3D Object Detection with Attention:Shell-Based Modeling

LIDAR point cloud-based 3D object detection aims to sense the surrounding environment by anchoring objects with the Bounding Box(BBox).However,under the three-dimensional space of autonomous driving scenes,the previous object detection methods,due to the pre-processing of the original LIDAR point cloud into voxels or pillars,lose the coordinate information of the original point cloud,slow detection speed,and gain inaccurate bounding box positioning.To address the issues above,this study proposes a new two-stage network structure to extract point cloud features directly by PointNet++,which effectively preserves the original point cloud coordinate information.To improve the detection accuracy,a shell-based modeling method is proposed.It roughly determines which spherical shell the coordinates belong to.Then,the results are refined to ground truth,thereby narrowing the localization range and improving the detection accuracy.To improve the recall of 3D object detection with bounding boxes,this paper designs a self-attention module for 3D object detection with a skip connection structure.Some of these features are highlighted by weighting them on the feature dimensions.After training,it makes the feature weights that are favorable for object detection get larger.Thus,the extracted features are more adapted to the object detection task.Extensive comparison experiments and ablation experiments conducted on the KITTI dataset verify the effectiveness of our proposed method in improving recall and precision.

Enhancing Dense Small Object Detection in UAV Images Based on Hybrid Transformer

Transformer-based models have facilitated significant advances in object detection.However,their extensive computational consumption and suboptimal detection of dense small objects curtail their applicability in unmanned aerial vehicle(UAV)imagery.Addressing these limitations,we propose a hybrid transformer-based detector,H-DETR,and enhance it for dense small objects,leading to an accurate and efficient model.Firstly,we introduce a hybrid transformer encoder,which integrates a convolutional neural network-based cross-scale fusion module with the original encoder to handle multi-scale feature sequences more efficiently.Furthermore,we propose two novel strategies to enhance detection performance without incurring additional inference computation.Query filter is designed to cope with the dense clustering inherent in drone-captured images by counteracting similar queries with a training-aware non-maximum suppression.Adversarial denoising learning is a novel enhancement method inspired by adversarial learning,which improves the detection of numerous small targets by counteracting the effects of artificial spatial and semantic noise.Extensive experiments on the VisDrone and UAVDT datasets substantiate the effectiveness of our approach,achieving a significant improvement in accuracy with a reduction in computational complexity.Our method achieves 31.9%and 21.1%AP on the VisDrone and UAVDT datasets,respectively,and has a faster inference speed,making it a competitive model in UAV image object detection.

Object-Oriented Modeling of the Variation of Acceleration and Deceleration Characteristics in Relation to Speed Bands for Railway Vehicles

Automated operation and artificial intelligence technology have become essential for ensuring the safety, efficiency, and punctuality of railways, with applications such as ATO (Automatic Train Operation). In this study, the authors propose a method to efficiently simulate the kinematic characteristics of railroad vehicles depending on their speed zone. They utilized the function overloading function supported by a programming language and applied the fourth-order Lunge-Kutta method for dynamic simulation. By constructing an object model, the authors calculated vehicle characteristics and TPS and compared them with actual values, verifying that the developed model represents the real-life vehicle characteristics accurately. The study highlights potential improvements in automated driving and energy consumption optimization in the railway industry.

Predictive modeling for postoperative delirium in elderly patients with abdominal malignancies using synthetic minority oversampling technique

BACKGROUND Postoperative delirium,particularly prevalent in elderly patients after abdominal cancer surgery,presents significant challenges in clinical management.AIM To develop a synthetic minority oversampling technique(SMOTE)-based model for predicting postoperative delirium in elderly abdominal cancer patients.METHODS In this retrospective cohort study,we analyzed data from 611 elderly patients who underwent abdominal malignant tumor surgery at our hospital between September 2020 and October 2022.The incidence of postoperative delirium was recorded for 7 d post-surgery.Patients were divided into delirium and non-delirium groups based on the occurrence of postoperative delirium or not.A multivariate logistic regression model was used to identify risk factors and develop a predictive model for postoperative delirium.The SMOTE technique was applied to enhance the model by oversampling the delirium cases.The model’s predictive accuracy was then validated.RESULTS In our study involving 611 elderly patients with abdominal malignant tumors,multivariate logistic regression analysis identified significant risk factors for postoperative delirium.These included the Charlson comorbidity index,American Society of Anesthesiologists classification,history of cerebrovascular disease,surgical duration,perioperative blood transfusion,and postoperative pain score.The incidence rate of postoperative delirium in our study was 22.91%.The original predictive model(P1)exhibited an area under the receiver operating characteristic curve of 0.862.In comparison,the SMOTE-based logistic early warning model(P2),which utilized the SMOTE oversampling algorithm,showed a slightly lower but comparable area under the curve of 0.856,suggesting no significant difference in performance between the two predictive approaches.CONCLUSION This study confirms that the SMOTE-enhanced predictive model for postoperative delirium in elderly abdominal tumor patients shows performance equivalent to that of traditional methods,effectively addressing data imbalance.

Volume-averaged modeling of multiphase solidification with equiaxed crystal sedimentation in a steel ingot

Macrosegregation is a critical factor that limits the mechanical properties of materials.The impact of equiaxed crystal sedimentation on macrosegregation has been extensively studied,as it plays a significant role in determining the distribution of alloying elements and impurities within a material.To improve macrosegregation in steel connecting shafts,a multiphase solidification model that couples melt flow,heat transfer,microstructure evolution,and solute transport was established based on the volume-averaged Eulerian-Eulerian approach.In this model,the effects of liquid phase,equiaxed crystals,columnar dendrites,and columnar-to-equiaxed transition(CET)during solidification and evolution of microstructure can be considered simultaneously.The sedimentation of equiaxed crystals contributes to negative macrosegregation,where regions between columnar dendrites and equiaxed crystals undergo significant A-type positive macrosegregation due to the CET.Additionally,noticeable positive macrosegregation occurs in the area of final solidification in the ingot.The improvement in macrosegregation is beneficial for enhancing the mechanical properties of connecting shafts.To mitigate the thermal convection of molten steel resulting from excessive superheating,reducing the superheating during casting without employing external fields or altering the design of the ingot mold is indeed an effective approach to control macrosegregation.

Background removal from global auroral images:Data-driven dayglow modeling

Global images of auroras obtained by cameras on spacecraft are a key tool for studying the near-Earth environment.However,the cameras are sensitive not only to auroral emissions produced by precipitating particles,but also to dayglow emissions produced by photoelectrons induced by sunlight.Nightglow emissions and scattered sunlight can contribute to the background signal.To fully utilize such images in space science,background contamination must be removed to isolate the auroral signal.Here we outline a data-driven approach to modeling the background intensity in multiple images by formulating linear inverse problems based on B-splines and spherical harmonics.The approach is robust,flexible,and iteratively deselects outliers,such as auroral emissions.The final model is smooth across the terminator and accounts for slow temporal variations and large-scale asymmetries in the dayglow.We demonstrate the model by using the three far ultraviolet cameras on the Imager for Magnetopause-to-Aurora Global Exploration(IMAGE)mission.The method can be applied to historical missions and is relevant for upcoming missions,such as the Solar wind Magnetosphere Ionosphere Link Explorer(SMILE)mission.

Surrogate modeling for unsaturated infiltration via the physics and equality-constrained artificial neural networks

Machine learning(ML)provides a new surrogate method for investigating groundwater flow dynamics in unsaturated soils.Traditional pure data-driven methods(e.g.deep neural network,DNN)can provide rapid predictions,but they do require sufficient on-site data for accurate training,and lack interpretability to the physical processes within the data.In this paper,we provide a physics and equalityconstrained artificial neural network(PECANN),to derive unsaturated infiltration solutions with a small amount of initial and boundary data.PECANN takes the physics-informed neural network(PINN)as a foundation,encodes the unsaturated infiltration physical laws(i.e.Richards equation,RE)into the loss function,and uses the augmented Lagrangian method to constrain the learning process of the solutions of RE by adding stronger penalty for the initial and boundary conditions.Four unsaturated infiltration cases are designed to test the training performance of PECANN,i.e.one-dimensional(1D)steady-state unsaturated infiltration,1D transient-state infiltration,two-dimensional(2D)transient-state infiltration,and 1D coupled unsaturated infiltration and deformation.The predicted results of PECANN are compared with the finite difference solutions or analytical solutions.The results indicate that PECANN can accurately capture the variations of pressure head during the unsaturated infiltration,and present higher precision and robustness than DNN and PINN.It is also revealed that PECANN can achieve the same accuracy as the finite difference method with fewer initial and boundary training data.Additionally,we investigate the effect of the hyperparameters of PECANN on solving RE problem.PECANN provides an effective tool for simulating unsaturated infiltration.

A methodology for damage evaluation of underground tunnels subjected to static loading using numerical modeling

We have proposed a methodology to assess the robustness of underground tunnels against potential failure.This involves developing vulnerability functions for various qualities of rock mass and static loading intensities.To account for these variations,we utilized a Monte Carlo Simulation(MCS)technique coupled with the finite difference code FLAC^(3D),to conduct two thousand seven hundred numerical simulations of a horseshoe tunnel located within a rock mass with different geological strength index system(GSIs)and subjected to different states of static loading.To quantify the severity of damage within the rock mass,we selected one stress-based(brittle shear ratio(BSR))and one strain-based failure criterion(plastic damage index(PDI)).Based on these criteria,we then developed fragility curves.Additionally,we used mathematical approximation techniques to produce vulnerability functions that relate the probabilities of various damage states to loading intensities for different quality classes of blocky rock mass.The results indicated that the fragility curves we obtained could accurately depict the evolution of the inner and outer shell damage around the tunnel.Therefore,we have provided engineers with a tool that can predict levels of damages associated with different failure mechanisms based on variations in rock mass quality and in situ stress state.Our method is a numerically developed,multi-variate approach that can aid engineers in making informed decisions about the robustness of underground tunnels.

Artificial intelligence-driven radiomics study in cancer:the role of feature engineering and modeling

Modern medicine is reliant on various medical imaging technologies for non-invasively observing patients’anatomy.However,the interpretation of medical images can be highly subjective and dependent on the expertise of clinicians.Moreover,some potentially useful quantitative information in medical images,especially that which is not visible to the naked eye,is often ignored during clinical practice.In contrast,radiomics performs high-throughput feature extraction from medical images,which enables quantitative analysis of medical images and prediction of various clinical endpoints.Studies have reported that radiomics exhibits promising performance in diagnosis and predicting treatment responses and prognosis,demonstrating its potential to be a non-invasive auxiliary tool for personalized medicine.However,radiomics remains in a developmental phase as numerous technical challenges have yet to be solved,especially in feature engineering and statistical modeling.In this review,we introduce the current utility of radiomics by summarizing research on its application in the diagnosis,prognosis,and prediction of treatment responses in patients with cancer.We focus on machine learning approaches,for feature extraction and selection during feature engineering and for imbalanced datasets and multi-modality fusion during statistical modeling.Furthermore,we introduce the stability,reproducibility,and interpretability of features,and the generalizability and interpretability of models.Finally,we offer possible solutions to current challenges in radiomics research.

Objective Model Selection in Physics: Exploring the Finite Information Quantity Approach

Traditional methods for selecting models in experimental data analysis are susceptible to researcher bias, hindering exploration of alternative explanations and potentially leading to overfitting. The Finite Information Quantity (FIQ) approach offers a novel solution by acknowledging the inherent limitations in information processing capacity of physical systems. This framework facilitates the development of objective criteria for model selection (comparative uncertainty) and paves the way for a more comprehensive understanding of phenomena through exploring diverse explanations. This work presents a detailed comparison of the FIQ approach with ten established model selection methods, highlighting the advantages and limitations of each. We demonstrate the potential of FIQ to enhance the objectivity and robustness of scientific inquiry through three practical examples: selecting appropriate models for measuring fundamental constants, sound velocity, and underwater electrical discharges. Further research is warranted to explore the full applicability of FIQ across various scientific disciplines.

Enhanced Object Detection and Classification via Multi-Method Fusion

Advances in machine vision systems have revolutionized applications such as autonomous driving,robotic navigation,and augmented reality.Despite substantial progress,challenges persist,including dynamic backgrounds,occlusion,and limited labeled data.To address these challenges,we introduce a comprehensive methodology toenhance image classification and object detection accuracy.The proposed approach involves the integration ofmultiple methods in a complementary way.The process commences with the application of Gaussian filters tomitigate the impact of noise interference.These images are then processed for segmentation using Fuzzy C-Meanssegmentation in parallel with saliency mapping techniques to find the most prominent regions.The Binary RobustIndependent Elementary Features(BRIEF)characteristics are then extracted fromdata derived fromsaliency mapsand segmented images.For precise object separation,Oriented FAST and Rotated BRIEF(ORB)algorithms areemployed.Genetic Algorithms(GAs)are used to optimize Random Forest classifier parameters which lead toimproved performance.Our method stands out due to its comprehensive approach,adeptly addressing challengessuch as changing backdrops,occlusion,and limited labeled data concurrently.A significant enhancement hasbeen achieved by integrating Genetic Algorithms(GAs)to precisely optimize parameters.This minor adjustmentnot only boosts the uniqueness of our system but also amplifies its overall efficacy.The proposed methodologyhas demonstrated notable classification accuracies of 90.9%and 89.0%on the challenging Corel-1k and MSRCdatasets,respectively.Furthermore,detection accuracies of 87.2%and 86.6%have been attained.Although ourmethod performed well in both datasets it may face difficulties in real-world data especially where datasets havehighly complex backgrounds.Despite these limitations,GAintegration for parameter optimization shows a notablestrength in enhancing the overall adaptability and performance of our system.

Multi-Objective Optimization of Fused Deposition Modeling for Mechanical Properties of Biopolymer Parts Using the Grey-Taguchi Method

The urgent need to develop customized functional products only possible by 3D printing had realized when faced with the unavailability of medical devices like surgical instruments during the coronavirus-19 disease and the ondemand necessity to perform surgery during space missions.Biopolymers have recently been the most appropriate option for fabricating surgical instruments via 3D printing in terms of cheaper and faster processing.Among all 3D printing techniques,fused deposition modelling(FDM)is a low-cost and more rapid printing technique.This article proposes the fabrication of surgical instruments,namely,forceps and hemostat using the fused deposition modeling(FDM)process.Excellent mechanical properties are the only indicator to judge the quality of the functional parts.The mechanical properties of FDM-processed parts depend on various process parameters.These parameters are layer height,infill pattern,top/bottom pattern,number of top/bottom layers,infill density,flow,number of shells,printing temperature,build plate temperature,printing speed,and fan speed.Tensile strength and modulus of elasticity are chosen as evaluation indexes to ascertain the mechanical properties of polylactic acid(PLA)parts printed by FDM.The experiments have performed through Taguchi’s L27orthogonal array(OA).Variance analysis(ANOVA)ascertains the significance of the process parameters and their percent contributions to the evaluation indexes.Finally,as a multiobjective optimization technique,grey relational analysis(GRA)obtains an optimal set of FDM process parameters to fabricate the best parts with comprehensive mechanical properties.Scanning electron microscopy(SEM)examines the types of defects and strong bonding between rasters.The proposed research ensures the successful fabrication of functional surgical tools with substantial ultimate tensile strength(42.6 MPa)and modulus of elasticity(3274 MPa).

Floating Waste Discovery by Request via Object-Centric Learning

Discovering floating wastes,especially bottles on water,is a crucial research problem in environmental hygiene.Nevertheless,real-world applications often face challenges such as interference from irrelevant objects and the high cost associated with data collection.Consequently,devising algorithms capable of accurately localizing specific objects within a scene in scenarios where annotated data is limited remains a formidable challenge.To solve this problem,this paper proposes an object discovery by request problem setting and a corresponding algorithmic framework.The proposed problem setting aims to identify specified objects in scenes,and the associated algorithmic framework comprises pseudo data generation and object discovery by request network.Pseudo-data generation generates images resembling natural scenes through various data augmentation rules,using a small number of object samples and scene images.The network structure of object discovery by request utilizes the pre-trained Vision Transformer(ViT)model as the backbone,employs object-centric methods to learn the latent representations of foreground objects,and applies patch-level reconstruction constraints to the model.During the validation phase,we use the generated pseudo datasets as training sets and evaluate the performance of our model on the original test sets.Experiments have proved that our method achieves state-of-the-art performance on Unmanned Aerial Vehicles-Bottle Detection(UAV-BD)dataset and self-constructed dataset Bottle,especially in multi-object scenarios.

Two-Layer Attention Feature Pyramid Network for Small Object Detection

Effective small object detection is crucial in various applications including urban intelligent transportation and pedestrian detection.However,small objects are difficult to detect accurately because they contain less information.Many current methods,particularly those based on Feature Pyramid Network(FPN),address this challenge by leveraging multi-scale feature fusion.However,existing FPN-based methods often suffer from inadequate feature fusion due to varying resolutions across different layers,leading to suboptimal small object detection.To address this problem,we propose the Two-layerAttention Feature Pyramid Network(TA-FPN),featuring two key modules:the Two-layer Attention Module(TAM)and the Small Object Detail Enhancement Module(SODEM).TAM uses the attention module to make the network more focused on the semantic information of the object and fuse it to the lower layer,so that each layer contains similar semantic information,to alleviate the problem of small object information being submerged due to semantic gaps between different layers.At the same time,SODEM is introduced to strengthen the local features of the object,suppress background noise,enhance the information details of the small object,and fuse the enhanced features to other feature layers to ensure that each layer is rich in small object information,to improve small object detection accuracy.Our extensive experiments on challenging datasets such as Microsoft Common Objects inContext(MSCOCO)and Pattern Analysis Statistical Modelling and Computational Learning,Visual Object Classes(PASCAL VOC)demonstrate the validity of the proposedmethod.Experimental results show a significant improvement in small object detection accuracy compared to state-of-theart detectors.

Progressive fragmentation of granular assemblies within rockslides: Insights from discrete-continuous numerical modeling

Rock fragmentation plays a critical role in rock avalanches,yet conventional approaches such as classical granular flow models or the bonded particle model have limitations in accurately characterizing the progressive disintegration and kinematics of multi-deformable rock blocks during rockslides.The present study proposes a discrete-continuous numerical model,based on a cohesive zone model,to explicitly incorporate the progressive fragmentation and intricate interparticle interactions inherent in rockslides.Breakable rock granular assemblies are released along an inclined plane and flow onto a horizontal plane.The numerical scenarios are established to incorporate variations in slope angle,initial height,friction coefficient,and particle number.The evolutions of fragmentation,kinematic,runout and depositional characteristics are quantitatively analyzed and compared with experimental and field data.A positive linear relationship between the equivalent friction coefficient and the apparent friction coefficient is identified.In general,the granular mass predominantly exhibits characteristics of a dense granular flow,with the Savage number exhibiting a decreasing trend as the volume of mass increases.The process of particle breakage gradually occurs in a bottom-up manner,leading to a significant increase in the angular velocities of the rock blocks with increasing depth.The simulation results reproduce the field observations of inverse grading and source stratigraphy preservation in the deposit.We propose a disintegration index that incorporates factors such as drop height,rock mass volume,and rock strength.Our findings demonstrate a consistent linear relationship between this index and the fragmentation degree in all tested scenarios.

A Secure and Cost-Effective Training Framework Atop Serverless Computing for Object Detection in Blasting

The data analysis of blasting sites has always been the research goal of relevant researchers.The rise of mobile blasting robots has aroused many researchers’interest in machine learning methods for target detection in the field of blasting.Serverless Computing can provide a variety of computing services for people without hardware foundations and rich software development experience,which has aroused people’s interest in how to use it in the field ofmachine learning.In this paper,we design a distributedmachine learning training application based on the AWS Lambda platform.Based on data parallelism,the data aggregation and training synchronization in Function as a Service(FaaS)are effectively realized.It also encrypts the data set,effectively reducing the risk of data leakage.We rent a cloud server and a Lambda,and then we conduct experiments to evaluate our applications.Our results indicate the effectiveness,rapidity,and economy of distributed training on FaaS.

SMSTracker:A Self-Calibration Multi-Head Self-Attention Transformer for Visual Object Tracking

Visual object tracking plays a crucial role in computer vision.In recent years,researchers have proposed various methods to achieve high-performance object tracking.Among these,methods based on Transformers have become a research hotspot due to their ability to globally model and contextualize information.However,current Transformer-based object tracking methods still face challenges such as low tracking accuracy and the presence of redundant feature information.In this paper,we introduce self-calibration multi-head self-attention Transformer(SMSTracker)as a solution to these challenges.It employs a hybrid tensor decomposition self-organizing multihead self-attention transformermechanism,which not only compresses and accelerates Transformer operations but also significantly reduces redundant data,thereby enhancing the accuracy and efficiency of tracking.Additionally,we introduce a self-calibration attention fusion block to resolve common issues of attention ambiguities and inconsistencies found in traditional trackingmethods,ensuring the stability and reliability of tracking performance across various scenarios.By integrating a hybrid tensor decomposition approach with a self-organizingmulti-head self-attentive transformer mechanism,SMSTracker enhances the efficiency and accuracy of the tracking process.Experimental results show that SMSTracker achieves competitive performance in visual object tracking,promising more robust and efficient tracking systems,demonstrating its potential to providemore robust and efficient tracking solutions in real-world applications.

Discontinuity development patterns and the challenges for 3D discrete fracture network modeling on complicated exposed rock surfaces

Natural slopes usually display complicated exposed rock surfaces that are characterized by complex and substantial terrain undulation and ubiquitous undesirable phenomena such as vegetation cover and rockfalls.This study presents a systematic outcrop research of fracture pattern variations in a complicated rock slope,and the qualitative and quantitative study of the complex phenomena impact on threedimensional(3D)discrete fracture network(DFN)modeling.As the studies of the outcrop fracture pattern have been so far focused on local variations,thus,we put forward a statistical analysis of global variations.The entire outcrop is partitioned into several subzones,and the subzone-scale variability of fracture geometric properties is analyzed(including the orientation,the density,and the trace length).The results reveal significant variations in fracture characteristics(such as the concentrative degree,the average orientation,the density,and the trace length)among different subzones.Moreover,the density of fracture sets,which is approximately parallel to the slope surface,exhibits a notably higher value compared to other fracture sets across all subzones.To improve the accuracy of the DFN modeling,the effects of three common phenomena resulting from vegetation and rockfalls are qualitatively analyzed and the corresponding quantitative data processing solutions are proposed.Subsequently,the 3D fracture geometric parameters are determined for different areas of the high-steep rock slope in terms of the subzone dimensions.The results show significant variations in the same set of 3D fracture parameters across different regions with density differing by up to tenfold and mean trace length exhibiting differences of 3e4 times.The study results present precise geological structural information,improve modeling accuracy,and provide practical solutions for addressing complex outcrop issues.

The relationship between compartment models and their stochastic counterparts:A comparative study with examples of the COVID-19 epidemic modeling

Deterministic compartment models(CMs)and stochastic models,including stochastic CMs and agent-based models,are widely utilized in epidemic modeling.However,the relationship between CMs and their corresponding stochastic models is not well understood.The present study aimed to address this gap by conducting a comparative study using the susceptible,exposed,infectious,and recovered(SEIR)model and its extended CMs from the coronavirus disease 2019 modeling literature.We demonstrated the equivalence of the numerical solution of CMs using the Euler scheme and their stochastic counterparts through theoretical analysis and simulations.Based on this equivalence,we proposed an efficient model calibration method that could replicate the exact solution of CMs in the corresponding stochastic models through parameter adjustment.The advancement in calibration techniques enhanced the accuracy of stochastic modeling in capturing the dynamics of epidemics.However,it should be noted that discrete-time stochastic models cannot perfectly reproduce the exact solution of continuous-time CMs.Additionally,we proposed a new stochastic compartment and agent mixed model as an alternative to agent-based models for large-scale population simulations with a limited number of agents.This model offered a balance between computational efficiency and accuracy.The results of this research contributed to the comparison and unification of deterministic CMs and stochastic models in epidemic modeling.Furthermore,the results had implications for the development of hybrid models that integrated the strengths of both frameworks.Overall,the present study has provided valuable epidemic modeling techniques and their practical applications for understanding and controlling the spread of infectious diseases.

Learning Discriminatory Information for Object Detection on Urine Sediment Image

In clinical practice,the microscopic examination of urine sediment is considered an important in vitro examination with many broad applications.Measuring the amount of each type of urine sediment allows for screening,diagnosis and evaluation of kidney and urinary tract disease,providing insight into the specific type and severity.However,manual urine sediment examination is labor-intensive,time-consuming,and subjective.Traditional machine learning based object detection methods require hand-crafted features for localization and classification,which have poor generalization capabilities and are difficult to quickly and accurately detect the number of urine sediments.Deep learning based object detection methods have the potential to address the challenges mentioned above,but these methods require access to large urine sediment image datasets.Unfortunately,only a limited number of publicly available urine sediment datasets are currently available.To alleviate the lack of urine sediment datasets in medical image analysis,we propose a new dataset named UriSed2K,which contains 2465 high-quality images annotated with expert guidance.Two main challenges are associated with our dataset:a large number of small objects and the occlusion between these small objects.Our manuscript focuses on applying deep learning object detection methods to the urine sediment dataset and addressing the challenges presented by this dataset.Specifically,our goal is to improve the accuracy and efficiency of the detection algorithm and,in doing so,provide medical professionals with an automatic detector that saves time and effort.We propose an improved lightweight one-stage object detection algorithm called Discriminatory-YOLO.The proposed algorithm comprises a local context attention module and a global background suppression module,which aid the detector in distinguishing urine sediment features in the image.The local context attention module captures context information beyond the object region,while the global background suppression module emphasizes objects in uninformative backgrounds.We comprehensively evaluate our method on the UriSed2K dataset,which includes seven categories of urine sediments,such as erythrocytes(red blood cells),leukocytes(white blood cells),epithelial cells,crystals,mycetes,broken erythrocytes,and broken leukocytes,achieving the best average precision(AP)of 95.3%while taking only 10 ms per image.The source code and dataset are available at https://github.com/binghuiwu98/discriminatoryyolov5.