基于改进Deformable-DETR的水下图像目标检测方法

针对由于水下复杂环境造成的目标检测效果较差、检测精度较低的问题,基于Deformable-DETR算法提出一种改进的水下目标检测算法Deformable-DETR-DA。使用空间注意力模块结合标准Transformer块设计了一个用于增加模型深度的深度特征金字塔(deep feature pyramid networks,DFPN)模块,将其嵌入到模型中提高模型对深层纹理信息的提取能力。使用注意力引导的方式对原模型中编码器部分进行改进,加强了对特征信息的聚合能力,提高了模型在复杂环境下的检测能力。针对URPC数据集,模型各交并比尺度的平均准确度(average precision,AP)为39.5%,相比原模型提升1%,与一些DETR(detection transformer)类的模型相比,不同目标尺度的平均准确度均有1%~4%左右的提高,表明改进的模型能够很好解决复杂环境的水下目标检测的问题。本文提出的模型可作为其他水下目标检测模型设计的参考。

基于改进Deformable DETR模型的多源局部放电识别方法及其应用

基于图像的局部放电识别方法大部分仅对单源局部放电谱图有效,无法识别多源局部放电谱图。为实现对多源局部放电谱图的识别,该文提出一种基于Transformer架构的局部放电Deformable DETR目标检测模型,收集典型单源局部放电和多源局部放电数据,生成局部放电相位角解析和极坐标相位分布解析谱图数据集。在Deformable DETR模型中引入去噪训练任务和贝叶斯优化算法,优化了局部放电目标检测模型;编写局部放电谱图采集和识别程序,并使用优化后的局部放电Deformable DETR模型对单源和多源局部放电谱图进行识别。结果表明:局部放电Deformable DETR模型不仅可有效识别出单源和多源局部放电的类型,而且大幅提升了局部放电类型识别的收敛速度和精度等性能。在对真实绝缘缺陷电动机的局部放电谱图识别中,局部放电Deformable DETR模型的识别准确率达到91%,证明该模型在实际应用中的有效性。

基于改进Deformable DETR的无人机视频流车辆目标检测算法

针对无人机视频流检测中小目标数量多、因图像传输质量较低而导致的上下文语义信息不充分、传统算法融合特征推理速度慢、数据集类别样本不均衡导致的训练效果差等问题,提出一种基于改进Deformable DETR的无人机视频流车辆目标检测算法。在模型结构方面,该算法设计了跨尺度特征融合模块以增大感受野,提升小目标检测能力,并采用针对object_query的挤压-激励模块提升关键目标的响应值,减少重要目标的漏检与错检率;在数据处理方面,使用了在线困难样本挖掘技术,改善数据集中类别样本分布不均的问题。在UAVDT数据集上进行了实验,实验结果表明,改进后的算法相较于基线算法在平均检测精度上提升了1.5%,在小目标检测精度上提升了0.8%,并在保持参数量较少增长的情况下,维持了原有的检测速度。

Deformable Catalytic Material Derived from Mechanical Flexibility for Hydrogen Evolution Reaction

Deformable catalytic material with excellent flexible structure is a new type of catalyst that has been applied in various chemical reactions,especially electrocatalytic hydrogen evolution reaction(HER).In recent years,deformable catalysts for HER have made great progress and would become a research hotspot.The catalytic activities of deformable catalysts could be adjustable by the strain engineering and surface reconfiguration.The surface curvature of flexible catalytic materials is closely related to the electrocatalytic HER properties.Here,firstly,we systematically summarized self-adaptive catalytic performance of deformable catalysts and various micro–nanostructures evolution in catalytic HER process.Secondly,a series of strategies to design highly active catalysts based on the mechanical flexibility of lowdimensional nanomaterials were summarized.Last but not least,we presented the challenges and prospects of the study of flexible and deformable micro–nanostructures of electrocatalysts,which would further deepen the understanding of catalytic mechanisms of deformable HER catalyst.

基于Deformable DETR的红外图像目标检测方法研究

基于Transformer架构的DETR系列网络在计算机视觉目标检测领域不断刷新目标检测的精度与速度。然而,基于红外图像的非合作目标检测的应用环境复杂,而且红外图像质量较差。针对该问题,提出了一种新的以Deformable DETR算法为基线的具有高检测精度的目标检测算法:首先设计了对红外图像进行图像增强处理的图像增强模块CLAHE-GB,并将其与Deformable DETR进行有机结合;然后在大型通用数据集上进行预训练;最后引入数据增强和迁移学习方法在自制的空中飞行物小型红外图像数据集中对检测头网络参数进行再训练,并对结果进行分析。结果表明:所提出的算法对红外图像数据具有较好的图像增强效果和检测精度。

Modeling Near-Field Impulsive Waves Generated by Deformable Landslide Using the HBP Model Based on the SPH Method

Landslide-generated impulsive waves(LGWs)in reservoir areas can seriously threaten waterway safety as well as hu-man life and properties around the two side slopes.The risk reduction and mitigation of such a hazard require the accurate prediction of near-field wave characteristics,such as wave amplitude and run-up.However,near-field LGW involves complicated fluid-solid interactions.Furthermore,the wave characteristics are closely related to various aspects,including the geometry and physical features of the slide,river,and body of water.However,the empirical or analytical methods used for rough estimation cannot derive accurate results,especially for deformable landslides,due to their significant geometry changes during the sliding process.In this study,the near-field waves generated by deformable landslides were simulated by smoothed particle hydrodynamics(SPH)based on multi-phase flow.The deformable landslides were generalized as a kind of viscous flow by adopting the Herschel-Bulkley-Papanastasiou(HBP)-based nonNewtonian rheology model.The HBP model is capable of producing deformable landslide dynamics even though the high-speed sliding process is involved.In this study,an idealized experiment case originating from Lituya LGW and a practical case of Gongjiafang LGW were reproduced for verification and demonstration.The simulation results of both cases show satisfactory consistency with the experiment/investigation data in terms of landslide movement and near-field impulsive wave characteristics,thus indicating the applicability and accuracy of the proposed method.Finally,the effects of the HBP model’s rheological parameters on the landslide dynamics and near-field wave characteristics are discussed,providing a parameter calibration method along with sug-gestions for further applications.

Study of deep transportation and plugging performance of deformable gel particles in porous media

Deformable gel particles(DGPs) possess the capability of deep profile control and flooding. However, the deep migration behavior and plugging mechanism along their path remain unclear. Breakage, an inevitable phenomenon during particle migration, significantly impacts the deep plugging effect. Due to the complexity of the process, few studies have been conducted on this subject. In this paper, we conducted DGP flow experiments using a physical model of a multi-point sandpack under various injection rates and particle sizes. Particle size and concentration tests were performed at each measurement point to investigate the transportation behavior of particles in the deep part of the reservoir. The residual resistance coefficient and concentration changes along the porous media were combined to analyze the plugging performance of DGPs. Furthermore, the particle breakage along their path was revealed by analyzing the changes in particle size along the way. A mathematical model of breakage and concentration changes along the path was established. The results showed that the passage after breakage is a significant migration behavior of particles in porous media. The particles were reduced to less than half of their initial size at the front of the porous media. Breakage is an essential reason for the continuous decreases in particle concentration, size, and residual resistance coefficient. However, the particles can remain in porous media after breakage and play a significant role in deep plugging. Higher injection rates or larger particle sizes resulted in faster breakage along the injection direction, higher degrees of breakage, and faster decreases in residual resistance coefficient along the path. These conditions also led to a weaker deep plugging ability. Smaller particles were more evenly retained along the path, but more particles flowed out of the porous media, resulting in a poor deep plugging effect. The particle size is a function of particle size before injection, transport distance, and different injection parameters(injection rate or the diameter ratio of DGP to throat). Likewise, the particle concentration is a function of initial concentration, transport distance, and different injection parameters. These models can be utilized to optimize particle injection parameters, thereby achieving the goal of fine-tuning oil displacement.

A Deformable Network with Attention Mechanism for Retinal Vessel Segmentation

The intensive application of deep learning in medical image processing has facilitated the advancement of automatic retinal vessel segmentation research.To overcome the limitation that traditional U-shaped vessel segmentation networks fail to extract features in fundus image sufficiently,we propose a novel network(DSeU-net)based on deformable convolution and squeeze excitation residual module.The deformable convolution is utilized to dynamically adjust the receptive field for the feature extraction of retinal vessel.And the squeeze excitation residual module is used to scale the weights of the low-level features so that the network learns the complex relationships of the different feature layers efficiently.We validate the DSeU-net on three public retinal vessel segmentation datasets including DRIVE,CHASEDB1,and STARE,and the experimental results demonstrate the satisfactory segmentation performance of the network.

High temperature deformation and recrystallization behavior of magnesium bicrystals with 90°<1010>and 90°<1120>tilt grain boundaries

The deformation mechanisms and dynamic recrystallization(DRX)behavior of specifically grown bicrystals with a symmetric 90°<1010>and 90°<1120>tilt grain boundary,respectively,were investigated under deformation in plane strain compression at 200℃and 400℃.The microstructures were analyzed by panoramic optical microscopy and large-area electron backscatter diffraction(EBSD)orientation mapping.The analysis employed a meticulous approach utilizing hundreds of individual,small EBSD maps with a small step size that were stitched together to provide comprehensive access to orientation and misorientation data on a macroscopic scale.Basal slip primarily governed the early stages of deformation at the two temperatures,and the resulting shear induced lattice rotation around the transverse direction(TD)of the sample.The existence of the grain boundary gave rise to dislocation pile-up in its vicinity,leading to much larger TD-lattice rotations within the boundary region compared to the bulk.With increasing temperature,the deformation was generally more uniform towards the bulk due to enhanced dislocation mobility and more uniform stress distribution.Dynamic recrystallization at 200℃was initiated in{1011}-compression twins at strains of 40%and higher.At 400℃,DRX consumed the entire grain boundary region and gradually replaced the deformed microstructure with progressing deformation.The recrystallized grains displayed characteristic orientations,such that their c-axes were perpendicular to the TD and additionally scattered between 0°and 60°from the loading axis.These recrystallized grains displayed mutual rotations of up to 30°around the c-axes of the initial grains,forming a discernible basal fiber texture component,prominently visible in the{1120}pole figure.It is noteworthy that the deformation and DRX behaviors of the two analyzed bicrystals exhibited marginal variations in response to strain and deformation temperature.

基于Deformable DETR的自然场景任意形状文本检测

自然场景下的文本区域形状复杂多变,直接使用轮廓坐标描述文本区域会使得建模不充分,导致文本检测准确性低。针对自然场景下文本区域不规则的问题,提出了一种基于Deformable DETR的任意形状文本检测模型,不同于传统的直接预测轮廓点的方法,使用B-样条对文字区域进行建模使得文本轮廓平滑精确的同时减少了需要预测的参数。提出的文本检测模型无需手工设计锚点、区域建议等组件,极大地简化了模型设计并提高了通用性。提出的模型在无需额外数据集的情况下在任意形状文本数据集CTW1500和Total-Text上的平均精度(F值)分别达到了85.4%和85.0%,证明了模型的有效性。

Contact detumbling toward a nutating target through deformable effectors and prescribed performance controller

Detumbling operation toward a rotating target with nutation is meaningful for debris removal but challenging. In this study, a deformable end-effector is first designed based on the requirements for contacting the nutating target. A dual-arm robotic system installed with the deformable end-effectors is modeled and the movement of the end-tips is analyzed. The complex operation of the contact toward a nutating target places strict requirements on control accuracy and controller robustness. Thus, an improvement of the tracking error transformation is proposed and an adaptive sliding mode controller with prescribed performance is designed to guarantee the fast and precise motion of the effector during the contact detumbling.Finally, by employing the proposed effector and the controller,numerical simulations are carried out to verify the effectiveness and efficiency of the contact detumbling toward a nutating target.

Fuzzy Proportional Integral Derivative control of a voice coil actuator system for adaptive deformable mirrors

Research on adaptive deformable mirror technology for voice coil actuators(VCAs)is an important trend in the development of large ground-based telescopes.A voice coil adaptive deformable mirror contains a large number of actuators,and there are problems with structural coupling and large temperature increases in their internal coils.Additionally,parameters of the traditional proportional integral derivative(PID)control cannot be adjusted in real-time to adapt to system changes.These problems can be addressed by introducing fuzzy control methods.A table lookup method is adopted to replace real-time calculations of the regular fuzzy controller during the control process,and a prototype platform has been established to verify the effectiveness and robustness of this process.Experimental tests compare the control performance of traditional and fuzzy proportional integral derivative(Fuzzy-PID)controllers,showing that,in system step response tests,the fuzzy control system reduces rise time by 20.25%,decreases overshoot by 78.24%,and shortens settling time by 67.59%.In disturbance rejection experiments,fuzzy control achieves a 46.09%reduction in the maximum deviation,indicating stronger robustness.The Fuzzy-PID controller,based on table lookup,outperforms the standard controller significantly,showing excellent potential for enhancing the dynamic performance and disturbance rejection capability of the voice coil motor actuator system.

GAN-DIRNet:A Novel Deformable Image Registration Approach for Multimodal Histological Images

Multi-modal histological image registration tasks pose significant challenges due to tissue staining operations causing partial loss and folding of tissue.Convolutional neural network(CNN)and generative adversarial network(GAN)are pivotal inmedical image registration.However,existing methods often struggle with severe interference and deformation,as seen in histological images of conditions like Cushing’s disease.We argue that the failure of current approaches lies in underutilizing the feature extraction capability of the discriminator inGAN.In this study,we propose a novel multi-modal registration approach GAN-DIRNet based on GAN for deformable histological image registration.To begin with,the discriminators of two GANs are embedded as a new dual parallel feature extraction module into the unsupervised registration networks,characterized by implicitly extracting feature descriptors of specific modalities.Additionally,modal feature description layers and registration layers collaborate in unsupervised optimization,facilitating faster convergence and more precise results.Lastly,experiments and evaluations were conducted on the registration of the Mixed National Institute of Standards and Technology database(MNIST),eight publicly available datasets of histological sections and the Clustering-Registration-Classification-Segmentation(CRCS)dataset on the Cushing’s disease.Experimental results demonstrate that our proposed GAN-DIRNet method surpasses existing approaches like DIRNet in terms of both registration accuracy and time efficiency,while also exhibiting robustness across different image types.

Multi-Layer Feature Extraction with Deformable Convolution for Fabric Defect Detection

In the textile industry,the presence of defects on the surface of fabric is an essential factor in determining fabric quality.Therefore,identifying fabric defects forms a crucial part of the fabric production process.Traditional fabric defect detection algorithms can only detect specific materials and specific fabric defect types;in addition,their detection efficiency is low,and their detection results are relatively poor.Deep learning-based methods have many advantages in the field of fabric defect detection,however,such methods are less effective in identifying multiscale fabric defects and defects with complex shapes.Therefore,we propose an effective algorithm,namely multilayer feature extraction combined with deformable convolution(MFDC),for fabric defect detection.In MFDC,multi-layer feature extraction is used to fuse the underlying location features with high-level classification features through a horizontally connected top-down architecture to improve the detection of multi-scale fabric defects.On this basis,a deformable convolution is added to solve the problem of the algorithm’s weak detection ability of irregularly shaped fabric defects.In this approach,Roi Align and Cascade-RCNN are integrated to enhance the adaptability of the algorithm in materials with complex patterned backgrounds.The experimental results show that the MFDC algorithm can achieve good detection results for both multi-scale fabric defects and defects with complex shapes,at the expense of a small increase in detection time.

Deformable DETR大模型在新农业的应用研究——以宁夏固原娃娃菜病虫害检测与识别为例

宁夏固原市露地娃娃菜种植规模逐年增加,市场需求广泛,促进了农村经济的发展。但传统的娃娃菜病虫害检测方法由于耗时长、效率低、成本高,已不能满足快速发展农业生产的需要。文中以固原市农科院露地娃娃菜为研究对象,基于Deformable DETR模型对霜霉病、软腐病、菜青虫3种主要病虫害进行检测与识别,不仅提高了病虫害检测的效率和准确性,且降低了管理成本,有助于固原市农业生产者实现更加高效、科学的病虫害防治。

Deformable分数阶Logistic微分方程的解

该文基于Deformable分数阶微积分定义框架,讨论一类分数阶Logistic微分方程初值问题,得到了解析解,并给出相应的数值模拟结果.

Size-dependent thermomechanical vibration characteristics of rotating pre-twisted functionally graded shear deformable microbeams

A three-dimensional(3D)thermomechanical vibration model is developed for rotating pre-twisted functionally graded(FG)microbeams according to the refined shear deformation theory(RSDT)and the modified couple stress theory(MCST).The material properties are assumed to follow a power-law distribution along the chordwise direction.The model introduces one axial stretching variable and four transverse deflection variables including two pure bending components and two pure shear ones.The complex modal analysis and assumed mode methods are used to solve the governing equations of motion under different boundary conditions(BCs).Several examples are presented to verify the effectiveness of the developed model.By coupling the slenderness ratio,gradient index,rotation speed,and size effect with the pre-twisted angle,the effects of these factors on the thermomechanical vibration of the microbeam with different BCs are investigated.It is found that with the increase in the pre-twisted angle,the critical slenderness ratio and gradient index corresponding to the thermal instability of the microbeam increase,while the critical material length scale parameter(MLSP)and rotation speed decrease.The sensitivity of the fundamental frequency to temperature increases with the increasing slenderness ratio and gradient index,and decreases with the other increasing parameters.Moreover,the size effect can suppress the dynamic stiffening effect and enhance the Coriolis effect.Finally,the mode transition is quantitatively demonstrated by a modal assurance criterion(MAC).

Machine Learning Techniques in Predicting Hot Deformation Behavior of Metallic Materials

In engineering practice,it is often necessary to determine functional relationships between dependent and independent variables.These relationships can be highly nonlinear,and classical regression approaches cannot always provide sufficiently reliable solutions.Nevertheless,Machine Learning(ML)techniques,which offer advanced regression tools to address complicated engineering issues,have been developed and widely explored.This study investigates the selected ML techniques to evaluate their suitability for application in the hot deformation behavior of metallic materials.The ML-based regression methods of Artificial Neural Networks(ANNs),Support Vector Machine(SVM),Decision Tree Regression(DTR),and Gaussian Process Regression(GPR)are applied to mathematically describe hot flow stress curve datasets acquired experimentally for a medium-carbon steel.Although the GPR method has not been used for such a regression task before,the results showed that its performance is the most favorable and practically unrivaled;neither the ANN method nor the other studied ML techniques provide such precise results of the solved regression analysis.

TOOTH CROWN DEFORMABLE DESIGN BASED ON DIRECT MANIPULATION OF FFD

A flexible and adaptable design method for the tooth crown is presented based on the direct manipulation of free-form deformation （FFD）. The correct shape of the tooth crown can be obtained by adding, modifying, and deleting constraint points or load points based on stretching and compressing operation. Finally, an example is given to illustrate the method to be efficient.

基于伪磁梯度张量混合方向tilt梯度的磁源定位方法及其应用:以内蒙古塔木素铀矿床为例

斜磁化磁源的准确定位是磁法勘探有效圈定场源分布及划分断裂构造的重要依据.但受磁化方向影响,总磁异常与地质体分布并无明显对应关系.为此,本文从总磁异常垂向积分出发,推导出了伪磁梯度张量,定义了伪磁梯度张量的混合方向tilt梯度及混合方向解析信号振幅,提出了基于伪磁梯度张量混合方向tilt梯度水平导数组合的场源定位方法.模型试验表明,相对于解析信号振幅、tilt梯度的总水平导数及方向tilt梯度的总水平导数模这3种受磁化方向影响较小的方法,本文提出的磁源定位方法能够更有效地反映出不同磁化方向的磁源位置,具有更强的识别能力和定位精度.在内蒙古塔木素铀矿床的航磁资料应用中,新方法较好地展示出了已知断裂带的展布及出露岩浆岩的分布,同时有效地提取了塔木素铀矿床上方的弱磁异常信息,处理结果为研究区圈定隐伏岩体、划分隐伏断裂及扩大铀矿找矿靶区提供了依据.