Various deep learning models have been proposed for the accurate assisted diagnosis of early-stage Alzheimer’s disease(AD).Most studies predominantly employ Convolutional Neural Networks(CNNs),which focus solely on l...Various deep learning models have been proposed for the accurate assisted diagnosis of early-stage Alzheimer’s disease(AD).Most studies predominantly employ Convolutional Neural Networks(CNNs),which focus solely on local features,thus encountering difficulties in handling global features.In contrast to natural images,Structural Magnetic Resonance Imaging(sMRI)images exhibit a higher number of channel dimensions.However,during the Position Embedding stage ofMulti Head Self Attention(MHSA),the coded information related to the channel dimension is disregarded.To tackle these issues,we propose theRepBoTNet-CESA network,an advanced AD-aided diagnostic model that is capable of learning local and global features simultaneously.It combines the advantages of CNN networks in capturing local information and Transformer networks in integrating global information,reducing computational costs while achieving excellent classification performance.Moreover,it uses the Cubic Embedding Self Attention(CESA)proposed in this paper to incorporate the channel code information,enhancing the classification performance within the Transformer structure.Finally,the RepBoTNet-CESA performs well in various AD-aided diagnosis tasks,with an accuracy of 96.58%,precision of 97.26%,and recall of 96.23%in the AD/NC task;an accuracy of 92.75%,precision of 92.84%,and recall of 93.18%in the EMCI/NC task;and an accuracy of 80.97%,precision of 83.86%,and recall of 80.91%in the AD/EMCI/LMCI/NC task.This demonstrates that RepBoTNet-CESA delivers outstanding outcomes in various AD-aided diagnostic tasks.Furthermore,our study has shown that MHSA exhibits superior performance compared to conventional attention mechanisms in enhancing ResNet performance.Besides,the Deeper RepBoTNet-CESA network fails to make further progress in AD-aided diagnostic tasks.展开更多
Applying homogeneous coordinates, we extend a newly appeared algorithm of best constrained multi-degree reduction for polynomial Bezier curves to the algorithms of constrained multi-degree reduction for rational Bezie...Applying homogeneous coordinates, we extend a newly appeared algorithm of best constrained multi-degree reduction for polynomial Bezier curves to the algorithms of constrained multi-degree reduction for rational Bezier curves. The idea is introducing two criteria, variance criterion and ratio criterion, for reparameterization of rational Bezier curves, which are used to make uniform the weights of the rational Bezier curves as accordant as possible, and then do multi-degree reduction for each component in homogeneous coordinates. Compared with the two traditional algorithms of ＂cancelling the best linear common divisor＂ and ＂shifted Chebyshev polynomial＂, the two new algorithms presented here using reparameterization have advantages of simplicity and fast computing, being able to preserve high degrees continuity at the end points of the curves, do multi-degree reduction at one time, and have good approximating effect.展开更多
A series of advantages of single difference (SD) and undifferenced (ZD) models are given as compared with the double difference (DD) model. However, rank defects exist in SD and ZD models. The reparameterization metho...A series of advantages of single difference (SD) and undifferenced (ZD) models are given as compared with the double difference (DD) model. However, rank defects exist in SD and ZD models. The reparameterization method is provided to resolve this rank defect problem by estimating some combinations of the unknowns rather than the unknowns themselves. The reparameterization of SD and ZD functional models is discussed in detail with their stochastic models. The theoretical confirmation of the equivalence of undifferenced and differenced models is described in a straightforward way. The relationship between SD and ZD residuals is given and verified for some special purposes, e.g. research on the stochastical properties of GPS observations.展开更多
Coincidence detection of two curves or two surfaces has wide application in computer-aided design(CAD)and computer-aided geometric design(CAGD).Proper reparameterization is the most complicated part in the detection.T...Coincidence detection of two curves or two surfaces has wide application in computer-aided design(CAD)and computer-aided geometric design(CAGD).Proper reparameterization is the most complicated part in the detection.This paper presents and proves the efficient and necessary coincidence condition for two rational Bézier curves in a new way.It also proposes an effective and efficient proper reparameterization method,Algorithm 1,for detecting a rational Bézier curve which can be degenerated into a new one of a lower degree.A numerical proper reparameterization method,Algorithm 2,and examples are also presented.Algorithm 1 is up to ten times faster than other prevailing methods,and Algorithm 2 is twice as fast and half as close as other prevailing methods.New CAD systems using Algorithm 1 and Algorithm 2 will hold accuracy and little computation time.展开更多
Many works have investigated the problem of reparameterizing rational B^zier curves or surfaces via MSbius transformation to adjust their parametric distribution as well as weights, such that the maximal ratio of weig...Many works have investigated the problem of reparameterizing rational B^zier curves or surfaces via MSbius transformation to adjust their parametric distribution as well as weights, such that the maximal ratio of weights becomes smallerthat some algebraic and computational properties of the curves or surfaces can be improved in a way. However, it is an indication of veracity and optimization of the reparameterization to do prior to judge whether the maximal ratio of weights reaches minimum, and verify the new weights after MSbius transfor- mation. What＇s more the users of computer aided design softwares may require some guidelines for designing rational B6zier curves or surfaces with the smallest ratio of weights. In this paper we present the necessary and sufficient conditions that the maximal ratio of weights of the curves or surfaces reaches minimum and also describe it by using weights succinctly and straightway. The weights being satisfied these conditions are called being in the stable state. Applying such conditions, any giving rational B6zier curve or surface can automatically be adjusted to come into the stable state by CAD system, that is, the curve or surface possesses its optimal para- metric distribution. Finally, we give some numerical examples for demonstrating our results in important applications of judging the stable state of weights of the curves or surfaces and designing rational B6zier surfaces with compact derivative bounds.展开更多
In this paper, the smooth connection between two B-spline surfaces is discussed. First, a brief proof of some simple sufficient conditions of Go and G1 continuity is given. On this basis, a novel method for Go or G1 c...In this paper, the smooth connection between two B-spline surfaces is discussed. First, a brief proof of some simple sufficient conditions of Go and G1 continuity is given. On this basis, a novel method for Go or G1 connection between two adjacent B-spline surfaces is presented. A reparameterization step is firstly taken for one of the surfaces such that they have the same parameterization in v direction, then, adjust their boundary control vertices to make them Go or Gl connected. The GI connection parameter is determined by an optimization problem. Compared with the existed methods, our method is simple and easy to be used in practice.展开更多
In this paper the amalgamation of (I+1)×J tables under consistent significance is considered and the sufficient condition is obtained. The sufficient and necessary conditions are given to avoid the famous paradox...In this paper the amalgamation of (I+1)×J tables under consistent significance is considered and the sufficient condition is obtained. The sufficient and necessary conditions are given to avoid the famous paradoxes 'YAP' and 'YRP'. A practical algorithm is given to compute the critical value when pooling tables with odds ratio less than 1.展开更多
This paper presents a method of generating a parametric G^n blending surfacebased on reparameterizing the partial surface patches in the base surfaces on the basis of ErichHartmann method. This method is expressed as ...This paper presents a method of generating a parametric G^n blending surfacebased on reparameterizing the partial surface patches in the base surfaces on the basis of ErichHartmann method. This method is expressed as follows Firstly, the partial region near contact curvesin both base surfaces is reparameterized. The contact curves are used as the boundaries of thereparameterized partial region respectively. The reparameterized partial region in two base surfacesis called the reparameterized local base surfaces. Then the parametric G^n blending surface isgenerated by a linear combination of the reparameterized local base surface patches depending on oneof the common parameters. Therefore, generating a Parametric G^n Blending Surface between two basesurfaces is translated into generating a Parametric G^n Blending Surface between the tworeparameterized local base surfaces. This paper illustrates the method to generate the G^n blendingsurface with some constraints by generating a G^2 blending surface between the aerofoil and the bodyof a missile with the constraints of the forward and rear fringe curves. When the G^n blendingsurface with some constraints is generated, the partial region near contact curves in both basesurfaces is reparameterized, and the scale factors, offset, balance factor and thumb weight aredefined by meeting the constraints through using an optimization method. Then the parametric G^nblending surface is generated by the linear combination of the reparameterized local base surfacepatches. The shape of the blending surface can be adjusted by changing the size of thereparameterized local base surface patches.展开更多
Deformable retinal image registration is crucial in clinical diagnosis and longitudinal studies of retinal diseases.Most existing deep deformable retinal image registration methods focus on fully convolutional network...Deformable retinal image registration is crucial in clinical diagnosis and longitudinal studies of retinal diseases.Most existing deep deformable retinal image registration methods focus on fully convolutional network(FCN)architecture design,which fails to model long-range dependencies among pixels-a significant factor in deformable retinal image registration.Transformers based on the self-attention mechanism,can capture global context dependencies,complementing local convolution.However,multi-scale spatial feature fusion and pixel-wise position selection are also crucial for the deformable retinal image registration,are often ignored by both FCNs and transformers.To fully leverage the merits of FCNs,multi-scale spatial attention and transformers,we propose a hierarchical hybrid architecture,reparameterized multi-scale transformer(RMFormer),for deformable retinal image registration.In RMFormer,we specifically develop a reparameterized multi-scale spatial attention to adaptively fuse multi-scale spatial features,with the assistance of the reparameterizing technique,thereby highlighting informative pixel-wise positions in a lightweight manner.The experimental results on two publicly available datasets demonstrate the superiority of our RMFormer over state-of-the-art methods and show that it is data-efficient in a limited medical image regime.Additionally,we are the first to provide a visualization analysis to explain how our proposed method affects the deformable retinal image registration process.The source code of our work is available at https://github.com/Tloops/RMFormer.展开更多
Geological model compression is crucial for making large and complex models more manageable.By reducing the size of these models,compression techniques enable efficient storage,enhance computational efficiency,making ...Geological model compression is crucial for making large and complex models more manageable.By reducing the size of these models,compression techniques enable efficient storage,enhance computational efficiency,making it feasible to perform complex simulations and analyses in a shorter time.This is particularly important in applications such as reservoir management,groundwater hydrology,and geological carbon storage,where large geomodels with millions of grid cells are common.This study presents a comprehensive overview of previous work on geomodel compression and introduces several autoencoder-based deep-learning architectures for low-dimensional representation of modified Bruggefield geomodels.The compression and reconstruction efficiencies of autoencoders(AE),variational autoencoders(VAE),vector-quantized variational autoencoders(VQ-VAE),and vector-quantized variational autoencoders 2(VQ-VAE2)were tested and compared to the traditional singular value decomposition(SVD)method.Results show that the deep-learning-based approaches significantly outperform SVD,achieving higher compression ratios while maintaining or even exceeding the reconstruction quality.Notably,VQ-VAE2 achieves the highest compression ratio of 667:1 with a structural similarity index metric(SSIM)of 0.92,far surpassing the 10:1 compression ratio of SVD with a SSIM of 0.9.The result of this work shows that,unlike traditional approaches,which often rely on linear transformations and can struggle to capture complex,non-linear relationships within geological data,VQ-VAE's use of vector quantization helps in preserving high-resolution details and enhances the model's ability to generalize across varying geological complexities.展开更多
Forest is the largest carbon reservoir and carbon absorber on earth.Thus,mapping forest cover change accurately is of great significance to achieving the global carbon neutrality goal.Accurate forest change informatio...Forest is the largest carbon reservoir and carbon absorber on earth.Thus,mapping forest cover change accurately is of great significance to achieving the global carbon neutrality goal.Accurate forest change information could be acquired by deep learning methods using high-resolution remote sensing images.However,deforestation detection based on deep learning on a large-scale region with high-resolution images required huge computational resources.Therefore,there was an urgent need for a fast and accurate deforestation detection model.In this study,we proposed an interesting but effective re-parameterization deforestation detection model,named RepDDNet.Unlike other existing models designed for deforestation detection,the main feature of RepDDNet was its decoupling feature,which means that it allowed the multi-branch structure in the training stages to be converted into a plain structure in the inference stage,thus the computation efficiency can be significantly improved in the inference stage while maintaining the accuracy unchanged.A large-scale experiment was carried out in Ankang city with 2-meter high-resolution remote sensing images(the total area of it was over 20,000 square kilometers),and the result indicated that the model computation efficiency could be improved by nearly 30%compared with the model without re-parameterization.Additionally,compared with other lightweight models,RepDDNet also displayed a trade-off between accuracy and computation efficiency.展开更多
Tea shoot segmentation is crucial for the automation of high-quality tea plucking.However,accurate segmentation of tea shoots in unstructured and complex environments presents significant challenges due to the small s...Tea shoot segmentation is crucial for the automation of high-quality tea plucking.However,accurate segmentation of tea shoots in unstructured and complex environments presents significant challenges due to the small size of the targets and the similarity in color between the shoots and their background.To address these challenges and achieve accurate recognition of tea shoots in complex settings,an advanced tea shoot segmentation network model is proposed based on You Only Look Once version 8 segmentation(YOLOv8-seg)network model.Firstly,to enhance the model’s segmentation capability for small targets,this study designed a feature fusion network that incorporates shallow,large-scale features extracted by the backbone network.Subsequently,the features extracted at different scales by the backbone network are fused to obtain both global and local features,thereby enhancing the overall information representation capability of the features.Furthermore,the Efficient Channel Attention mechanism was integrated into the feature fusion process and combined with a reparameterization technique to refine and improve the efficiency of the fusion process.Finally,Wise-IoU with a dynamic non-monotonic aggregation mechanism was employed to assign varying gradient gains to anchor boxes of differing qualities.Experimental results demonstrate that the improved network model increases the AP50 of box and mask by 4.33%and 4.55%,respectively,while maintaining a smaller parameter count and reduced computational demand.Compared to other classical segmentation algorithms models,the proposed model excels in tea shoot segmentation.Overall,the advancements proposed in this study effectively segment tea shoots in complex environments,offering significant theoretical and practical contributions to the automated plucking of high-quality tea.展开更多
基金the Key Project of Zhejiang Provincial Natural Science Foundation under Grants LD21F020001,Z20F020022the National Natural Science Foundation of China under Grants 62072340,62076185the Major Project of Wenzhou Natural Science Foundation under Grants 2021HZSY0071,ZS2022001.
文摘Various deep learning models have been proposed for the accurate assisted diagnosis of early-stage Alzheimer’s disease(AD).Most studies predominantly employ Convolutional Neural Networks(CNNs),which focus solely on local features,thus encountering difficulties in handling global features.In contrast to natural images,Structural Magnetic Resonance Imaging(sMRI)images exhibit a higher number of channel dimensions.However,during the Position Embedding stage ofMulti Head Self Attention(MHSA),the coded information related to the channel dimension is disregarded.To tackle these issues,we propose theRepBoTNet-CESA network,an advanced AD-aided diagnostic model that is capable of learning local and global features simultaneously.It combines the advantages of CNN networks in capturing local information and Transformer networks in integrating global information,reducing computational costs while achieving excellent classification performance.Moreover,it uses the Cubic Embedding Self Attention(CESA)proposed in this paper to incorporate the channel code information,enhancing the classification performance within the Transformer structure.Finally,the RepBoTNet-CESA performs well in various AD-aided diagnosis tasks,with an accuracy of 96.58%,precision of 97.26%,and recall of 96.23%in the AD/NC task;an accuracy of 92.75%,precision of 92.84%,and recall of 93.18%in the EMCI/NC task;and an accuracy of 80.97%,precision of 83.86%,and recall of 80.91%in the AD/EMCI/LMCI/NC task.This demonstrates that RepBoTNet-CESA delivers outstanding outcomes in various AD-aided diagnostic tasks.Furthermore,our study has shown that MHSA exhibits superior performance compared to conventional attention mechanisms in enhancing ResNet performance.Besides,the Deeper RepBoTNet-CESA network fails to make further progress in AD-aided diagnostic tasks.
基金Project supported by the National Basic Research Program (973) of China (No. 2004CB719400)the National Natural Science Founda-tion of China (Nos. 60673031 and 60333010)the National Natural Science Foundation for Innovative Research Groups of China (No. 60021201)
文摘Applying homogeneous coordinates, we extend a newly appeared algorithm of best constrained multi-degree reduction for polynomial Bezier curves to the algorithms of constrained multi-degree reduction for rational Bezier curves. The idea is introducing two criteria, variance criterion and ratio criterion, for reparameterization of rational Bezier curves, which are used to make uniform the weights of the rational Bezier curves as accordant as possible, and then do multi-degree reduction for each component in homogeneous coordinates. Compared with the two traditional algorithms of ＂cancelling the best linear common divisor＂ and ＂shifted Chebyshev polynomial＂, the two new algorithms presented here using reparameterization have advantages of simplicity and fast computing, being able to preserve high degrees continuity at the end points of the curves, do multi-degree reduction at one time, and have good approximating effect.
文摘A series of advantages of single difference (SD) and undifferenced (ZD) models are given as compared with the double difference (DD) model. However, rank defects exist in SD and ZD models. The reparameterization method is provided to resolve this rank defect problem by estimating some combinations of the unknowns rather than the unknowns themselves. The reparameterization of SD and ZD functional models is discussed in detail with their stochastic models. The theoretical confirmation of the equivalence of undifferenced and differenced models is described in a straightforward way. The relationship between SD and ZD residuals is given and verified for some special purposes, e.g. research on the stochastical properties of GPS observations.
基金supported by the National Natural Science Foundation of China under Grant Nos.61972120 and 62021002 and the Haihe Lab of ITAI Project(22HHXCJC00006)supported by the National Key Research and Development Program of China under Grant Nos.2020YFB1708900 and TC190A4DA/3.
文摘Coincidence detection of two curves or two surfaces has wide application in computer-aided design(CAD)and computer-aided geometric design(CAGD).Proper reparameterization is the most complicated part in the detection.This paper presents and proves the efficient and necessary coincidence condition for two rational Bézier curves in a new way.It also proposes an effective and efficient proper reparameterization method,Algorithm 1,for detecting a rational Bézier curve which can be degenerated into a new one of a lower degree.A numerical proper reparameterization method,Algorithm 2,and examples are also presented.Algorithm 1 is up to ten times faster than other prevailing methods,and Algorithm 2 is twice as fast and half as close as other prevailing methods.New CAD systems using Algorithm 1 and Algorithm 2 will hold accuracy and little computation time.
基金Supported by the National Nature Science Foundations of China(61070065)
文摘Many works have investigated the problem of reparameterizing rational B^zier curves or surfaces via MSbius transformation to adjust their parametric distribution as well as weights, such that the maximal ratio of weights becomes smallerthat some algebraic and computational properties of the curves or surfaces can be improved in a way. However, it is an indication of veracity and optimization of the reparameterization to do prior to judge whether the maximal ratio of weights reaches minimum, and verify the new weights after MSbius transfor- mation. What＇s more the users of computer aided design softwares may require some guidelines for designing rational B6zier curves or surfaces with the smallest ratio of weights. In this paper we present the necessary and sufficient conditions that the maximal ratio of weights of the curves or surfaces reaches minimum and also describe it by using weights succinctly and straightway. The weights being satisfied these conditions are called being in the stable state. Applying such conditions, any giving rational B6zier curve or surface can automatically be adjusted to come into the stable state by CAD system, that is, the curve or surface possesses its optimal para- metric distribution. Finally, we give some numerical examples for demonstrating our results in important applications of judging the stable state of weights of the curves or surfaces and designing rational B6zier surfaces with compact derivative bounds.
基金Supported by the Natural Science Foundation of Hebei Province(No.F2012202041)Youth Research Foundation of Science and Technology of Hebei Education Departmen(No.Q2012022)
文摘In this paper, the smooth connection between two B-spline surfaces is discussed. First, a brief proof of some simple sufficient conditions of Go and G1 continuity is given. On this basis, a novel method for Go or G1 connection between two adjacent B-spline surfaces is presented. A reparameterization step is firstly taken for one of the surfaces such that they have the same parameterization in v direction, then, adjust their boundary control vertices to make them Go or Gl connected. The GI connection parameter is determined by an optimization problem. Compared with the existed methods, our method is simple and easy to be used in practice.
文摘In this paper the amalgamation of (I+1)×J tables under consistent significance is considered and the sufficient condition is obtained. The sufficient and necessary conditions are given to avoid the famous paradoxes 'YAP' and 'YRP'. A practical algorithm is given to compute the critical value when pooling tables with odds ratio less than 1.
文摘This paper presents a method of generating a parametric G^n blending surfacebased on reparameterizing the partial surface patches in the base surfaces on the basis of ErichHartmann method. This method is expressed as follows Firstly, the partial region near contact curvesin both base surfaces is reparameterized. The contact curves are used as the boundaries of thereparameterized partial region respectively. The reparameterized partial region in two base surfacesis called the reparameterized local base surfaces. Then the parametric G^n blending surface isgenerated by a linear combination of the reparameterized local base surface patches depending on oneof the common parameters. Therefore, generating a Parametric G^n Blending Surface between two basesurfaces is translated into generating a Parametric G^n Blending Surface between the tworeparameterized local base surfaces. This paper illustrates the method to generate the G^n blendingsurface with some constraints by generating a G^2 blending surface between the aerofoil and the bodyof a missile with the constraints of the forward and rear fringe curves. When the G^n blendingsurface with some constraints is generated, the partial region near contact curves in both basesurfaces is reparameterized, and the scale factors, offset, balance factor and thumb weight aredefined by meeting the constraints through using an optimization method. Then the parametric G^nblending surface is generated by the linear combination of the reparameterized local base surfacepatches. The shape of the blending surface can be adjusted by changing the size of thereparameterized local base surface patches.
基金supported in part by the National Natural Science Foundation of China(No.82272086)the Leading Goose Program of Zhejiang,China(No.2023C03079)the Shenzhen Natural Science Fund,China(No.JCYJ20200109140820699).
文摘Deformable retinal image registration is crucial in clinical diagnosis and longitudinal studies of retinal diseases.Most existing deep deformable retinal image registration methods focus on fully convolutional network(FCN)architecture design,which fails to model long-range dependencies among pixels-a significant factor in deformable retinal image registration.Transformers based on the self-attention mechanism,can capture global context dependencies,complementing local convolution.However,multi-scale spatial feature fusion and pixel-wise position selection are also crucial for the deformable retinal image registration,are often ignored by both FCNs and transformers.To fully leverage the merits of FCNs,multi-scale spatial attention and transformers,we propose a hierarchical hybrid architecture,reparameterized multi-scale transformer(RMFormer),for deformable retinal image registration.In RMFormer,we specifically develop a reparameterized multi-scale spatial attention to adaptively fuse multi-scale spatial features,with the assistance of the reparameterizing technique,thereby highlighting informative pixel-wise positions in a lightweight manner.The experimental results on two publicly available datasets demonstrate the superiority of our RMFormer over state-of-the-art methods and show that it is data-efficient in a limited medical image regime.Additionally,we are the first to provide a visualization analysis to explain how our proposed method affects the deformable retinal image registration process.The source code of our work is available at https://github.com/Tloops/RMFormer.
基金supported by ConocoPhillips through the DataDriven Intelligent Engineering and Characterization(DICE)research group.
文摘Geological model compression is crucial for making large and complex models more manageable.By reducing the size of these models,compression techniques enable efficient storage,enhance computational efficiency,making it feasible to perform complex simulations and analyses in a shorter time.This is particularly important in applications such as reservoir management,groundwater hydrology,and geological carbon storage,where large geomodels with millions of grid cells are common.This study presents a comprehensive overview of previous work on geomodel compression and introduces several autoencoder-based deep-learning architectures for low-dimensional representation of modified Bruggefield geomodels.The compression and reconstruction efficiencies of autoencoders(AE),variational autoencoders(VAE),vector-quantized variational autoencoders(VQ-VAE),and vector-quantized variational autoencoders 2(VQ-VAE2)were tested and compared to the traditional singular value decomposition(SVD)method.Results show that the deep-learning-based approaches significantly outperform SVD,achieving higher compression ratios while maintaining or even exceeding the reconstruction quality.Notably,VQ-VAE2 achieves the highest compression ratio of 667:1 with a structural similarity index metric(SSIM)of 0.92,far surpassing the 10:1 compression ratio of SVD with a SSIM of 0.9.The result of this work shows that,unlike traditional approaches,which often rely on linear transformations and can struggle to capture complex,non-linear relationships within geological data,VQ-VAE's use of vector quantization helps in preserving high-resolution details and enhances the model's ability to generalize across varying geological complexities.
基金supported by the Shenzhen Science and Technology Innovation Project(No.ZDSYS20210623091808026)supported in part by the National Natural Science Foundation of China(General Program,No.42071351)+1 种基金the National Key Research and Development Program of China(No.2020YFA0608501)the Chongqing Science and Technology Bureau technology innovation and application development special(cstc2021jscx-gksb0116).
文摘Forest is the largest carbon reservoir and carbon absorber on earth.Thus,mapping forest cover change accurately is of great significance to achieving the global carbon neutrality goal.Accurate forest change information could be acquired by deep learning methods using high-resolution remote sensing images.However,deforestation detection based on deep learning on a large-scale region with high-resolution images required huge computational resources.Therefore,there was an urgent need for a fast and accurate deforestation detection model.In this study,we proposed an interesting but effective re-parameterization deforestation detection model,named RepDDNet.Unlike other existing models designed for deforestation detection,the main feature of RepDDNet was its decoupling feature,which means that it allowed the multi-branch structure in the training stages to be converted into a plain structure in the inference stage,thus the computation efficiency can be significantly improved in the inference stage while maintaining the accuracy unchanged.A large-scale experiment was carried out in Ankang city with 2-meter high-resolution remote sensing images(the total area of it was over 20,000 square kilometers),and the result indicated that the model computation efficiency could be improved by nearly 30%compared with the model without re-parameterization.Additionally,compared with other lightweight models,RepDDNet also displayed a trade-off between accuracy and computation efficiency.
基金supported by the National Natural Science Foundation of China(Grant No.52305289,Grant No.U23A20175)the Open Fund of State Key Laboratory of Tea Plant Biology and Utilization(Grant No.NKLTOF20240103)the earmarked fund for CARS-19 and the Open Project Program of Fujian Key Laboratory of Big Data Application and Intellectualization for Tea Industry,Wuyi University.
文摘Tea shoot segmentation is crucial for the automation of high-quality tea plucking.However,accurate segmentation of tea shoots in unstructured and complex environments presents significant challenges due to the small size of the targets and the similarity in color between the shoots and their background.To address these challenges and achieve accurate recognition of tea shoots in complex settings,an advanced tea shoot segmentation network model is proposed based on You Only Look Once version 8 segmentation(YOLOv8-seg)network model.Firstly,to enhance the model’s segmentation capability for small targets,this study designed a feature fusion network that incorporates shallow,large-scale features extracted by the backbone network.Subsequently,the features extracted at different scales by the backbone network are fused to obtain both global and local features,thereby enhancing the overall information representation capability of the features.Furthermore,the Efficient Channel Attention mechanism was integrated into the feature fusion process and combined with a reparameterization technique to refine and improve the efficiency of the fusion process.Finally,Wise-IoU with a dynamic non-monotonic aggregation mechanism was employed to assign varying gradient gains to anchor boxes of differing qualities.Experimental results demonstrate that the improved network model increases the AP50 of box and mask by 4.33%and 4.55%,respectively,while maintaining a smaller parameter count and reduced computational demand.Compared to other classical segmentation algorithms models,the proposed model excels in tea shoot segmentation.Overall,the advancements proposed in this study effectively segment tea shoots in complex environments,offering significant theoretical and practical contributions to the automated plucking of high-quality tea.
