CFM-UNet:A Joint CNN and Transformer Network via Cross Feature Modulation for Remote Sensing Images Segmentation

The semantic segmentation methods based on CNN have made great progress,but there are still some shortcomings in the application of remote sensing images segmentation,such as the small receptive field can not effectively capture global context.In order to solve this problem,this paper proposes a hybrid model based on ResNet50 and swin transformer to directly capture long-range dependence,which fuses features through Cross Feature Modulation Module(CFMM).Experimental results on two publicly available datasets,Vaihingen and Potsdam,are mIoU of 70.27%and 76.63%,respectively.Thus,CFM-UNet can maintain a high segmentation performance compared with other competitive networks.

DT-Net:Joint Dual-Input Transformer and CNN for Retinal Vessel Segmentation

Retinal vessel segmentation in fundus images plays an essential role in the screening,diagnosis,and treatment of many diseases.The acquired fundus images generally have the following problems:uneven illumination,high noise,and complex structure.It makes vessel segmentation very challenging.Previous methods of retinal vascular segmentation mainly use convolutional neural networks on U Network(U-Net)models,and they have many limitations and shortcomings,such as the loss of microvascular details at the end of the vessels.We address the limitations of convolution by introducing the transformer into retinal vessel segmentation.Therefore,we propose a hybrid method for retinal vessel segmentation based on modulated deformable convolution and the transformer,named DT-Net.Firstly,multi-scale image features are extracted by deformable convolution and multi-head selfattention(MHSA).Secondly,image information is recovered,and vessel morphology is refined by the proposed transformer decoder block.Finally,the local prediction results are obtained by the side output layer.The accuracy of the vessel segmentation is improved by the hybrid loss function.Experimental results show that our method obtains good segmentation performance on Specificity(SP),Sensitivity(SE),Accuracy(ACC),Curve(AUC),and F1-score on three publicly available fundus datasets such as DRIVE,STARE,and CHASE_DB1.

A photogrammetric approach for quantifying the evolution of rock joint void geometry under varying contact states

Accurate measurement of the evolution of rock joint void geometry is essential for comprehending the distribution characteristics of asperities responsible for shear and seepage behaviors.However,existing techniques often require specialized equipment and skilled operators,posing practical challenges.In this study,a cost-effective photogrammetric approach is proposed.Particularly,local coordinate systems are established to facilitate the alignment and precise quantification of the relative position between two halves of a rock joint.Push/pull tests are conducted on rock joints with varying roughness levels to induce different contact states.A high-precision laser scanner serves as a benchmark for evaluating the photogrammetry method.Despite certain deviations exist,the measured evolution of void geometry is generally consistent with the qualitative findings of previous studies.The photogrammetric measurements yield comparable accuracy to laser scanning,with maximum errors of 13.2%for aperture and 14.4%for void volume.Most joint matching coefficient(JMC)measurement errors are below 20%.Larger measurement errors occur primarily in highly mismatched rock joints with JMC values below 0.2,but even in cases where measurement errors exceed 80%,the maximum JMC error is only 0.0434.Thus,the proposed photogrammetric approach holds promise for widespread application in void geometry measurements in rock joints.

Shear behavior and off-fault damage of saw-cut smooth and tension-induced rough joints in granite

The damage of rock joints or fractures upon shear includes the surface damage occurring at the contact asperities and the damage beneath the shear surface within the host rock.The latter is commonly known as off-fault damage and has been much less investigated than the surface damage.The main contribution of this study is to compare the results of direct shear tests conducted on saw-cut planar joints and tension-induced rough granite joints under normal stresses ranging from 1 MPa to 50 MPa.The shear-induced off-fault damages are quantified and compared with the optical microscope observation.Our results clearly show that the planar joints slip stably under all the normal stresses except under 50 MPa,where some local fractures and regular stick-slip occur towards the end of the test.Both post-peak stress drop and stick-slip occur for all the rough joints.The residual shear strength envelopes for the rough joints and the peak shear strength envelope for the planar joints almost overlap.The root mean square(RMS)of asperity height for the rough joints decreases while it increases for the planar joint after shear,and a larger normal stress usually leads to a more significant decrease or increase in RMS.Besides,the extent of off-fault damage(or damage zone)increases with normal stress for both planar and rough joints,and it is restricted to a very thin layer with limited micro-cracks beneath the planar joint surface.In comparison,the thickness of the damage zone for the rough joints is about an order of magnitude larger than that of the planar joints,and the coalesced micro-cracks are generally inclined to the shear direction with acute angles.The findings obtained in this study contribute to a better understanding on the frictional behavior and damage characteristics of rock joints or fractures with different roughness.

Visual Semantic Segmentation Based on Few/Zero-Shot Learning:An Overview

Visual semantic segmentation aims at separating a visual sample into diverse blocks with specific semantic attributes and identifying the category for each block,and it plays a crucial role in environmental perception.Conventional learning-based visual semantic segmentation approaches count heavily on largescale training data with dense annotations and consistently fail to estimate accurate semantic labels for unseen categories.This obstruction spurs a craze for studying visual semantic segmentation with the assistance of few/zero-shot learning.The emergence and rapid progress of few/zero-shot visual semantic segmentation make it possible to learn unseen categories from a few labeled or even zero-labeled samples,which advances the extension to practical applications.Therefore,this paper focuses on the recently published few/zero-shot visual semantic segmentation methods varying from 2D to 3D space and explores the commonalities and discrepancies of technical settlements under different segmentation circumstances.Specifically,the preliminaries on few/zeroshot visual semantic segmentation,including the problem definitions,typical datasets,and technical remedies,are briefly reviewed and discussed.Moreover,three typical instantiations are involved to uncover the interactions of few/zero-shot learning with visual semantic segmentation,including image semantic segmentation,video object segmentation,and 3D segmentation.Finally,the future challenges of few/zero-shot visual semantic segmentation are discussed.

RB-DEM Modeling and Simulation of Non-Persisting Rough Open Joints Based on the IFS-Enhanced Method

When the geological environment of rock masses is disturbed,numerous non-persisting open joints can appear within it.It is crucial to investigate the effect of open joints on the mechanical properties of rock mass.However,it has been challenging to generate realistic open joints in traditional experimental tests and numerical simulations.This paper presents a novel solution to solve the problem.By utilizing the stochastic distribution of joints and an enhanced-fractal interpolation system(IFS)method,rough curves with any orientation can be generated.The Douglas-Peucker algorithm is then applied to simplify these curves by removing unnecessary points while preserving their fundamental shape.Subsequently,open joints are created by connecting points that move to both sides of rough curves based on the aperture distribution.Mesh modeling is performed to construct the final mesh model.Finally,the RB-DEM method is applied to transform the mesh model into a discrete element model containing geometric information about these open joints.Furthermore,this study explores the impacts of rough open joint orientation,aperture,and number on rock fracture mechanics.This method provides a realistic and effective approach for modeling and simulating these non-persisting open joints.

Analysis of the joint detection capability of the SMILE satellite and EISCAT-3D radar

The Solar wind Magnetosphere Ionosphere Link Explorer(SMILE)satellite is a small magnetosphere–ionosphere link explorer developed cooperatively between China and Europe.It pioneers the use of X-ray imaging technology to perform large-scale imaging of the Earth’s magnetosheath and polar cusp regions.It uses a high-precision ultraviolet imager to image the overall configuration of the aurora and monitor changes in the source of solar wind in real time,using in situ detection instruments to improve human understanding of the relationship between solar activity and changes in the Earth’s magnetic field.The SMILE satellite is scheduled to launch in 2025.The European Incoherent Scatter Sciences Association(EISCAT)-3D radar is a new generation of European incoherent scatter radar constructed by EISCAT and is the most advanced ground-based ionospheric experimental device in the high-latitude polar region.It has multibeam and multidirectional quasi-real-time three-dimensional(3D)imaging capabilities,continuous monitoring and operation capabilities,and multiple-baseline interferometry capabilities.Joint detection by the SMILE satellite and the EISCAT-3D radar is of great significance for revealing the coupling process of the solar wind–magnetosphere–ionosphere.Therefore,we performed an analysis of the joint detection capability of the SMILE satellite and EISCAT-3D,analyzed the period during which the two can perform joint detection,and defined the key scientific problems that can be solved by joint detection.In addition,we developed Web-based software to search for and visualize the joint detection period of the SMILE satellite and EISCAT-3D radar,which lays the foundation for subsequent joint detection experiments and scientific research.

On the calibration of a shear stress criterion for rock joints to represent the full stress-strain profile

Conventional numerical solutions developed to describe the geomechanical behavior of rock interfaces subjected to differential load emphasize peak and residual shear strengths.The detailed analysis of preand post-peak shear stress-displacement behavior is central to various time-dependent and dynamic rock mechanic problems such as rockbursts and structural instabilities in highly stressed conditions.The complete stress-displacement surface(CSDS)model was developed to describe analytically the pre-and post-peak behavior of rock interfaces under differential loads.Original formulations of the CSDS model required extensive curve-fitting iterations which limited its practical applicability and transparent integration into engineering tools.The present work proposes modifications to the CSDS model aimed at developing a comprehensive and modern calibration protocol to describe the complete shear stressdisplacement behavior of rock interfaces under differential loads.The proposed update to the CSDS model incorporates the concept of mobilized shear strength to enhance the post-peak formulations.Barton’s concepts of joint roughness coefficient(JRC)and joint compressive strength(JCS)are incorporated to facilitate empirical estimations for peak shear stress and normal closure relations.Triaxial/uniaxial compression test and direct shear test results are used to validate the updated model and exemplify the proposed calibration method.The results illustrate that the revised model successfully predicts the post-peak and complete axial stressestrain and shear stressedisplacement curves for rock joints.

Subsequent total joint arthroplasty: Are we learning from the first stage?

BACKGROUND With the increasing incidence of total joint arthroplasty(TJA),there is a desire to reduce peri-operative complications and resource utilization.As degenerative conditions progress in multiple joints,many patients undergo multiple proce-dures.AIM To determine if both physicians and patients learn from the patient’s initial arth-roplasty,resulting in improved outcomes following the second procedure.METHODS The institutional database was retrospectively queried for primary total hip arth-roplasty(THA)and total knee arthroplasty(TKA).Patients with only unilateral THA or TKA,and patients undergoing same-day bilateral TJA,were excluded.Patient demographics,comorbidities,and implant sizes were collected at the time of each procedure and patients were stratified by first vs second surgery.Outcome metrics evaluated included operative time,length of stay(LOS),disposition,90-d readmissions and emergency department(ED)visits.RESULTS A total of 642 patients,including 364 undergoing staged bilateral TKA and 278 undergoing bilateral THA,were analyzed.There was no significant difference in demographics or comorbidities between the first and second procedure,which were separated by a mean of 285 d.For THA and TKA,LOS was significantly less for the second surgery,with 66%of patients having a shorter hospitalization(P<0.001).THA patients had significantly decreased operative time only when the same sized implant was utilized(P=0.025).The vast majority(93.3%)of patients were discharged to the same type of location following their second surgery.However,when a change in disposition was present from the first surgery,patients were significantly more likely to be discharged to home after the second procedure(P=0.033).There was no difference between procedures for post-operative readmissions(P=0.438)or ED visits(P=0.915).CONCLUSION After gaining valuable experience recovering from the initial surgery,a patient’s perioperative outcomes are improved for their second TJA.This may be the result of increased confidence and decreased anxiety,and it supports the theory that enhanced patient education pre-operatively may improve outcomes.For the surgical team,the second procedure of a staged THA is more efficient,although this finding did not hold for TKA.

CrossFormer Embedding DeepLabv3+ for Remote Sensing Images Semantic Segmentation

High-resolution remote sensing image segmentation is a challenging task. In urban remote sensing, the presenceof occlusions and shadows often results in blurred or invisible object boundaries, thereby increasing the difficultyof segmentation. In this paper, an improved network with a cross-region self-attention mechanism for multi-scalefeatures based onDeepLabv3+is designed to address the difficulties of small object segmentation and blurred targetedge segmentation. First,we use CrossFormer as the backbone feature extraction network to achieve the interactionbetween large- and small-scale features, and establish self-attention associations between features at both large andsmall scales to capture global contextual feature information. Next, an improved atrous spatial pyramid poolingmodule is introduced to establish multi-scale feature maps with large- and small-scale feature associations, andattention vectors are added in the channel direction to enable adaptive adjustment of multi-scale channel features.The proposed networkmodel is validated using the PotsdamandVaihingen datasets. The experimental results showthat, compared with existing techniques, the network model designed in this paper can extract and fuse multiscaleinformation, more clearly extract edge information and small-scale information, and segment boundariesmore smoothly. Experimental results on public datasets demonstrate the superiority of ourmethod compared withseveral state-of-the-art networks.

Part-Whole Relational Few-Shot 3D Point Cloud Semantic Segmentation

This paper focuses on the task of few-shot 3D point cloud semantic segmentation.Despite some progress,this task still encounters many issues due to the insufficient samples given,e.g.,incomplete object segmentation and inaccurate semantic discrimination.To tackle these issues,we first leverage part-whole relationships into the task of 3D point cloud semantic segmentation to capture semantic integrity,which is empowered by the dynamic capsule routing with the module of 3D Capsule Networks(CapsNets)in the embedding network.Concretely,the dynamic routing amalgamates geometric information of the 3D point cloud data to construct higher-level feature representations,which capture the relationships between object parts and their wholes.Secondly,we designed a multi-prototype enhancement module to enhance the prototype discriminability.Specifically,the single-prototype enhancement mechanism is expanded to the multi-prototype enhancement version for capturing rich semantics.Besides,the shot-correlation within the category is calculated via the interaction of different samples to enhance the intra-category similarity.Ablation studies prove that the involved part-whole relations and proposed multi-prototype enhancement module help to achieve complete object segmentation and improve semantic discrimination.Moreover,under the integration of these two modules,quantitative and qualitative experiments on two public benchmarks,including S3DIS and ScanNet,indicate the superior performance of the proposed framework on the task of 3D point cloud semantic segmentation,compared to some state-of-the-art methods.

Adaptive Segmentation for Unconstrained Iris Recognition

In standard iris recognition systems,a cooperative imaging framework is employed that includes a light source with a near-infrared wavelength to reveal iris texture,look-and-stare constraints,and a close distance requirement to the capture device.When these conditions are relaxed,the system’s performance significantly deteriorates due to segmentation and feature extraction problems.Herein,a novel segmentation algorithm is proposed to correctly detect the pupil and limbus boundaries of iris images captured in unconstrained environments.First,the algorithm scans the whole iris image in the Hue Saturation Value(HSV)color space for local maxima to detect the sclera region.The image quality is then assessed by computing global features in red,green and blue(RGB)space,as noisy images have heterogeneous characteristics.The iris images are accordingly classified into seven categories based on their global RGB intensities.After the classification process,the images are filtered,and adaptive thresholding is applied to enhance the global contrast and detect the outer iris ring.Finally,to characterize the pupil area,the algorithm scans the cropped outer ring region for local minima values to identify the darkest area in the iris ring.The experimental results show that our method outperforms existing segmentation techniques using the UBIRIS.v1 and v2 databases and achieved a segmentation accuracy of 99.32 on UBIRIS.v1 and an error rate of 1.59 on UBIRIS.v2.

Dynamic SLAM Visual Odometry Based on Instance Segmentation:A Comprehensive Review

Dynamic Simultaneous Localization and Mapping(SLAM)in visual scenes is currently a major research area in fields such as robot navigation and autonomous driving.However,in the face of complex real-world envi-ronments,current dynamic SLAM systems struggle to achieve precise localization and map construction.With the advancement of deep learning,there has been increasing interest in the development of deep learning-based dynamic SLAM visual odometry in recent years,and more researchers are turning to deep learning techniques to address the challenges of dynamic SLAM.Compared to dynamic SLAM systems based on deep learning methods such as object detection and semantic segmentation,dynamic SLAM systems based on instance segmentation can not only detect dynamic objects in the scene but also distinguish different instances of the same type of object,thereby reducing the impact of dynamic objects on the SLAM system’s positioning.This article not only introduces traditional dynamic SLAM systems based on mathematical models but also provides a comprehensive analysis of existing instance segmentation algorithms and dynamic SLAM systems based on instance segmentation,comparing and summarizing their advantages and disadvantages.Through comparisons on datasets,it is found that instance segmentation-based methods have significant advantages in accuracy and robustness in dynamic environments.However,the real-time performance of instance segmentation algorithms hinders the widespread application of dynamic SLAM systems.In recent years,the rapid development of single-stage instance segmentationmethods has brought hope for the widespread application of dynamic SLAM systems based on instance segmentation.Finally,possible future research directions and improvementmeasures are discussed for reference by relevant professionals.

Real-Time Detection and Instance Segmentation of Strawberry in Unstructured Environment

The real-time detection and instance segmentation of strawberries constitute fundamental components in the development of strawberry harvesting robots.Real-time identification of strawberries in an unstructured envi-ronment is a challenging task.Current instance segmentation algorithms for strawberries suffer from issues such as poor real-time performance and low accuracy.To this end,the present study proposes an Efficient YOLACT(E-YOLACT)algorithm for strawberry detection and segmentation based on the YOLACT framework.The key enhancements of the E-YOLACT encompass the development of a lightweight attention mechanism,pyramid squeeze shuffle attention(PSSA),for efficient feature extraction.Additionally,an attention-guided context-feature pyramid network(AC-FPN)is employed instead of FPN to optimize the architecture’s performance.Furthermore,a feature-enhanced model(FEM)is introduced to enhance the prediction head’s capabilities,while efficient fast non-maximum suppression(EF-NMS)is devised to improve non-maximum suppression.The experimental results demonstrate that the E-YOLACT achieves a Box-mAP and Mask-mAP of 77.9 and 76.6,respectively,on the custom dataset.Moreover,it exhibits an impressive category accuracy of 93.5%.Notably,the E-YOLACT also demonstrates a remarkable real-time detection capability with a speed of 34.8 FPS.The method proposed in this article presents an efficient approach for the vision system of a strawberry-picking robot.

A semantic segmentation-based underwater acoustic image transmission framework for cooperative SLAM

With the development of underwater sonar detection technology,simultaneous localization and mapping(SLAM)approach has attracted much attention in underwater navigation field in recent years.But the weak detection ability of a single vehicle limits the SLAM performance in wide areas.Thereby,cooperative SLAM using multiple vehicles has become an important research direction.The key factor of cooperative SLAM is timely and efficient sonar image transmission among underwater vehicles.However,the limited bandwidth of underwater acoustic channels contradicts a large amount of sonar image data.It is essential to compress the images before transmission.Recently,deep neural networks have great value in image compression by virtue of the powerful learning ability of neural networks,but the existing sonar image compression methods based on neural network usually focus on the pixel-level information without the semantic-level information.In this paper,we propose a novel underwater acoustic transmission scheme called UAT-SSIC that includes semantic segmentation-based sonar image compression(SSIC)framework and the joint source-channel codec,to improve the accuracy of the semantic information of the reconstructed sonar image at the receiver.The SSIC framework consists of Auto-Encoder structure-based sonar image compression network,which is measured by a semantic segmentation network's residual.Considering that sonar images have the characteristics of blurred target edges,the semantic segmentation network used a special dilated convolution neural network(DiCNN)to enhance segmentation accuracy by expanding the range of receptive fields.The joint source-channel codec with unequal error protection is proposed that adjusts the power level of the transmitted data,which deal with sonar image transmission error caused by the serious underwater acoustic channel.Experiment results demonstrate that our method preserves more semantic information,with advantages over existing methods at the same compression ratio.It also improves the error tolerance and packet loss resistance of transmission.

Improved Convolutional Neural Network for Traffic Scene Segmentation

In actual traffic scenarios,precise recognition of traffic participants,such as vehicles and pedestrians,is crucial for intelligent transportation.This study proposes an improved algorithm built on Mask-RCNN to enhance the ability of autonomous driving systems to recognize traffic participants.The algorithmincorporates long and shortterm memory networks and the fused attention module(GSAM,GCT,and Spatial Attention Module)to enhance the algorithm’s capability to process both global and local information.Additionally,to increase the network’s initial operation stability,the original network activation function was replaced with Gaussian error linear unit.Experiments were conducted using the publicly available Cityscapes dataset.Comparing the test results,it was observed that the revised algorithmoutperformed the original algorithmin terms of AP_(50),AP_(75),and othermetrics by 8.7%and 9.6%for target detection and 12.5%and 13.3%for segmentation.

ProNet Adaptive Retinal Vessel Segmentation Algorithm Based on Improved UperNet Network

This paper proposes a new network structure,namely the ProNet network.Retinal medical image segmentation can help clinical diagnosis of related eye diseases and is essential for subsequent rational treatment.The baseline model of the ProNet network is UperNet(Unified perceptual parsing Network),and the backbone network is ConvNext(Convolutional Network).A network structure based on depth-separable convolution and 1×1 convolution is used,which has good performance and robustness.We further optimise ProNet mainly in two aspects.One is data enhancement using increased noise and slight angle rotation,which can significantly increase the diversity of data and help the model better learn the patterns and features of the data and improve the model’s performance.Meanwhile,it can effectively expand the training data set,reduce the influence of noise and abnormal data in the data set on the model,and improve the accuracy and reliability of the model.Another is the loss function aspect,and we finally use the focal loss function.The focal loss function is well suited for complex tasks such as object detection.The function will penalise the loss carried by samples that the model misclassifies,thus enabling better training of the model to avoid these errors while solving the category imbalance problem as a way to improve image segmentation density and segmentation accuracy.From the experimental results,the evaluation metrics mIoU(mean Intersection over Union)enhanced by 4.47%,and mDice enhanced by 2.92% compared to the baseline network.Better generalization effects and more accurate image segmentation are achieved.

Semantic segmentation via pixel-to-center similarity calculation

Since the fully convolutional network has achieved great success in semantic segmentation,lots of works have been proposed to extract discriminative pixel representations.However,the authors observe that existing methods still suffer from two typical challenges:(i)The intra-class feature variation between different scenes may be large,leading to the difficulty in maintaining the consistency between same-class pixels from different scenes;(ii)The inter-class feature distinction in the same scene could be small,resulting in the limited performance to distinguish different classes in each scene.The authors first rethink se-mantic segmentation from a perspective of similarity between pixels and class centers.Each weight vector of the segmentation head represents its corresponding semantic class in the whole dataset,which can be regarded as the embedding of the class center.Thus,the pixel-wise classification amounts to computing similarity in the final feature space between pixels and the class centers.Under this novel view,the authors propose a Class Center Similarity(CCS)layer to address the above-mentioned challenges by generating adaptive class centers conditioned on each scenes and supervising the similarities between class centers.The CCS layer utilises the Adaptive Class Center Module to generate class centers conditioned on each scene,which adapt the large intra-class variation between different scenes.Specially designed Class Distance Loss(CD Loss)is introduced to control both inter-class and intra-class distances based on the predicted center-to-center and pixel-to-center similarity.Finally,the CCS layer outputs the processed pixel-to-center similarity as the segmentation prediction.Extensive experiments demonstrate that our model performs favourably against the state-of-the-art methods.

Multilevel Attention Unet Segmentation Algorithmfor Lung Cancer Based on CT Images

Lung cancer is a malady of the lungs that gravely jeopardizes human health.Therefore,early detection and treatment are paramount for the preservation of human life.Lung computed tomography(CT)image sequences can explicitly delineate the pathological condition of the lungs.To meet the imperative for accurate diagnosis by physicians,expeditious segmentation of the region harboring lung cancer is of utmost significance.We utilize computer-aided methods to emulate the diagnostic process in which physicians concentrate on lung cancer in a sequential manner,erect an interpretable model,and attain segmentation of lung cancer.The specific advancements can be encapsulated as follows:1)Concentration on the lung parenchyma region:Based on 16-bit CT image capturing and the luminance characteristics of lung cancer,we proffer an intercept histogram algorithm.2)Focus on the specific locus of lung malignancy:Utilizing the spatial interrelation of lung cancer,we propose a memory-based Unet architecture and incorporate skip connections.3)Data Imbalance:In accordance with the prevalent situation of an overabundance of negative samples and a paucity of positive samples,we scrutinize the existing loss function and suggest a mixed loss function.Experimental results with pre-existing publicly available datasets and assembled datasets demonstrate that the segmentation efficacy,measured as Area Overlap Measure(AOM)is superior to 0.81,which markedly ameliorates in comparison with conventional algorithms,thereby facilitating physicians in diagnosis.

Multi-Level Parallel Network for Brain Tumor Segmentation

Accurate automatic segmentation of gliomas in various sub-regions,including peritumoral edema,necrotic core,and enhancing and non-enhancing tumor core from 3D multimodal MRI images,is challenging because of its highly heterogeneous appearance and shape.Deep convolution neural networks(CNNs)have recently improved glioma segmentation performance.However,extensive down-sampling such as pooling or stridden convolution in CNNs significantly decreases the initial image resolution,resulting in the loss of accurate spatial and object parts information,especially information on the small sub-region tumors,affecting segmentation performance.Hence,this paper proposes a novel multi-level parallel network comprising three different level parallel subnetworks to fully use low-level,mid-level,and high-level information and improve the performance of brain tumor segmentation.We also introduce the Combo loss function to address input class imbalance and false positives and negatives imbalance in deep learning.The proposed method is trained and validated on the BraTS 2020 training and validation dataset.On the validation dataset,ourmethod achieved a mean Dice score of 0.907,0.830,and 0.787 for the whole tumor,tumor core,and enhancing tumor core,respectively.Compared with state-of-the-art methods,the multi-level parallel network has achieved competitive results on the validation dataset.