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。

基于改进的Transformer细粒度图像识别算法研究

针对细粒度图像识别存在类间差异小、难以区分等问题,本文通过提升网络对图像细节特征的表达能力,来改善这一问题。为此,设计了一种基于改进的Transformer细粒度识别算法。首先,可变形卷积令牌嵌入通过自适应调整采样点的位置,来改变卷积操作范围及其卷积核的形状,从而增强网络模型对空间信息的感知能力,以获取更为精准的空间信息;其次,高效相关通道注意力机制通过对通道的自动选择,将通道注意力的计算从通道相邻转换成语义相似,来捕获语义相似的通道信息。而精准的空间信息和语义相似的通道信息将有效提升网络模型局部特征感知能力。实验结果表明,与基线算法相比,本文方法在CUB-200-2011、StanfordCars和StanfordDogs三个数据集上的识别结果分别提升了1.5%、2.4%、1.5%。结果表明,本文提出的方法通过提升细粒度图像细节特征的表达能力,从而有效提高了细粒度图像识别的有效性。

基于Transformer网络多模态融合的密集视频描述方法

针对目前的密集视频描述模型大多使用两阶段的方法存在效率较低、忽略音频及语义信息,描述结果不全面的问题。提出了一种基于Transformer网络多模态和语义信息融合的密集视频描述方法。提取自适应R(2+1)D网络提取视觉特征,设计了语义探测器生成语义信息,加入音频特征进行补充,建立了多尺度可变形注意力模块,应用并行的预测头,加快模型收敛速度,提高模型精度。实验结果表明:模型在2个基准数据集上性能均有很好的表现,评价指标BLEU4上达到了2.17。

基于边缘引导和动态可变形Transformer的遥感图像变化检测

卷积神经网络(Convolutional Neural Network,CNN)和Transformer的混合架构能够有效建模图像的局部与全局特征,已成为遥感图像变化检测任务的主流网络.然而这类网络仍面临着一些挑战. CNN分支中的卷积和池化运算通常会抑制遥感图像中的高频信息,降低目标边界的精度;此外,Transformer分支对图像像素进行等同长程依赖关系建模,忽略了变化目标的形状及语义关联信息,导致网络对变化目标特征的表达不足.为解决上述问题,提出了基于边缘引导和动态可变形Transformer的遥感图像变化检测网络.在CNN分支中设计了边缘信息引导模块,利用高频信息增强目标区域的边缘信息,从而改善变化目标的轮廓精度.同时设计了一种新颖的动态可变形Transformer,能够自适应地匹配形状不同的变化目标,选择与变化相关的特征建模长程依赖关系,以提高网络的特征表达能力.实验结果表明,提出的方法在三个公开数据集LEVIR-CD、CDD和DSIFN-CD上显著提高了检测精度,在变化目标的边界精度和内部完整性方面都明显优于当前的主流网络.

结合Transformer和动态特征融合的低照度目标检测

针对现有低照度目标检测算法模型参数量与计算量大、检测实时性差、难以应用于移动设备等问题,提出一种基于YOLOv8的改进轻量模型DarkYOLOv8的低照度目标检测方法。使用MobileNet v2替换YOLOv8的主干网络,增加模型的特征提取能力;使用Transformer注意力机制,获取图像的全局信息,并且基于目标标记信息作为标签训练Transformer模块参数,增强目标区域内的权重,从而提高模型在低照度条件下提取目标特征的能力;对颈部网络使用动态特征融合注意力模块(dynamic feature fusion attention,DFFA),动态融合浅层和深层特征,同时使用YOLOv8X算法+CBAM对DFFA模块中CBAM空间注意力权重进行监督训练。实验结果表明,在ExDark数据集上,DarkYOLOv8在GFLOPs仅为8.53的情况下mAP50指标达到70.1%,相比YOLOv8n提高了3.9个百分点。

Pre-training transformer with dual-branch context content module for table detection in document images

Background Document images such as statistical reports and scientific journals are widely used in information technology.Accurate detection of table areas in document images is an essential prerequisite for tasks such as information extraction.However,because of the diversity in the shapes and sizes of tables,existing table detection methods adapted from general object detection algorithms,have not yet achieved satisfactory results.Incorrect detection results might lead to the loss of critical information.Methods Therefore,we propose a novel end-to-end trainable deep network combined with a self-supervised pretraining transformer for feature extraction to minimize incorrect detections.To better deal with table areas of different shapes and sizes,we added a dualbranch context content attention module(DCCAM)to high-dimensional features to extract context content information,thereby enhancing the network's ability to learn shape features.For feature fusion at different scales,we replaced the original 3×3 convolution with a multilayer residual module,which contains enhanced gradient flow information to improve the feature representation and extraction capability.Results We evaluated our method on public document datasets and compared it with previous methods,which achieved state-of-the-art results in terms of evaluation metrics such as recall and F1-score.https://github.com/Yong Z-Lee/TD-DCCAM.

基于LSTM与Transformer的大坝变形预测研究

变形是大坝运行过程中受内外荷载作用下其状态的直观表现,构建高精度的变形预测模型对大坝安全预警与运行状态评估有着重要的意义。针对已有的大坝位移模型训练时间长,预测模型预测精度和泛化能力一般,无法满足大坝位移中短期准确预测的问题,耦合长短时记忆网络(LSTM)与Transformer框架,引入改进的粒子群优化算法(IPSO)进行优化,建立了IPSO-LSTM-Transformer(ILT)大坝变形预测模型。以紧水滩拱坝正垂线11-1测点为例,选取6150组变形时间序列数据进行分析与预测。研究结果表明,模型预测精度会随着预测期的增大而出现一定程度的下降,但在预测步长10以内均具有良好的预测能力;与传统粒子群优化算法相比,ILT模型显著提升了模型的寻优精度和收敛速度;与RNN、LSTM、IPSO-Transformer神经网络模型单步与多步预测结果对比,ILT模型具有更高的精度和更好的稳定性,即使在训练数据较少时也能保证较好的预测效果。研究成果为实现运行期大坝位移的中短期精确预测提供了新的技术手段。

Effect of Austenite Deformation on Continuous Cooling Transformation Microstructures for 22CrSH Gear Steel

The effect of compressive deformation of austenite on continuous cooling transformation microstructures for 22CrSH gear steel has been investigated using a Gleeble 1500 thermal simulator. The experimental results show that the deformation of austenite promotes the formation of proeutectoid ferrite and pearlite, and leads to the increase of critical cooling rate of proeutectoid ferrite plus pearlite microstructure. The grain boundary allotriomorphic ferrite occupies the austenite grain surfaces when the prior deformation takes place or the cooling rate is decreased, which causes a transition from bainite to acicular ferrite. The deformation enhances the stability of transformation from austenite to acicular ferrite, which results in an increase of M/A constituent.

MECHANICAL STABILIZATION OF DEFORMED AUSTENITEDURING CONTINUOUS COOLING TRANSFORMATION IN AC-Mn-Cr-Ni-Mo PLASTIC DIE STEEL

The influence of prior austenite deformed at different temperature on the subsequent continuous cooling bainitic transformation has been investigated in an C-Ma-Cr-Ni-Mo plastic die steel. The results show that the prior deformation in low temperature region of austenite retards significantly the bainitic transformation. For the same continuous cooling schedule, as austenite deformed at lower temperature, the quantity of the classical sheaf-like bainite becomes less. The present results show that severe deformation leads to mechanical stabilization of austenite and causes the difficulty of bainitic ferrite propagation into the austenite.

INFLUENCE OF HOT DEFORMATION ON CONTINUOUSCOOLING BAINITIC TRANSFORMATION IN A LOWCARBON STEEL

The influence of hot deformation conditions on continuous cooling bainitic transformation has been investigated for a low carbon microalloyed steel. The CCT diagrams show that deformation in unrecrpstallized austcnite ation can accelerate transformation process. Bainitic transformation in intermediate transformation temperature region is prominent, and the proeutectoid polygonal ferrite transformation at evelated high temperature is suppressed. According to optical and TEM analyses, low carbon bainitic ferrite is characterized by granular and lathlike ferrite, based on the cooling rate and deformation conditions. For nondeformation, groaps of coarse parallel ferrite lath form from the prior austenite grain boundaries with the same crystallographic orientation. For heavy deformaton, cell structure within the austenite grains due to the high dislocation density formed, which provides more nucleation sites for bainite ferrite. So deformation can discontinue the growth of ferrite laths and decrease the length of ferrite laths.

DEFORMATION ENHANCED FERRITE TRANSFORMATION IN PLAIN LOW CARBON STEEL

The microstructure evolution during deformation enhanced transformation of undercooled austenite of a plain low carbon steel has been investigated by means of hot compression simulation experiment under various conditions of strain rate, deformation temperature and strain. The effect of austenite grain size on the strain enhanced ferrite transformation has been studied. The ferrite dynamic recrystallization involved in successive hot deformation has been explored.

Geochronology Constraints on Transformation Age from Ductile to Brittle Deformation of the Shangma Fault and Its Tectonic Significance,Dabieshan,Central China

By a detailed investigation of geometry and kinematics of the Shangma （商麻） fault in Dabieshan （大别山）, three different crust levels of extension movement have been recognized in sequence from the deep to the shallow：① low-angle ductile detachment shearing with top to the NW; ② low-angle normal fault with top to the NW or NWW in brittle or brittle-ductile transition domain; ③ high-angle brittle normal fault with top to the W or NWW. Two samples were chosen for zircon U-Pb age dating to constrain the activity age of the Shangma fault. A bedding intrusive granitoid pegmatite vein that is parallel to the foliation of the low-angle ductile detachment shear zone of the country rock exhibits a lotus-joint type of boudinage deformation, showing syn-tectonic emplacing at the end of the ductile deformation period and deformation in the brittle-ductile transition domain. The zircon U-Pb dating of this granitoid pegmatite vein gives an age of （125.9±4.2） Ma, which expresses the extension in the brittle-ductile transition domain of the Shangma fault. The other sample, which is collected from a granite pluton cutting the foliation of the low-angle ductile detachment shear zone, gives a zircon U-Pb age of （118.8±4.1） Ma, constraining the end of the ductile detachment shearing. Then the transformation age from ductile to brittle deformation can be constrained between 126-119 Ma. Combined with the previous researches, the formation of the Luotian （罗田） dome, which is locatedto the east of the Shangma fault, can be constrained during 150-126 Ma. This study gives a new time constraint to the evolution of the Dabie orogenic belt.

Effect of V and V-N Microalloying on Deformation-Induced Ferrite Transformation in Low Carbon Steels

Deformation-induced ferrite transformation （DIFT） has been proved to be an effective approach to refine ferrite grains. This paper shows that the ferrite grains can further be refined through combination of DIFT and V or V-N microalloying. Vanadium dissolved in γ matrix restrains DIFT. During deformation, vanadium carbonitrides rapidly precipitate due to strain-induced precipitation, which causes decrease in vanadium dissolved in matrix and indirectly accelerates DIFT. Under heavy deformation, deformation induced ferrite （DIF） grains in V microalloyed steel were finer than those in V free steel. The more V added to steel, the finer DIF grains obtained. Moreover, the addition of N to V microalloyed steels can remarkably accelerate precipitation of V, and then promote DIFT. However, DIF grains in V-N microalloyed steel easily coarsen.

HOT DEFORMATION AND MARTENSITIC TRANSFORMATION BEHAVIORS OF Fe-32%Ni ALLOY

The Hot deformation and martensitic transformation behaviors of Fe-32%Ni alloy was investigated by measurements of electrical resistance and X-ray diffraction. With the increase in strain, the austenite goes through from the work-hardened to the partial dynamcally re-crystallized and then to the completed dynamically re-crystallized. The martensitic transformation characteristics depend on the austenite states. The work-hardening in small strain is helpful to martensitic transformation due to the low dislocation density and little lattice distortion, while the high dislocation density and severe lattice distortion by the increase in strain will hinder the martensitic nucleation. Once dynamic re-crystallization （ DRX ） takes place, the martensitic transformation will be enhanced again, which is related to the heterogeneous dynamic substructures. The growing DRX grain can enhance the martensitic nucleation due to the low dislocation density near its grain boundary.

Effect of silicon and prior deformation of austenite on isothermal transformation in low carbon steels

Isothermal transformation （TTT） behavior of the low carbon steels with two Si contents （0.50 wt pct and 1.35 wt pct） was investigated with and without the prior deformation. The results show that Si and the prior deformation of the austenite have significant effects on the transformation of the ferrite and bainite. The addition of Si refines the ferrite grains, accelerates the polygonal ferrite transformation and the formation of M/A constituents, leading to the improvement of the strength. The ferrite grains formed under the prior deformation of the austenite become more homogeneous and refined. However, the influence of deformation on the tensile strength of both steels is dependent on the isothermal temperatures. Thermodynamic calculation indicates that Si and prior deformation reduce the incubation time of both ferrite and bainite transformation, but the effect is weakened by the decrease of the isothermal temperatures.

Detection of Mechanical Deformation in Old Aged Power Transformer Using Cross Correlation Co-Efficient Analysis Method

Detection of minor faults in power transformer active part is essential because minor faults may develop and lead to major faults and finally irretrievable damages occur. Sweep Frequency Response Analysis (SFRA) is an effective low-voltage, off-line diagnostic tool used for finding out any possible winding displacement or mechanical deterioration inside the Transformer, due to large electromechanical forces occurring from the fault currents or due to Transformer transportation and relocation. In this method, the frequency response of a transformer is taken both at manufacturing industry and concern site. Then both the response is compared to predict the fault taken place in active part. But in old aged transformers, the primary reference response is unavailable. So Cross Correlation Co-Efficient (CCF) measurement technique can be a vital process for fault detection in these transformers. In this paper, theoretical background of SFRA technique has been elaborated and through several case studies, the effectiveness of CCF parameter for fault detection has been represented.

Deformation at Room and Low Temperatures and Martensite Transformation in Resistance Spot Welding Duplexγ&α(δ) Materials of 301L Stainless Steel

Transformation induced plasticity （TRIP） and twinning induced plasticity （TWlP） effects had been widely studied in single austenite steel. But in duplex γ ＆ α（δ） phase, such as welding materials of stainless steel, they had been less studied. Tensile shear loading experiment of resistance spot welding specimens prepared with 2 mm 301L sheets, was carried out at 15℃ and -50℃. Optical microscopy and scanning electron microscopy （SEM） as well as X-ray diffraction （XRD） were used to investigate the microstructure of weld nugget, and specimens fracture surface. The results showed that the initial weld nugget was composed of 8.4% α（δ） ferrite and 91.6% austenite. Tensile shear load bearing capacity of spot welding specimen at -50℃ was 24.8 kN, 17.7% higher than that at 15℃. About 78.5 vol. pct. martensite transformation was induced by plastic deformation at -50℃, while about 67.9 vol. pct transformation induced at 15℃. The plasticity of spot welding joint decreased with the decline of experimental temperature.

Effect of hot deformation on α→β phase transformation in 47Zr-45Ti-5Al-3V alloy

The effect of strain rate and deformation temperature on theα→βphase transformation in 47Zr-45Ti-5Al-3V alloy with an initial widmanstattenαstructure was investigated.At the deformation temperature of 550°C,the volume fraction ofαphase decreased with increasing strain rate.At 600 and 650°C,the volume fraction ofαphase firstly increased to a maximum value with increasing strain rate from 1×10-3 to 1×10-2 s-1,and then decreased.At 700°C,the microstructure consisted of singleβphase.At a given strain rate,the volume fraction ofαphase decreased with increasing deformation temperature.With decreasing strain rate and increasing deformation temperature,the volume fraction and size of globularαphase increased.At 650°C and 1×10-3 s-1,the lamellarαphase was fully globularized.The variation in the volume fraction and morphology ofαphase with strain rate and deformation temperature significantly affected the hardness of 47Zr-45Ti-5Al-3V alloy.

Effects of Plastic Deformation and Stresses on Dilatation duringthe Martensitic Transformation in a B-bearing Steel

To provide data for improved modelling of the behaviour of steel components in a simultaneous forming and quenching process, the effects of plastic deformation and stresses on dilatation during the martensitic transformation in a B-bearing steel were investigated. It was found that plastic deformation of austenite at high temperatures enhances ferrite formation significantly, and consequently, the dilatation decreases markedly even at a cooling rate of 280'C/s. The created ferritic-martensitic microstructure possesses clearly lower hardness and strength than the martensitic structure. Elastic stresses cause the preferred orientation in martensite to be formed so that diametric dilatation can increase by nearly 200% under axial compression.

Effect of Nb on the transformation kinetics of low carbon(manganese) steel during deformation of undercooled austenite

The hot compression tests using Gleeble 1500 were performed by varying the true strain up to 1.6 （80% reduction） in Nbfree and Nb-microalloyed steels. The effect of Nb addition on the transformation kinetics during deformation of undercooled austenite was investigated. It was found that as compared with Nb-free steel, the transformation incubation period of Nb-bearing steel was prolonged and the transformation kinetics curves parallelly moved to higher strain because of the solute Nb drag effect. Studies on kinetics also showed that the deformation-enhanced ferrite transformation （DEFT） of the two steels were composed of three stages, which can be expressed by the J-M-A equations individually. However, the parameter n related to the mode of nucleation and growth is somewhat different in the first and second stages of the two steels, and the same in the third stage for both the steels corresponding to the nucleation Of retained austenite.