separating the segmentation and classification processes.This enhancement significantly boosts the network’s overall detection capability.Thirdly,for a comprehensive assessment of our proposed algorithm and its applicability in real-world industrial settings,we curated a line-scan image dataset featuring weld seams.This dataset,named the Standardized Linear Depth Profile(SLDP)dataset,was collected from actual industrial sites where autonomous robots are in operation.Ultimately,we conducted experiments utilizing the SLDP dataset,achieving an average accuracy of 92.7%.Our proposed approach exhibited a remarkable performance improvement over the prior method on the identical dataset.Moreover,we have successfully deployed the proposed algorithm in genuine industrial environments,fulfilling the prerequisites of unmanned robot operations.

Applying Source Parameter Imaging (SPI) to Aeromagnetic Data to Estimate Depth to Magnetic Sources in the Mamfe Sedimentary Basin

Aeromagnetic data over the Mamfe Basin have been processed. A regional magnetic gridded dataset was obtained from the Total Magnetic Intensity (TMI) data grid using a 3 × 3 convolution (Hanning) filter to remove regional trends. Major similarities in magnetic field orientation and intensities were observed at identical locations on both the regional and TMI data grids. From the regional and TMI gridded datasets, the residual dataset was generated which represents the very shallow geological features of the basin. Processing this residual data grid using the Source Parameter Imaging (SPI) for magnetic depth suggests that the estimated depths to magnetic sources in the basin range from about 271 m to 3552 m. The highest depths are located in two main locations somewhere around the central portion of the study area which correspond to the area with positive magnetic susceptibilities, as well as the areas extending outwards across the eastern boundary of the study area. Shallow magnetic depths are prominent towards the NW portion of the basin and also correspond to areas of negative magnetic susceptibilities. The basin generally exhibits a variation in depth of magnetic sources with high, average and shallow depths. The presence of intrusive igneous rocks was also observed in this basin. This characteristic is a pointer to the existence of geologic resources of interest for exploration in the basin.

Edge enhanced depth perception with binocular meta-lens

The increasing popularity of the metaverse has led to a growing interest and market size in spatial computing from both academia and industry.Developing portable and accurate imaging and depth sensing systems is crucial for advancing next-generation virtual reality devices.This work demonstrates an intelligent,lightweight,and compact edge-enhanced depth perception system that utilizes a binocular meta-lens for spatial computing.The miniaturized system comprises a binocular meta-lens,a 532 nm filter,and a CMOS sensor.For disparity computation,we propose a stereo-matching neural network with a novel H-Module.The H-Module incorporates an attention mechanism into the Siamese network.The symmetric architecture,with cross-pixel interaction and cross-view interaction,enables a more comprehensive analysis of contextual information in stereo images.Based on spatial intensity discontinuity,the edge enhancement eliminates illposed regions in the image where ambiguous depth predictions may occur due to a lack of texture.With the assistance of deep learning,our edge-enhanced system provides prompt responses in less than 0.15 seconds.This edge-enhanced depth perception meta-lens imaging system will significantly contribute to accurate 3D scene modeling,machine vision,autonomous driving,and robotics development.

Seasonal influence of freshwater discharge on spatio-temporal variations in primary productivity, sea surface temperature, and euphotic zone depth in the northern Bay of Bengal

Ocean productivity is the foundation of marine food web,which continuously removes atmospheric carbon dioxide and supports life at sea and on land.Spatio-temporal variability of net primary productivity(NPP),sea surface temperature(SST),sea surface salinity(SSS),mixed layer depth(MLD),and euphotic zone depth(EZD) in the northern B ay of Bengal(BoB) during three monsoon seasons were examined in this study based on remote sensing data for the period 2005 to 2020.To compare the NPP distribution between the coastal zones and open BoB,the study area was divided into five zones(Z1-Z5).Results suggest that most productive zones Z2 and Zl are located at the head bay area and are directly influenced by freshwater discharge together with riverine sediment and nutrient loads.Across Z1-Z5,the NPP ranges from 5 315.38 mg/(m^(2)·d) to 346.7 mg/(m^(2)·d)(carbon,since then the same).The highest monthly average NPP of 5 315.38 mg/(m^(2)·d) in February and 5 039.36 mg/(m^(2)·d) in June were observed from Z2,while the lowest monthly average of 346.72 mg/(m^(2)·d) was observed in March from Z4,which is an oceanic zone.EZD values vary from 6-154 m for the study area,and it has an inverse correlation with NPP concentration.EZD is deeper during the summer season and shallower during the wintertime,with a corresponding increase in productivity.Throughout the year,monthly SST shows slight fluctuation for the entire study area,and statistical analysis shows a significant correlation among NPP,and EZD,overall positive between NPP and MLD,whereas no significant correlation among SSS,and SST for the northern BoB.Long-term trends in SST and productivity were significantly po sitive in head bay zones but negatively productive in the open ocean.The findings in this study on the distribution of NPP,SST,SSS,MLD,and EZD and their seasonal variability in five different zones of BoB can be used to further improve the management of marine resources and overall environmental condition in response to climate changes in BoB as they are of utmost relevance to the fisheries for the three bordering countries.

Spatiotemporal Changes of Snow Depth in Western Jilin,China from 1987 to 2018

Seasonal snow cover is a key global climate and hydrological system component drawing considerable attention due to glob-al warming conditions.However,the spatiotemporal snow cover patterns are challenging in western Jilin,China due to natural condi-tions and sparse observation.Hence,this study investigated the spatiotemporal patterns of snow cover using fine-resolution passive mi-crowave(PMW)snow depth(SD)data from 1987 to 2018,and revealed the potential influence of climate factors on SD variations.The results indicated that the interannual range of SD was between 2.90 cm and 9.60 cm during the snowy winter seasons and the annual mean SD showed a slightly increasing trend(P>0.05)at a rate of 0.009 cm/yr.In snowmelt periods,the snow cover contributed to an increase in volumetric soil water,and the change in SD was significantly affected by air temperature.The correlation between SD and air temperature was negative,while the correlation between SD and precipitation was positive during December and March.In March,the correlation coefficient exceeded 0.5 in Zhenlai,Da’an,Qianan,and Qianguo counties.However,the SD and precipitation were neg-atively correlated over western Jilin in October,and several subregions presented a negative correlation between SD and precipitation in November and April.

Automatic depth matching method of well log based on deep reinforcement learning

In the traditional well log depth matching tasks,manual adjustments are required,which means significantly labor-intensive for multiple wells,leading to low work efficiency.This paper introduces a multi-agent deep reinforcement learning(MARL)method to automate the depth matching of multi-well logs.This method defines multiple top-down dual sliding windows based on the convolutional neural network(CNN)to extract and capture similar feature sequences on well logs,and it establishes an interaction mechanism between agents and the environment to control the depth matching process.Specifically,the agent selects an action to translate or scale the feature sequence based on the double deep Q-network(DDQN).Through the feedback of the reward signal,it evaluates the effectiveness of each action,aiming to obtain the optimal strategy and improve the accuracy of the matching task.Our experiments show that MARL can automatically perform depth matches for well-logs in multiple wells,and reduce manual intervention.In the application to the oil field,a comparative analysis of dynamic time warping(DTW),deep Q-learning network(DQN),and DDQN methods revealed that the DDQN algorithm,with its dual-network evaluation mechanism,significantly improves performance by identifying and aligning more details in the well log feature sequences,thus achieving higher depth matching accuracy.

Improving path planning efficiency for underwater gravity-aided navigation based on a new depth sorting fast search algorithm

This study focuses on the improvement of path planning efficiency for underwater gravity-aided navigation.Firstly,a Depth Sorting Fast Search(DSFS)algorithm was proposed to improve the planning speed of the Quick Rapidly-exploring Random Trees*(Q-RRT*)algorithm.A cost inequality relationship between an ancestor and its descendants was derived,and the ancestors were filtered accordingly.Secondly,the underwater gravity-aided navigation path planning system was designed based on the DSFS algorithm,taking into account the fitness,safety,and asymptotic optimality of the routes,according to the gravity suitability distribution of the navigation space.Finally,experimental comparisons of the computing performance of the ChooseParent procedure,the Rewire procedure,and the combination of the two procedures for Q-RRT*and DSFS were conducted under the same planning environment and parameter conditions,respectively.The results showed that the computational efficiency of the DSFS algorithm was improved by about 1.2 times compared with the Q-RRT*algorithm while ensuring correct computational results.

Responses of the Southern Ocean mixed layer depth to the eastern and central Pacific El Niño events during austral winter

Based on the Ocean Reanalysis System version 5(ORAS5)and the fifth-generation reanalysis datasets derived from European Centre for Medium-Range Weather Forecasts(ERA5),we investigate the different impacts of the central Pacific(CP)El Niño and the eastern Pacific(EP)El Niño on the Southern Ocean(SO)mixed layer depth(MLD)during austral winter.The MLD response to the EP El Niño shows a dipole pattern in the South Pacific,namely the MLD dipole,which is the leading El Niño-induced MLD variability in the SO.The tropical Pacific warm sea surface temperature anomaly(SSTA)signal associated with the EP El Niño excites a Rossby wave train propagating southeastward and then enhances the Amundsen Sea low(ASL).This results in an anomalous cyclone over the Amundsen Sea.As a result,the anomalous southerly wind to the west of this anomalous cyclone advects colder and drier air into the southeast of New Zealand,leading to surface cooling through less total surface heat flux,especially surface sensible heat(SH)flux and latent heat(LH)flux,and thus contributing to the mix layer(ML)deepening.The east of the anomalous cyclone brings warmer and wetter air to the southwest of Chile,but the total heat flux anomaly shows no significant change.The warm air promotes the sea ice melting and maintains fresh water,which strengthens stratification.This results in a shallower MLD.During the CP El Niño,the response of MLD shows a separate negative MLD anomaly center in the central South Pacific.The Rossby wave train triggered by the warm SSTA in the central Pacific Ocean spreads to the Amundsen Sea,which weakens the ASL.Therefore,the anomalous anticyclone dominates the Amundsen Sea.Consequently,the anomalous northerly wind to the west of anomalous anticyclone advects warmer and wetter air into the central and southern Pacific,causing surface warming through increased SH,LH,and longwave radiation flux,and thus contributing to the ML shoaling.However,to the east of the anomalous anticyclone,there is no statistically significant impact on the MLD.