Separating and imaging diffractions in dip domain on the basis of slope analysis

Fracture-cave reservoirs in carbonate rocks are characterized by a large difference in fracture and cavity size,and a sharp variation in lithology and velocity,thereby resulting in complex diffraction responses.Some small-scale fractures and caves cause weak diffraction energy and would be obscured by the continuous reflection layer in the imaging section,thereby making them difficult to identify.This paper develops a diffraction wave imaging method in the dip domain,which can improve the resolution of small-scale diffractors in the imaging section.Common imaging gathers(CIGs)in the dip domain are extracted by Gaussian beam migration.In accordance with the geometric differences of the diffraction being quasilinear and the reflection being quasiparabolic in the dip-domain CIGs,we use slope analysis technique to filter waves and use Hanning window function to improve the diffraction wave separation level.The diffraction dip-domain CIGs are stacked horizontally to obtain diffraction imaging results.Wavefield separation analysis and numerical modeling results show that the slope analysis method,together with Hanning window filtering,can better suppress noise to obtain the diffraction dip-domain CIGs,thereby improving the clarity of the diffractors in the diffraction imaging section.

Diffraction separation and imaging based on double sparse transforms

Reflection imaging results generally reveal large-scale continuous geological information,and it is difficult to identify small-scale geological bodies such as breakpoints,pinch points,small fault blocks,caves,and fractures,etc.Diffraction imaging is an important method to identify small-scale geological bodies and it has higher resolution than reflection imaging.In the common-offset domain,reflections are mostly expressed as smooth linear events,whereas diffractions are characterized by hyperbolic events.This paper proposes a diffraction extraction method based on double sparse transforms.The linear events can be sparsely expressed by the high-resolution linear Radon transform,and the curved events can be sparsely expressed by the Curvelet transform.A sparse inversion model is built and the alternating direction method is used to solve the inversion model.Simulation data and field data experimental results proved that the diffractions extraction method based on double sparse transforms can effectively improve the imaging quality of faults and other small-scale geological bodies.

Separation of cortical arteries and veins in optical neurovascular imaging

Separation of arteries and veins in the cerebral cortex is of significant importance in the studies of cortical hemodynamics,such as the changes of cerebral blood flow,perfusion or oxygen con-centration in arteries and veins under different pathological and physiological conditions.Yet the cerebral vessel segmentation and vessel-type separation are challenging due to the complexity of cortical vessel characteristics and low spatial signal-to-noise ratio.In this work,we presented an effective full-field method to differentiate arteries and veins in cerebral cortex using dual-modal optical imaging technology including laser speckle imaging(LSI)and optical intrinsic signals(OIS)imaging.The raw contrast images were acquired by LSI and processed with enhanced laser speckle contrast analysis(eLASCA),algorithm.The vascular pattern was extracted and seg-mented using region growing algorithm from the eLASCA-based LSI.Meanwhile,OIS imageswere acquired altermatively with 630 and 870 nm to obtain an oxy hemoglobin concentration mapover cerebral cortex.Then the separation of arteries and veins was accomplished by Otsuthreshold segmentation algorithm based on the OIS information and segmentation of LSI.Finally,the segmentation and separation performances were assessed using area overlap measure(AOM).The segmentation and separation of cerebral vessels in cortical optical imaging have great potential applications in full-field cerebral hemodynamics monitoring and pathological study of cerebral vascular diseases,as well as in clinical intraoperative monitoring.

A Lightweight Convolutional Neural Network with Hierarchical Multi-Scale Feature Fusion for Image Classification

Convolutional neural networks (CNNs) are widely used in image classification tasks, but their increasing model size and computation make them challenging to implement on embedded systems with constrained hardware resources. To address this issue, the MobileNetV1 network was developed, which employs depthwise convolution to reduce network complexity. MobileNetV1 employs a stride of 2 in several convolutional layers to decrease the spatial resolution of feature maps, thereby lowering computational costs. However, this stride setting can lead to a loss of spatial information, particularly affecting the detection and representation of smaller objects or finer details in images. To maintain the trade-off between complexity and model performance, a lightweight convolutional neural network with hierarchical multi-scale feature fusion based on the MobileNetV1 network is proposed. The network consists of two main subnetworks. The first subnetwork uses a depthwise dilated separable convolution (DDSC) layer to learn imaging features with fewer parameters, which results in a lightweight and computationally inexpensive network. Furthermore, depthwise dilated convolution in DDSC layer effectively expands the field of view of filters, allowing them to incorporate a larger context. The second subnetwork is a hierarchical multi-scale feature fusion (HMFF) module that uses parallel multi-resolution branches architecture to process the input feature map in order to extract the multi-scale feature information of the input image. Experimental results on the CIFAR-10, Malaria, and KvasirV1 datasets demonstrate that the proposed method is efficient, reducing the network parameters and computational cost by 65.02% and 39.78%, respectively, while maintaining the network performance compared to the MobileNetV1 baseline.

Modeling and inversions of acoustic reflection logging imaging using the combined monopole–dipole measurement mode

In this paper, we theoretically and numerically study a combined monopole–dipole measurement mode to show its capability to overcome the issues encountered in conventional single-well imaging, i.e., the low signal-to-noise ratio of the reflections and azimuth ambiguity. First, the azimuth ambiguity, which exists extensively in conventional single-well imaging, is solved with an improved imaging procedure using combined monopole–dipole logging data in addition to conventional logging data. Furthermore, we demonstrate that the direct waves propagating along the boreholes with strong energy, can be effectively eliminated with the proposed combined monopole–dipole measurement mode. The reflections are therefore predominant in the combined monopole–dipole data even before the signals are filtered; thus, the reflections' arrival times in each receiver are identified, which may help minimize the difficulties in filtering conventional logging data. The optimized processing flow of the combined measurement mode logging image is given in this paper. The proposed combined monopole–dipole measurement mode may improve the accuracy of single-well imaging.

SADDLE-POINT BASED SEPARATION OF TOUCHED OBJECTS IN 2-D IMAGE

In many image analysis and processing problems, discriminating the size and shape of each individual object in an aggregate pile projected in an image is an important practice. It is relatively easy to distinguish these features among the objects already separated from each other. The problems will be undoubtedly more complex and of greater challenge if the objects are touched or/and overlapped. This letter presents an algorithm that can be used to separate the touches and overlaps existing in the objects within a 2-D image. The approach is first to convert the gray-scale image to its corresponding binary one and then to the 3-D topographic one using the erosion operations. A template (or mask) is engineered to search the topographic surface for the saddle point, from which the segmenting orientation is determined followed by the desired separating operation. The algorithm is tested on a real image and the running result is adequately satisfying and encouraging.

Blind-restoration-based blind separation method for permuted motion blurred images

A novel single-channel blind separation algorithm for permuted motion blurred images is proposed by using blind restoration in this paper. Both the motion direction and the length of the point spread function （PSF） are estimated by Radon transformation and extrema a detection. Using the estimated blur parameters, the permuted image is restored by performing the L-R blind restoration method. The permutation mixing matrices can be accurately estimated by classifying the ringing effect in the restored image, thereby the source images can be separated. Simulation results show a better separation efficiency for the permuted motion blurred image with various permutation operations. The proposed algorithm indicates a better performance on the robustness against Gaussian noise and lossy JPEG compression.

An improved fast converted-wave imaging method

The conventional fast converted-wave imaging method directly uses backward Pand converted S-wavefield to produce joint images. However, this image is accompanied by strong background noises, because the wavefi elds in all propagation directions contribute to it. Given this issue, we improve the conventional imaging method in the two aspects. First, the amplitude-preserved P-and S-wavef ield are obtained by using an improved space-domain wavef ield separation scheme to decouple the original elastic wavef ield. Second, a convertedwave imaging condition is constructed based on the directional-wavefield separation and only the wavefields propagating in the same directions used for cross-correlation imaging, resulting in effectively eliminating the imaging artifacts of the wavefields with different directions;Complex-wavefi eld extrapolation is adopted to decompose the decoupled P-and S-wavefield into directional-wavefields during backward propagation, this improves the eff iciency of the directional-wavef ield separation. Experiments on synthetic data show that the improved method generates more accurate converted-wave images than the conventional one. Moreover, the improved method has application potential in micro-seismic and passive-source exploration due to its source-independent characteristic.

CMAC Based Color Separation in Printing Images

To over come the drawbacks existing in current measurement methods for detecting and controlling colors in printing process, a new medal for color separation and dot recognition is proposed from a view of digital image processing and patter recognition. In this model, firstly data samples are collected from some color patches by the Fuzzy C-Means （FCM） method; then a classifier based on the Cerebellar Model Articulation Controller （CMAC） is constructed which is used to recognize color pattern of each pixel in a microscopic halftone image. The principle of color separation and the algorithm model are introduced and the experiments show the effectiveness of the CMAC-based classifier as opposed to the BP network.

A Separating Algorithm for Overlapping Cell Images

The cell overlapping and adhesion phenomenon often exists in cell image processing. Separating overlapped cell into single ones is of great important and difficult in cell image quantitative analysis and automatic recognition. In this paper, an algorithm based on concave region extraction and erosion limit has been proposed to judge and separate overlapping cell images. Experimental results show that the proposed algorithm has a good separation effects on analog cell images. Then the method is applying in actual cervical cell image and obtains good separation result.

Dimensionality Reduction Using Optimized Self-Organized Map Technique for Hyperspectral Image Classification

The high dimensionalhyperspectral image classification is a challenging task due to the spectral feature vectors.The high correlation between these features and the noises greatly affects the classification performances.To overcome this,dimensionality reduction techniques are widely used.Traditional image processing applications recently propose numerous deep learning models.However,in hyperspectral image classification,the features of deep learning models are less explored.Thus,for efficient hyperspectral image classification,a depth-wise convolutional neural network is presented in this research work.To handle the dimensionality issue in the classification process,an optimized self-organized map model is employed using a water strider optimization algorithm.The network parameters of the self-organized map are optimized by the water strider optimization which reduces the dimensionality issues and enhances the classification performances.Standard datasets such as Indian Pines and the University of Pavia(UP)are considered for experimental analysis.Existing dimensionality reduction methods like Enhanced Hybrid-Graph Discriminant Learning(EHGDL),local geometric structure Fisher analysis(LGSFA),Discriminant Hyper-Laplacian projection(DHLP),Group-based tensor model(GBTM),and Lower rank tensor approximation(LRTA)methods are compared with proposed optimized SOM model.Results confirm the superior performance of the proposed model of 98.22%accuracy for the Indian pines dataset and 98.21%accuracy for the University of Pavia dataset over the existing maximum likelihood classifier,and Support vector machine(SVM).

基于计算机视觉的岩石裂隙识别表征与软件研制

岩石裂隙特征是评判岩体结构及其完整性的核心指标,也是评估岩石工程安全稳定性的重要因素。针对岩石裂隙识别,采用深度学习方法,通过引入混合注意力机制对Unet模型进行了改进,有效提高了岩石裂隙识别的精度。针对交叉岩石裂隙的分离与特征提取,提出了一种基于迹线方向判定的裂隙分离与表征算法,依据裂隙分离的结果形式,采用重合追踪法或断裂追踪法分离交叉裂隙骨架,继而使用微分累加法、方框法、线性回归法求得裂隙的长度、宽度及倾角等几何特征指标。基于提出的算法,研制了一套具有图形用户界面的岩石裂隙图像智能识别与表征软件系统,实现了从深度学习模型参数选择、模型训练、裂隙识别、量化分析到结果可视化的完整功能。最后对岩石裂隙识别与分离表征算法的性能进行了评判,结果表明,改进Unet模型对复杂分布的裂隙识别效果最好,其总体识别性能要优于其他网络;骨架分离算法对常见类型交叉裂隙能够取得预期结果,表征算法对分离裂隙与交叉裂隙的表征精度高,对实际岩石裂隙图像的应用效果较好。研究成果可为基于计算机视觉的岩石工程试验与岩体结构检测技术研发提供参考依据。

一种基于YOLOX_s的雾天场景目标检测方法

文中提出了一个基于深度可分离卷积和注意力机制的雾天目标检测模型,旨在实现在雾天场景中对目标的快速、准确检测。该模型由去雾模块和检测模块组成,并在训练过程中共同训练。为确保模型在雾天场景中检测的准确性和实时性,在去雾模块方面,采用AODNet对输入图像进行去雾处理,以降低雾对图像中待检测目标的干扰,在检测模块中使用改进后的YOLOX_s模型,输出目标的分类置信度和位置坐标。为提升网络的检测性能,在YOLOX_s基础上采用深度可分离卷积和注意力机制来提高特征提取能力,扩大特征图感受野。所提模型能提高有雾场景中模型的检测精度,且不增加模型参数量和计算量。实验结果表明,所提模型在RTTS数据集和合成有雾目标检测数据集上均表现出色,有效提高了模型在雾天场景中的检测精度。与基准模型相比,平均精度(mAP@50_95)分别提升了1.9%和2.37%。

基于多尺度特征融合的轻量化人脸图像修复算法

针对当前遮挡的人脸图像修复中修复图像质量差和模型参数量大的问题,提出了一种基于多尺度特征融合的改进U-Net的轻量化人脸图像修复模型——LM-UNET。首先,使用深度可分离卷积替换原有卷积,增强模型对不同通道和上下文信息的特征表达能力,实现模型轻量化;其次,在跳跃连接中设计了多尺度特征注意力融合模块,充分融合不同尺度特征的信息,内嵌残差块减少特征间语义差距,提高模型修复准确率;最后,引入了位置注意力模块,增强人脸图像的显著信息,提升模型对人脸位置像素信息的有效提取能力。在基于CK+数据集生成的遮挡人脸数据集MFD上对该算法进行训练、验证和测试,修复后的图像的峰值信噪比(PSNR)达到30.49dB,结构相似性(SSIM)达到96.85%,与其他模型的对比实验结果表明,该模型对存在遮挡的人脸修复图像质量和视觉效果更好。

不同MR设备CSE MR序列定量骨髓质子密度脂肪分数的一致性分析

目的探讨以单体素氢质子磁共振波谱成像(MRS)为参照,验证不同厂家3T化学位移编码水/脂分离MR序列(CSE-MRI)定量腰椎骨髓质子密度脂肪分数(PDFF)的可重复性与一致性。方法选取38例受试者进行腰椎MRS(PDFF_(MRS))、联影CSE-MRI(PDFF_(FACT))与西门子CSE-MRI(PDFF_(Dixion))扫描。以均方根误差(RMSE)和均方根变异系数(RMS-CV)评价CSE-MRI定量PDFF的精密度及可重复性。使用组内相关系数(ICC)评估不同观察者及观察者内测定PDFF值的一致性。采用Linear regression及Bland-Altman分析评价不同CSE-MRI序列与MRS定量PDFF的可代替性及不同公司间CSE-MRI序列定量PDFF的一致性。结果PDFF_(MRS)、PDFF_(FACT)及PDFF_(Dixion)定量骨髓PDFF值分别为58.2%(38.7%,67.0%)、56.7%(37.2%,66.6%)、56.3%(36.8%,65.4%)。PDFF_(FACT)与PDFF_(Dixion)RMSE分别为1.48%、1.57%,RMS-CV分别为2.37%、2.25%,可重复性良好。观察者内及观察者间测定PDFF具有实质性的一致性,ICC均大于0.99。MRS、FACT及Dixon VIBE序列定量骨髓PDFF之间的线性回归分析,R^(2)=0.985~0.989,差异有统计学意义(P<0.001)。Bland-Altman分析PDFF_(FACT)与PDFF_(MRS)差值97.4%的点落在95%可信区间内,PDFFDixon与PDFF_(MRS)差值92.1%的点落在95%可信区间内,PDFF_(FACT)与PDFFDixon差值92.1%的点落在95%可信区间内,显示了良好的一致性。结论联影与西门子CSE MR序列定量腰椎PDFF可重复性良好,其准确性可等同于MRS。

改进YOLACT的服装图像实例分割方法

针对服装图像实例分割精度与速度较低的问题,提出一种基于改进YOLACT的服装图像实例分割方法。以YOLACT为基础模型,首先在ResNet101网络中采用深度可分离卷积代替传统卷积,减少模型计算量和模型参数,加快模型速度;然后,在模板生成网络后引入高效通道注意力模块,优化输出特征,捕获服装图像的跨通道交互信息,加强对掩膜分支的特征提取能力;最后,训练过程采用LeakyReLU激活函数,避免反向传播时权值信息得不到及时更新,提升模型对服装图像负值特征信息的提取能力。结果表明:与原模型相比,所提方法能有效减少模型参数量,在提升速度的同时提高了精度,其速度提升了4.82帧/s,平均精度提升了5.4%。

基于改进SE-Net和深度可分离残差的高光谱图像分类

针对目前常见的用于高光谱图像分类的卷积神经网络参数数量多,训练时间长,对样本数量依赖性大的问题,提出一种适用于有限训练样本条件下基于改进压缩激活网络和深度可分离残差的分类网络MDSR&SE-Net.首先使用主成分分析对原始高光谱图像进行通道降维,然后通过三维卷积神经网络连接多特征残差结构,同时嵌入改进的SE模块提取高光谱图像的空间和光谱细节特征,最后将提取到的特征数据输入Softmax分类器激活分类.为了使网络更加轻量,通过在残差结构中使用深度可分离卷积和引入全局平均池化减少参数数量.实验结果显示,使用有限训练样本在三种常见高光谱数据集上总体分类精度均达到99%以上.

基于改进ResNet50的钨矿石双能X射线图像分选方法

文中提出一种基于深度扩张可分离卷积和注意力机制的残差网络模型(DWAtt-ResNet),通过实验对比表明,该模型在钨矿石双能X射线图像数据集上准确率、F1分数、AUC值和AP值均优于ConvNeXt、DenseNet121和EfficientNet_b4等主流的图像分类模型。通过消融实验表明,该模型准确率达到87.4%,计算量为2.7GFLOPs,参数量为16.95M,相比ResNet50准确率提高3%,计算量降低1.42 GFLOPs,参数量降低6.56M,准确率提升的同时,效率大幅提升,更适合工业生产的矿石快速分拣需求。

改进MetaFormer的轻量化模型设计

为了提升传统CNN的远距离特征提取能力,将深度可分离卷积操作嵌入到改进MetaFormer架构中,添加了通道混洗操作,提出了一种创新的轻量化CNN网络模型:ViTNet。通过这种融合策略,不仅保留了ViTs的灵活性和扩展性,还增强了CNN模型的图像特征提取能力。ViTNet的轻量化优势,使其能够高效运行于计算资源有限的设备上。实验结果表明:在cifar10上,ViTNet-1.0×比MobileNetV2提高了1.8%准确率,延迟降低了32%,表现出良好的竞争力。

面向深部地热岩体的弹性波逆时偏移成像方法

地热能作为一种清洁能源正受到全世界的日益关注,因此对于深部地热岩体的高精度成像格外重要。传统的弹性波逆时偏移方法是利用耦合纵横波直接成像,从而得到耦合波场的成像结果;然而这种方法可能会出现串扰假象,导致成像结果不够清晰;为了解决这个问题,采用基于解耦的弹性波方程,实现纵横波场的分离;通过利用内积成像条件,对两个典型的干热岩模型进行数值测试。结果表明,相较于耦合波场成像结果,基于解耦方程分离得到的弹性波成像剖面具有更清晰的同相轴,深部能量更加均衡,该方法能够实现对深部地热岩体的高质量成像。