Human pose estimation aims to localize the body joints from image or video data.With the development of deeplearning,pose estimation has become a hot research topic in the field of computer vision.In recent years,huma...Human pose estimation aims to localize the body joints from image or video data.With the development of deeplearning,pose estimation has become a hot research topic in the field of computer vision.In recent years,humanpose estimation has achieved great success in multiple fields such as animation and sports.However,to obtainaccurate positioning results,existing methods may suffer from large model sizes,a high number of parameters,and increased complexity,leading to high computing costs.In this paper,we propose a new lightweight featureencoder to construct a high-resolution network that reduces the number of parameters and lowers the computingcost.We also introduced a semantic enhancement module that improves global feature extraction and networkperformance by combining channel and spatial dimensions.Furthermore,we propose a dense connected spatialpyramid pooling module to compensate for the decrease in image resolution and information loss in the network.Finally,ourmethod effectively reduces the number of parameters and complexitywhile ensuring high performance.Extensive experiments show that our method achieves a competitive performance while dramatically reducing thenumber of parameters,and operational complexity.Specifically,our method can obtain 89.9%AP score on MPIIVAL,while the number of parameters and the complexity of operations were reduced by 41%and 36%,respectively.展开更多
High Spatial and Spectral Resolution(HSSR)remote-sensing images can provide rich spectral bands and detailed ground information,but there is a relative lack of research on this new type of remote-sensing data.Although...High Spatial and Spectral Resolution(HSSR)remote-sensing images can provide rich spectral bands and detailed ground information,but there is a relative lack of research on this new type of remote-sensing data.Although there are already some HSSR datasets for deep learning model training and testing,the data volume of these datasets is small,resulting in low classification accuracy and weak generalization ability of the trained models.In this paper,an HSSR dataset Luojia-HSSR is constructed based on aerial hyperspectral imagery of southern Shenyang City of Liaoning Province in China.To our knowledge,it is the largest HSSR dataset to date,with 6438 pairs of 256×256 sized samples(including 3480 pairs in the training set,2209 pairs in the test set,and 749 pairs in the validation set),covering area of 161 km2 with spatial resolution 0.75 m,249 Visible and Near-Infrared(VNIR)spectral bands,and corresponding to 23 classes of field-validated ground coverage.It is an ideal experimental data for spatial-spectral feature extraction.Furthermore,a new deep learning model 3D-HRNet for interpreting HSSR images is proposed.The conv-neck in HRNet is modified to better mine the spatial information of the images.Then,a 3D convolution module with attention mechanism is designed to capture the global-local fine spectral information simultaneously.Subsequently,the 3D convolution is inserted into the HRNet to optimize the performance.The experiments show that the 3D-HRNet model has good interpreting ability for the Luojia-HSSR dataset with the Frequency Weighted Intersection over Union(FWIoU)reaching 80.54%,indicating that the Luojia-HSSR dataset constructed in this paper and the proposed 3D-HRnet model have good applicable prospects for processing HSSR remote sensing images.展开更多
Artifi cial neural network is a kind of artificial intelligence method to simulate the function of human brain, and deep learning technology can establish a depth network model with hierarchical structure on the basis...Artifi cial neural network is a kind of artificial intelligence method to simulate the function of human brain, and deep learning technology can establish a depth network model with hierarchical structure on the basis of artificial neural network. Deep learning brings new development direction to artificial neural network. Convolution neural network is a new artificial neural network method, which combines artificial neural network and deep learning technology, and this new neural network is widely used in many fields of computer vision. Modern image recognition algorithm requires classifi cation system to adapt to different types of tasks, and deep network and convolution neural network is a hot research topic in neural networks. According to the characteristics of satellite digital image, we use the convolution neural network to classify the image, which combines texture features with spectral features. The experimental results show that the convolution neural network algorithm can effectively classify the image.展开更多
As digital image techniques have been widely used, the requirements for high-resolution images become increasingly stringent. Traditional single-frame interpolation techniques cannot add new high frequency information...As digital image techniques have been widely used, the requirements for high-resolution images become increasingly stringent. Traditional single-frame interpolation techniques cannot add new high frequency information to the expanded images, and cannot improve resolution in deed. Multiframe-based techniques are effective ways for high-resolution image reconstruction, but their computation complexities and the difficulties in achieving image sequences limit their applications. An original method using an artificial neural network is proposed in this paper. Using the inherent merits in neural network, we can establish the mapping between high frequency components in low-resolution images and high-resolution images. Example applications and their results demonstrated the images reconstructed by our method are aesthetically and quantitatively (using the criteria of MSE and MAE) superior to the images acquired by common methods. Even for infrared images this method can give satisfactory results with high definition. In addition, a single-layer linear neural network is used in this paper, the computational complexity is very low, and this method can be realized in real time.展开更多
We develop a high resolution ground penetrating radar system （LANRCS-GPR） based on the E5071B Vector Network Analyzer （VNA）. This system takes advantage of a wideband and adjustable frequency domain ground penetra...We develop a high resolution ground penetrating radar system （LANRCS-GPR） based on the E5071B Vector Network Analyzer （VNA）. This system takes advantage of a wideband and adjustable frequency domain ground penetrating radar system and adds the characteristics of a network analyzer with ultra-wideband and high precision measurement. It adopts the LAN mode to concatenate system control that reduces construction cost and makes the system easy to expand. The high resolution ground penetrating radar system carries out real time imaging using F-K migration with high calculation efficiency. The experiment results of the system indicate that the LANRCS-GPR system provides high resolution and precision, high signal-to-noise ratio, and great dynamic range. Furthermore, the LANRCS-GPR system is flexible and reliable to operate with easy to expand system functions. The research and development of the LANRCS-GPR provide the theoretical and experimental foundation for future frequency domain ground penetrating radar production and also can serve as an experimental platform with high data gathering precision, enormous information capability, wide application, and convenient operation for electromagnetic wave research and electromagnetic exploration.展开更多
Small-object detection has long been a challenge.High-megapixel cameras are used to solve this problem in industries.However,current detectors are inefficient for high-resolution images.In this work,we propose a new m...Small-object detection has long been a challenge.High-megapixel cameras are used to solve this problem in industries.However,current detectors are inefficient for high-resolution images.In this work,we propose a new module called Pre-Locate Net,which is a plug-and-play structure that can be combined with most popular detectors.We inspire the use of classification ideas to obtain candidate regions in images,greatly reducing the amount of calculation,and thus achieving rapid detection in high-resolution images.Pre-Locate Net mainly includes two parts,candidate region classification and behavior classification.Candidate region classification is used to obtain a candidate region,and behavior classification is used to estimate the scale of an object.Different follow-up processing is adopted according to different scales to balance the variance of the network input.Different from the popular candidate region generation method,we abandon the idea of regression of a bounding box and adopt the concept of classification,so as to realize the prediction of a candidate region in the shallow network.We build a high-resolution dataset of aircraft and landing gears covering complex scenes to verify the effectiveness of our method.Compared to state-of-the-art detectors(e.g.,Guided Anchoring,Libra-RCNN,and FASF),our method achieves the best m AP of 94.5 on 1920×1080 images at 16.7 FPS.展开更多
Light-field imaging has wide applications in various domains,including microscale life science imaging,mesoscale neuroimaging,and macroscale fluid dynamics imaging.The development of deep learning-based reconstruction...Light-field imaging has wide applications in various domains,including microscale life science imaging,mesoscale neuroimaging,and macroscale fluid dynamics imaging.The development of deep learning-based reconstruction methods has greatly facilitated high-resolution light-field image processing,however,current deep learning-based light-field reconstruction methods have predominantly concentrated on the microscale.Considering the multiscale imaging capacity of light-field technique,a network that can work over variant scales of light-field image reconstruction will significantly benefit the development of volumetric imaging.Unfortunately,to our knowledge,no one has reported a universal high-resolution light-field image reconstruction algorithm that is compatible with microscale,mesoscale,and macroscale.To fill this gap,we present a real-time and universal network(RTU-Net)to reconstruct high-resolution light-field images at any scale.RTU-Net,as the first network that works over multiscale light-field image reconstruction,employs an adaptive loss function based on generative adversarial theory and consequently exhibits strong generalization capability.We comprehensively assessed the performance of RTU-Net through the reconstruction of multiscale light-field images,including microscale tubulin and mitochondrion dataset,mesoscale synthetic mouse neuro dataset,and macroscale light-field particle imaging velocimetry dataset.The results indicated that RTU-Net has achieved real-time and high-resolution light-field image reconstruction for volume sizes ranging from 300μm×300μm×12μm to 25 mm×25 mm×25 mm,and demonstrated higher resolution when compared with recently reported light-field reconstruction networks.The high-resolution,strong robustness,high efficiency,and especially the general applicability of RTU-Net will significantly deepen our insight into high-resolution and volumetric imaging.展开更多
The high resolution 3D nonlinear integrated inversion method is based on nonlinear theory. Under layer control, the log data from several wells （or all wells） in the study area and seismic trace data adjacent to the...The high resolution 3D nonlinear integrated inversion method is based on nonlinear theory. Under layer control, the log data from several wells （or all wells） in the study area and seismic trace data adjacent to the wells are input to a network with multiple inputs and outputs and are integratedly trained to obtain an adaptive weight function of the entire study area. Integrated nonlinear mapping relationships are built and updated by the lateral and vertical geologic variations of the reservoirs. Therefore, the inversion process and its inversion results can be constrained and controlled and a stable seismic inversion section with high resolution with velocity inversion, impedance inversion, and density inversion sections, can be gained. Good geologic effects have been obtained in model computation tests and real data processing, which verified that this method has high precision, good practicality, and can be used for quantitative reservoir analysis.展开更多
Increasing the resolution of seismic data has long been a major topic in seismic exploration.Due to the effect of high-frequency noises,traditional methods could only improve the resolution limitedly.To end this,this ...Increasing the resolution of seismic data has long been a major topic in seismic exploration.Due to the effect of high-frequency noises,traditional methods could only improve the resolution limitedly.To end this,this paper newly proposed a high-resolution seismic data processing method based on welleseismic combination after summarizing the research status on high resolution.Synthetic record and seismogram are similar in effective signals but dissimilar in noises.Their effective signals are regular and noises are irregular.And they are similar in adjacent frequency.Based on these“three-regularity”characteristics,the relationship between synthetic record and seismogram was established using the neural network algorithm.Then,the corresponding extrapolation algorithm was proposed based on the self-adaptive geological and geophysical variation of multi-layer network structure.And a model was established by virtue of this method and the theoretical simulation was carried out.In addition,it was tested from the aspects of frequency component and amplitude energy recovery,phase correction,regularity elimination and stochastic noise.And the following research results were obtained.First,this new method can extract high-frequency information as much as possible and remain middle and low-frequency effective information while eliminating the noises.Second,in this method,the idea of traditional methods to denoisefirst and then expand frequency is changed completely and the limitation of traditional methods is broken.It establishes the idea of expanding frequency and denoising simultaneously and increases the resolution to the uttermost.Third,this new method has been applied to a variety of reservoir descriptions and the high-resolution processing results have been improved significantly in precision and accuracy.展开更多
目的步态识别是交通管理、监控安防领域的关键技术,为了解决现有步态识别算法无法充分捕捉和利用人体生物特征,在协变量干扰下模型精度降低的问题,本文提出一种深度提取和融合步态特征与身形特征的高精度步态识别方法。方法首先使用高...目的步态识别是交通管理、监控安防领域的关键技术,为了解决现有步态识别算法无法充分捕捉和利用人体生物特征,在协变量干扰下模型精度降低的问题,本文提出一种深度提取和融合步态特征与身形特征的高精度步态识别方法。方法首先使用高分辨率网络(high resolution network,HRNet)提取出人体骨架关键点;以残差神经网络ResNet-50(residual network)为主干,利用深度残差模块的复杂特征学习能力,从骨架信息中充分提取相对稳定的身形特征与提供显性高效运动本质表达的步态特征;设计多分支特征融合(multi-branch feature fusion,MFF)模块,进行不同通道间的尺寸对齐与权重优化,通过动态权重矩阵调节各分支贡献,把身形特征和步态特征融合为区分度更强的总体特征。结果室内数据集采用跨视角多状态CASIA-B(Institute of Automation,Chinese Academy of Sciences)数据集,本文方法在跨视角实验中表现稳健;在多状态实验中,常规组的识别准确率为94.52%,外套干扰组在同类算法中的识别性能最佳。在开放场景数据集中,模型同样体现出较高的泛化能力,相比于现有算法,本文方法的准确率提升了4.1%。结论本文设计的步态识别方法充分利用了深度残差模块的特征提取能力与多特征融合的互补优势,面向复杂识别场景仍具有较高的模型识别精度与泛化能力。展开更多
基金the National Natural Science Foundation of China(Grant Number 62076246).
文摘Human pose estimation aims to localize the body joints from image or video data.With the development of deeplearning,pose estimation has become a hot research topic in the field of computer vision.In recent years,humanpose estimation has achieved great success in multiple fields such as animation and sports.However,to obtainaccurate positioning results,existing methods may suffer from large model sizes,a high number of parameters,and increased complexity,leading to high computing costs.In this paper,we propose a new lightweight featureencoder to construct a high-resolution network that reduces the number of parameters and lowers the computingcost.We also introduced a semantic enhancement module that improves global feature extraction and networkperformance by combining channel and spatial dimensions.Furthermore,we propose a dense connected spatialpyramid pooling module to compensate for the decrease in image resolution and information loss in the network.Finally,ourmethod effectively reduces the number of parameters and complexitywhile ensuring high performance.Extensive experiments show that our method achieves a competitive performance while dramatically reducing thenumber of parameters,and operational complexity.Specifically,our method can obtain 89.9%AP score on MPIIVAL,while the number of parameters and the complexity of operations were reduced by 41%and 36%,respectively.
基金supported by the Major Program of the National Natural Science Foundation of China[grant number 92038301]The research was also supported by the National Natural Science Foundation of China[grant number 41971295]+1 种基金the Foundation for Innovative Research Groups of the Natural Science Foundation of Hubei Province[grant number 2020CFA003]the Special Fund of Hubei Luojia Laboratory.
文摘High Spatial and Spectral Resolution(HSSR)remote-sensing images can provide rich spectral bands and detailed ground information,but there is a relative lack of research on this new type of remote-sensing data.Although there are already some HSSR datasets for deep learning model training and testing,the data volume of these datasets is small,resulting in low classification accuracy and weak generalization ability of the trained models.In this paper,an HSSR dataset Luojia-HSSR is constructed based on aerial hyperspectral imagery of southern Shenyang City of Liaoning Province in China.To our knowledge,it is the largest HSSR dataset to date,with 6438 pairs of 256×256 sized samples(including 3480 pairs in the training set,2209 pairs in the test set,and 749 pairs in the validation set),covering area of 161 km2 with spatial resolution 0.75 m,249 Visible and Near-Infrared(VNIR)spectral bands,and corresponding to 23 classes of field-validated ground coverage.It is an ideal experimental data for spatial-spectral feature extraction.Furthermore,a new deep learning model 3D-HRNet for interpreting HSSR images is proposed.The conv-neck in HRNet is modified to better mine the spatial information of the images.Then,a 3D convolution module with attention mechanism is designed to capture the global-local fine spectral information simultaneously.Subsequently,the 3D convolution is inserted into the HRNet to optimize the performance.The experiments show that the 3D-HRNet model has good interpreting ability for the Luojia-HSSR dataset with the Frequency Weighted Intersection over Union(FWIoU)reaching 80.54%,indicating that the Luojia-HSSR dataset constructed in this paper and the proposed 3D-HRnet model have good applicable prospects for processing HSSR remote sensing images.
文摘Artifi cial neural network is a kind of artificial intelligence method to simulate the function of human brain, and deep learning technology can establish a depth network model with hierarchical structure on the basis of artificial neural network. Deep learning brings new development direction to artificial neural network. Convolution neural network is a new artificial neural network method, which combines artificial neural network and deep learning technology, and this new neural network is widely used in many fields of computer vision. Modern image recognition algorithm requires classifi cation system to adapt to different types of tasks, and deep network and convolution neural network is a hot research topic in neural networks. According to the characteristics of satellite digital image, we use the convolution neural network to classify the image, which combines texture features with spectral features. The experimental results show that the convolution neural network algorithm can effectively classify the image.
文摘As digital image techniques have been widely used, the requirements for high-resolution images become increasingly stringent. Traditional single-frame interpolation techniques cannot add new high frequency information to the expanded images, and cannot improve resolution in deed. Multiframe-based techniques are effective ways for high-resolution image reconstruction, but their computation complexities and the difficulties in achieving image sequences limit their applications. An original method using an artificial neural network is proposed in this paper. Using the inherent merits in neural network, we can establish the mapping between high frequency components in low-resolution images and high-resolution images. Example applications and their results demonstrated the images reconstructed by our method are aesthetically and quantitatively (using the criteria of MSE and MAE) superior to the images acquired by common methods. Even for infrared images this method can give satisfactory results with high definition. In addition, a single-layer linear neural network is used in this paper, the computational complexity is very low, and this method can be realized in real time.
基金This project was supported by National Natural Science Foundation of china (No.40474042)
文摘We develop a high resolution ground penetrating radar system （LANRCS-GPR） based on the E5071B Vector Network Analyzer （VNA）. This system takes advantage of a wideband and adjustable frequency domain ground penetrating radar system and adds the characteristics of a network analyzer with ultra-wideband and high precision measurement. It adopts the LAN mode to concatenate system control that reduces construction cost and makes the system easy to expand. The high resolution ground penetrating radar system carries out real time imaging using F-K migration with high calculation efficiency. The experiment results of the system indicate that the LANRCS-GPR system provides high resolution and precision, high signal-to-noise ratio, and great dynamic range. Furthermore, the LANRCS-GPR system is flexible and reliable to operate with easy to expand system functions. The research and development of the LANRCS-GPR provide the theoretical and experimental foundation for future frequency domain ground penetrating radar production and also can serve as an experimental platform with high data gathering precision, enormous information capability, wide application, and convenient operation for electromagnetic wave research and electromagnetic exploration.
基金the National Science Fund for Distinguished Young Scholars of China (No. 51625501)the Aeronautical Science Foundation of China (No. 201946051002)
文摘Small-object detection has long been a challenge.High-megapixel cameras are used to solve this problem in industries.However,current detectors are inefficient for high-resolution images.In this work,we propose a new module called Pre-Locate Net,which is a plug-and-play structure that can be combined with most popular detectors.We inspire the use of classification ideas to obtain candidate regions in images,greatly reducing the amount of calculation,and thus achieving rapid detection in high-resolution images.Pre-Locate Net mainly includes two parts,candidate region classification and behavior classification.Candidate region classification is used to obtain a candidate region,and behavior classification is used to estimate the scale of an object.Different follow-up processing is adopted according to different scales to balance the variance of the network input.Different from the popular candidate region generation method,we abandon the idea of regression of a bounding box and adopt the concept of classification,so as to realize the prediction of a candidate region in the shallow network.We build a high-resolution dataset of aircraft and landing gears covering complex scenes to verify the effectiveness of our method.Compared to state-of-the-art detectors(e.g.,Guided Anchoring,Libra-RCNN,and FASF),our method achieves the best m AP of 94.5 on 1920×1080 images at 16.7 FPS.
基金supported by National Natural Science Foundation of China(12402336,82201637,U20A2070,and 12025202)National High-Level Talent Project(YQR23069)+6 种基金Natural Science Foundation of Jiangsu Province(BK20230876)the Young Elite Scientist Sponsorship Program by CAST(YESS20210238)Forwardlooking layout projects(1002-ILB24009)Zhejang Provincial Medical and Health Technology Project(Grant No.2024KY246,2025KY180)Scientific Research Foundation of Hangzhou City University(No.J-202402)Open Research Fund of the State Key Laboratory of Brain-Machine Intelligence,Zhejiang University(Grant No.BMI2400025)Hangzhou Science and Technology Bureau.
文摘Light-field imaging has wide applications in various domains,including microscale life science imaging,mesoscale neuroimaging,and macroscale fluid dynamics imaging.The development of deep learning-based reconstruction methods has greatly facilitated high-resolution light-field image processing,however,current deep learning-based light-field reconstruction methods have predominantly concentrated on the microscale.Considering the multiscale imaging capacity of light-field technique,a network that can work over variant scales of light-field image reconstruction will significantly benefit the development of volumetric imaging.Unfortunately,to our knowledge,no one has reported a universal high-resolution light-field image reconstruction algorithm that is compatible with microscale,mesoscale,and macroscale.To fill this gap,we present a real-time and universal network(RTU-Net)to reconstruct high-resolution light-field images at any scale.RTU-Net,as the first network that works over multiscale light-field image reconstruction,employs an adaptive loss function based on generative adversarial theory and consequently exhibits strong generalization capability.We comprehensively assessed the performance of RTU-Net through the reconstruction of multiscale light-field images,including microscale tubulin and mitochondrion dataset,mesoscale synthetic mouse neuro dataset,and macroscale light-field particle imaging velocimetry dataset.The results indicated that RTU-Net has achieved real-time and high-resolution light-field image reconstruction for volume sizes ranging from 300μm×300μm×12μm to 25 mm×25 mm×25 mm,and demonstrated higher resolution when compared with recently reported light-field reconstruction networks.The high-resolution,strong robustness,high efficiency,and especially the general applicability of RTU-Net will significantly deepen our insight into high-resolution and volumetric imaging.
基金supported by the Key Project of the National Natural Scientific Foundation(Grant No.40839909)
文摘The high resolution 3D nonlinear integrated inversion method is based on nonlinear theory. Under layer control, the log data from several wells （or all wells） in the study area and seismic trace data adjacent to the wells are input to a network with multiple inputs and outputs and are integratedly trained to obtain an adaptive weight function of the entire study area. Integrated nonlinear mapping relationships are built and updated by the lateral and vertical geologic variations of the reservoirs. Therefore, the inversion process and its inversion results can be constrained and controlled and a stable seismic inversion section with high resolution with velocity inversion, impedance inversion, and density inversion sections, can be gained. Good geologic effects have been obtained in model computation tests and real data processing, which verified that this method has high precision, good practicality, and can be used for quantitative reservoir analysis.
文摘Increasing the resolution of seismic data has long been a major topic in seismic exploration.Due to the effect of high-frequency noises,traditional methods could only improve the resolution limitedly.To end this,this paper newly proposed a high-resolution seismic data processing method based on welleseismic combination after summarizing the research status on high resolution.Synthetic record and seismogram are similar in effective signals but dissimilar in noises.Their effective signals are regular and noises are irregular.And they are similar in adjacent frequency.Based on these“three-regularity”characteristics,the relationship between synthetic record and seismogram was established using the neural network algorithm.Then,the corresponding extrapolation algorithm was proposed based on the self-adaptive geological and geophysical variation of multi-layer network structure.And a model was established by virtue of this method and the theoretical simulation was carried out.In addition,it was tested from the aspects of frequency component and amplitude energy recovery,phase correction,regularity elimination and stochastic noise.And the following research results were obtained.First,this new method can extract high-frequency information as much as possible and remain middle and low-frequency effective information while eliminating the noises.Second,in this method,the idea of traditional methods to denoisefirst and then expand frequency is changed completely and the limitation of traditional methods is broken.It establishes the idea of expanding frequency and denoising simultaneously and increases the resolution to the uttermost.Third,this new method has been applied to a variety of reservoir descriptions and the high-resolution processing results have been improved significantly in precision and accuracy.
文摘目的步态识别是交通管理、监控安防领域的关键技术,为了解决现有步态识别算法无法充分捕捉和利用人体生物特征,在协变量干扰下模型精度降低的问题,本文提出一种深度提取和融合步态特征与身形特征的高精度步态识别方法。方法首先使用高分辨率网络(high resolution network,HRNet)提取出人体骨架关键点;以残差神经网络ResNet-50(residual network)为主干,利用深度残差模块的复杂特征学习能力,从骨架信息中充分提取相对稳定的身形特征与提供显性高效运动本质表达的步态特征;设计多分支特征融合(multi-branch feature fusion,MFF)模块,进行不同通道间的尺寸对齐与权重优化,通过动态权重矩阵调节各分支贡献,把身形特征和步态特征融合为区分度更强的总体特征。结果室内数据集采用跨视角多状态CASIA-B(Institute of Automation,Chinese Academy of Sciences)数据集,本文方法在跨视角实验中表现稳健;在多状态实验中,常规组的识别准确率为94.52%,外套干扰组在同类算法中的识别性能最佳。在开放场景数据集中,模型同样体现出较高的泛化能力,相比于现有算法,本文方法的准确率提升了4.1%。结论本文设计的步态识别方法充分利用了深度残差模块的特征提取能力与多特征融合的互补优势,面向复杂识别场景仍具有较高的模型识别精度与泛化能力。
