Microseismic monitoring and signal recognition constitute critical technologies for accurately assessing rockburst risks and ensuring the safe construction of underground rock engineering.This study developed a"S...Microseismic monitoring and signal recognition constitute critical technologies for accurately assessing rockburst risks and ensuring the safe construction of underground rock engineering.This study developed a"Surface+Underground"microseismic intelligent monitoring system to evaluate dynamic disaster risks during construction at the Beishan High-level Radioactive Waste Geological Disposal Laboratory in China.The datasets of four typical one-dimensional time-domain microseismic signals,including rock fracture,blasting,TBM tunneling,and drilling,were constructed,and the BO-CNN-LSTM model is developed to identify and classify these signals..Based on the classification results,the typical time-frequency domain characteristics of the four types of signals are analyzed.The classification results of BO-CNN-LSTM,CNN,LSTM and CNN-LSTM models show that the recognition accuracy of the four models is 98.0%,85.7%,66.7%and 91.0%,respectively.Among all types,the four models demonstrate the highest effectiveness in identifying rock fracture and borehole signals.The findings further confirm that the BO-CNN-LSTM model efficiently recognizes and extracts microseismic signal features,demonstrating superior performance and stability in the classification task.Finally,the study suggests several future research directions,particularly in the areas of raw signal denoising,and automation of feature extraction.展开更多
Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empir...Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empirical achievements.In this paper,the first filter learning framework with convergence-guaranteed learning laws for end-to-end learning of deep CNNs is proposed.Novel update laws with convergence analysis are formulated based on the mathematical representation of each layer in convolutional neural networks.The proposed learning laws enable concurrent updates of weights across all layers of the deep convolutional neural network and the analysis shows that the training errors converge to certain bounds which are dependent on the approximation errors.Case studies are conducted on benchmark datasets and the results show that the proposed concurrent filter learning framework guarantees the convergence and offers more consistent and reliable results during training with a trade-off in performance compared to stochastic gradient descent methods.This framework represents a significant step towards enhancing the reliability and effectiveness of deep convolutional neural network by developing a theoretical analysis which allows practical implementation of the learning laws with automatic tuning of the learning rate to guarantee the convergence during training.展开更多
The neutron diffusion equation plays a pivotal role in nuclear reactor analysis.Nevertheless,employing the physics-informed neural network(PINN)method for its solution entails certain limitations.Conventional PINN app...The neutron diffusion equation plays a pivotal role in nuclear reactor analysis.Nevertheless,employing the physics-informed neural network(PINN)method for its solution entails certain limitations.Conventional PINN approaches generally utilize a fully connected network(FCN)architecture that is susceptible to overfitting,training instability,and gradient vanishing as the network depth increases.These challenges result in accuracy bottlenecks in the solution.In response to these issues,the residual-based resample physics-informed neural network(R2-PINN)is proposed.It is an improved PINN architecture that replaces the FCN with a convolutional neural network with a shortcut(S-CNN).It incorporates skip connections to facilitate gradient propagation between network layers.Additionally,the incorporation of the residual adaptive resampling(RAR)mechanism dynamically increases the number of sampling points.This,in turn,enhances the spatial representation capabilities and overall predictive accuracy of the model.The experimental results illustrate that our approach significantly improves the convergence capability of the model and achieves high-precision predictions of the physical fields.Compared with conventional FCN-based PINN methods,R 2-PINN effectively overcomes the limitations inherent in current methods.Thus,it provides more accurate and robust solutions for neutron diffusion equations.展开更多
Recently the depth estimation methods based on deep learning(DL)retain challenging to estimate a high-precision depth map in fringe projection structured light three-dimensional(3D)measurement with limited information...Recently the depth estimation methods based on deep learning(DL)retain challenging to estimate a high-precision depth map in fringe projection structured light three-dimensional(3D)measurement with limited information from a single-frame fringe pattern.In this letter,we proposed a FDSUNet++convolutional neural network(CNN),which consists of a UNet++base model,an improved squeeze-andexcitation(ISE)block,a Fourier transform(FT)data preprocessing block,and a discrete wavelet transform(DWT)block.The proposed ISE block can improve the ability of feature extraction and the designed FT data preprocessing block preserves the key features of the fringe pattern by FT.The introduced DWT block reduces the complexity and training cost of the model.By integrating these three blocks into the UNet++,it can better achieve depth estimation.Experimental results from two structured light datasets demonstrate that the proposed FDSUNet++outperforms the state-of-the-art networks,achieving the best performance in both qualitative and quantitative evaluation.展开更多
Skin cancer is the most prevalent cancer globally,primarily due to extensive exposure to Ultraviolet(UV)radiation.Early identification of skin cancer enhances the likelihood of effective treatment,as delays may lead t...Skin cancer is the most prevalent cancer globally,primarily due to extensive exposure to Ultraviolet(UV)radiation.Early identification of skin cancer enhances the likelihood of effective treatment,as delays may lead to severe tumor advancement.This study proposes a novel hybrid deep learning strategy to address the complex issue of skin cancer diagnosis,with an architecture that integrates a Vision Transformer,a bespoke convolutional neural network(CNN),and an Xception module.They were evaluated using two benchmark datasets,HAM10000 and Skin Cancer ISIC.On the HAM10000,the model achieves a precision of 95.46%,an accuracy of 96.74%,a recall of 96.27%,specificity of 96.00%and an F1-Score of 95.86%.It obtains an accuracy of 93.19%,a precision of 93.25%,a recall of 92.80%,a specificity of 92.89%and an F1-Score of 93.19%on the Skin Cancer ISIC dataset.The findings demonstrate that the model that was proposed is robust and trustworthy when it comes to the classification of skin lesions.In addition,the utilization of Explainable AI techniques,such as Grad-CAM visualizations,assists in highlighting the most significant lesion areas that have an impact on the decisions that are made by the model.展开更多
基金Project supported by the China Atomic Energy Authority(CAEA)through the Geological Disposal ProgramProjects(U24A20616,U24B2038)supported by the National Natural Science Foundation of ChinaProject(2025-05)supported by the Guangdong Provincial Water Conservancy Science and Technology Innovation Project,China。
摘要Microseismic monitoring and signal recognition constitute critical technologies for accurately assessing rockburst risks and ensuring the safe construction of underground rock engineering.This study developed a"Surface+Underground"microseismic intelligent monitoring system to evaluate dynamic disaster risks during construction at the Beishan High-level Radioactive Waste Geological Disposal Laboratory in China.The datasets of four typical one-dimensional time-domain microseismic signals,including rock fracture,blasting,TBM tunneling,and drilling,were constructed,and the BO-CNN-LSTM model is developed to identify and classify these signals..Based on the classification results,the typical time-frequency domain characteristics of the four types of signals are analyzed.The classification results of BO-CNN-LSTM,CNN,LSTM and CNN-LSTM models show that the recognition accuracy of the four models is 98.0%,85.7%,66.7%and 91.0%,respectively.Among all types,the four models demonstrate the highest effectiveness in identifying rock fracture and borehole signals.The findings further confirm that the BO-CNN-LSTM model efficiently recognizes and extracts microseismic signal features,demonstrating superior performance and stability in the classification task.Finally,the study suggests several future research directions,particularly in the areas of raw signal denoising,and automation of feature extraction.
基金supported by the Ministry of Education(MOE)Singapore,Academic Research Fund(AcRF)Tier 1(RG65/22)。
摘要Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empirical achievements.In this paper,the first filter learning framework with convergence-guaranteed learning laws for end-to-end learning of deep CNNs is proposed.Novel update laws with convergence analysis are formulated based on the mathematical representation of each layer in convolutional neural networks.The proposed learning laws enable concurrent updates of weights across all layers of the deep convolutional neural network and the analysis shows that the training errors converge to certain bounds which are dependent on the approximation errors.Case studies are conducted on benchmark datasets and the results show that the proposed concurrent filter learning framework guarantees the convergence and offers more consistent and reliable results during training with a trade-off in performance compared to stochastic gradient descent methods.This framework represents a significant step towards enhancing the reliability and effectiveness of deep convolutional neural network by developing a theoretical analysis which allows practical implementation of the learning laws with automatic tuning of the learning rate to guarantee the convergence during training.
摘要针对不同磁密幅值、频率、谐波组合等复杂激励工况下磁致伸缩建模面临的精准性问题,该文利用空间注意力机制(spatial attention mechanism,SAM)对传统的卷积神经网络(convolutional neural network,CNN)进行改进,将SAM嵌套入CNN网络中,建立SAMCNN改进型网络。再结合双向长短期记忆(bidirectional long short-term memory,BiLSTM)网络,提出电工钢片SAMCNN-BiLSTM磁致伸缩模型。首先,利用灰狼优化算法(grey wolf optimization,GWO)寻优神经网络结构的参数,实现复杂工况下磁致伸缩效应的准确表征;然后,建立中低频范围单频与叠加谐波激励等复杂工况下的磁致伸缩应变数据库,开展数据预处理与特征分析;最后,对SAMCNN-BiLSTM模型开展对比验证。对比叠加3次谐波激励下的磁致伸缩应变频谱主要分量,SAMCNN-BiLSTM模型计算值最大相对误差为3.70%,其比Jiles-Atherton-Sablik(J-A-S)、二次畴转等模型能更精确地表征电工钢片的磁致伸缩效应。
基金supported by the Science and Technology on Reactor System Design Technology Laboratory(No.LRSDT12023108)supported in part by the Chongqing Postdoctoral Science Foundation(No.cstc2021jcyj-bsh0252)+2 种基金the National Natural Science Foundation of China(No.12005030)Sichuan Province to unveil the list of marshal industry common technology research projects(No.23jBGOV0001)Special Program for Stabilizing Support to Basic Research of National Basic Research Institutes(No.WDZC-2023-05-03-05).
摘要The neutron diffusion equation plays a pivotal role in nuclear reactor analysis.Nevertheless,employing the physics-informed neural network(PINN)method for its solution entails certain limitations.Conventional PINN approaches generally utilize a fully connected network(FCN)architecture that is susceptible to overfitting,training instability,and gradient vanishing as the network depth increases.These challenges result in accuracy bottlenecks in the solution.In response to these issues,the residual-based resample physics-informed neural network(R2-PINN)is proposed.It is an improved PINN architecture that replaces the FCN with a convolutional neural network with a shortcut(S-CNN).It incorporates skip connections to facilitate gradient propagation between network layers.Additionally,the incorporation of the residual adaptive resampling(RAR)mechanism dynamically increases the number of sampling points.This,in turn,enhances the spatial representation capabilities and overall predictive accuracy of the model.The experimental results illustrate that our approach significantly improves the convergence capability of the model and achieves high-precision predictions of the physical fields.Compared with conventional FCN-based PINN methods,R 2-PINN effectively overcomes the limitations inherent in current methods.Thus,it provides more accurate and robust solutions for neutron diffusion equations.
基金supported by the National Natural Science Foundation of China(No.61905178)。
摘要Recently the depth estimation methods based on deep learning(DL)retain challenging to estimate a high-precision depth map in fringe projection structured light three-dimensional(3D)measurement with limited information from a single-frame fringe pattern.In this letter,we proposed a FDSUNet++convolutional neural network(CNN),which consists of a UNet++base model,an improved squeeze-andexcitation(ISE)block,a Fourier transform(FT)data preprocessing block,and a discrete wavelet transform(DWT)block.The proposed ISE block can improve the ability of feature extraction and the designed FT data preprocessing block preserves the key features of the fringe pattern by FT.The introduced DWT block reduces the complexity and training cost of the model.By integrating these three blocks into the UNet++,it can better achieve depth estimation.Experimental results from two structured light datasets demonstrate that the proposed FDSUNet++outperforms the state-of-the-art networks,achieving the best performance in both qualitative and quantitative evaluation.
摘要Skin cancer is the most prevalent cancer globally,primarily due to extensive exposure to Ultraviolet(UV)radiation.Early identification of skin cancer enhances the likelihood of effective treatment,as delays may lead to severe tumor advancement.This study proposes a novel hybrid deep learning strategy to address the complex issue of skin cancer diagnosis,with an architecture that integrates a Vision Transformer,a bespoke convolutional neural network(CNN),and an Xception module.They were evaluated using two benchmark datasets,HAM10000 and Skin Cancer ISIC.On the HAM10000,the model achieves a precision of 95.46%,an accuracy of 96.74%,a recall of 96.27%,specificity of 96.00%and an F1-Score of 95.86%.It obtains an accuracy of 93.19%,a precision of 93.25%,a recall of 92.80%,a specificity of 92.89%and an F1-Score of 93.19%on the Skin Cancer ISIC dataset.The findings demonstrate that the model that was proposed is robust and trustworthy when it comes to the classification of skin lesions.In addition,the utilization of Explainable AI techniques,such as Grad-CAM visualizations,assists in highlighting the most significant lesion areas that have an impact on the decisions that are made by the model.