Artificial intelligence(AI)[1,2]allows computers to think and behave like humans,so it is now becoming more and more influential in almost every field[3].Hence,users in businesses,industries,hospitals[4],etc.,need to ...Artificial intelligence(AI)[1,2]allows computers to think and behave like humans,so it is now becoming more and more influential in almost every field[3].Hence,users in businesses,industries,hospitals[4],etc.,need to understand how these AI models work[5]and the potential impact of using them.展开更多
The cavitation in axial piston pumps threatens the reliability and safety of the overall hydraulic system.Vibration signal can reflect the cavitation conditions in axial piston pumps and it has been combined with mach...The cavitation in axial piston pumps threatens the reliability and safety of the overall hydraulic system.Vibration signal can reflect the cavitation conditions in axial piston pumps and it has been combined with machine learning to detect the pump cavitation.However,the vibration signal usually contains noise in real working conditions,which raises concerns about accurate recognition of cavitation in noisy environment.This paper presents an intelligent method to recognise the cavitation in axial piston pumps in noisy environment.First,we train a convolutional neural network(CNN)using the spectrogram images transformed from raw vibration data under different cavitation conditions.Second,we employ the technique of gradient-weighted class activation mapping(Grad-CAM)to visualise class-discriminative regions in the spectrogram image.Finally,we propose a novel image processing method based on Grad-CAM heatmap to automatically remove entrained noise and enhance class features in the spectrogram image.The experimental results show that the proposed method greatly improves the diagnostic performance of the CNN model in noisy environments.The classification accuracy of cavitation conditions increases from 0.50 to 0.89 and from 0.80 to 0.92 at signal-to-noise ratios of 4 and 6 dB,respectively.展开更多
Corona Virus(COVID-19)is a novel virus that crossed an animal-human barrier and emerged in Wuhan,China.Until now it has affected more than 119 million people.Detection of COVID-19 is a critical task and due to a large...Corona Virus(COVID-19)is a novel virus that crossed an animal-human barrier and emerged in Wuhan,China.Until now it has affected more than 119 million people.Detection of COVID-19 is a critical task and due to a large number of patients,a shortage of doctors has occurred for its detection.In this paper,a model has been suggested that not only detects the COVID-19 using X-ray and CT-Scan images but also shows the affected areas.Three classes have been defined;COVID-19,normal,and Pneumonia for X-ray images.For CT-Scan images,2 classes have been defined COVID-19 and non-COVID-19.For classi-fication purposes,pretrained models like ResNet50,VGG-16,and VGG19 have been used with some tuning.For detecting the affected areas Gradient-weighted Class Activation Mapping(GradCam)has been used.As the X-rays and ct images are taken at different intensities,so the contrast limited adaptive histogram equalization(CLAHE)has been applied to see the effect on the training of the models.As a result of these experiments,we achieved a maximum validation accuracy of 88.10%with a training accuracy of 88.48%for CT-Scan images using the ResNet50 model.While for X-ray images we achieved a maximum validation accuracy of 97.31%with a training accuracy of 95.64%using the VGG16 model.展开更多
鉴于Single Shot Multibox Detector(SSD)算法对中小目标检测时会出现漏检甚至错检的情况,提出一种改进的SSD目标检测算法,以提高中小目标检测的准确性.运用Gradient-weighted Class Activation Mapping(Grad-CAM)技术对检测过程中的细...鉴于Single Shot Multibox Detector(SSD)算法对中小目标检测时会出现漏检甚至错检的情况,提出一种改进的SSD目标检测算法,以提高中小目标检测的准确性.运用Gradient-weighted Class Activation Mapping(Grad-CAM)技术对检测过程中的细节作可视化处理,并以类激活图的形式呈现各检测层细节,分析各检测层的类激活图发现SSD算法中待检测目标的错检以及中小目标的漏检现象与回归损失函数相关.据此,采用Kullback-Leibler(KL)边框回归损失策略,利用Non Maximum Suppression(NMS)算法输出最终预测框.实验结果表明,改进算法相较于已有检测算法具有更高的准确率以及稳定性.展开更多
采用2016—2020年福建台网所记录的爆破和天然地震事件以及背景噪声数据集,使用CNN模型、Inception10模型、ResNet18模型和Vgg16模型4种深度学习网络模型进行分类研究。针对深度学习网络模型的“黑盒”问题,将梯度类激活映射(Gradient-w...采用2016—2020年福建台网所记录的爆破和天然地震事件以及背景噪声数据集,使用CNN模型、Inception10模型、ResNet18模型和Vgg16模型4种深度学习网络模型进行分类研究。针对深度学习网络模型的“黑盒”问题,将梯度类激活映射(Gradient-weighted Class Activation Mapping,Grad-CAM)算法引入这4种分类模型中,得到每个模型的可视化图。通过可视化图可以直观地看出模型在做出分类决策时对于不同波形特征的依赖权重,为模型的可解释性提供依据,进而提高模型的可信度。通过对模型的可视化图分析得出,分类效果更好的CNN模型和Vgg16模型在做出决策时更依赖于地震波形的震相特征,对于震前和震后的波段关注较小;而ResNet18模型和Inception10模型对于震相特征的关注不够敏锐。通过Grad-CAM算法对模型进行可视化分析得到的结果能够很好地反映模型的分类效果,对于改进和选择合适的分类模型具有重要意义。展开更多
基金Sino-UK Education Fund(OP202006)Royal Society(RP202G0230)+8 种基金MRC(MC_PC_17171)BHF(AA/18/3/34220)Hope Foundation for Cancer Research(RM60G0680)GCRF(P202PF11)BBSRC(RM32G0178B8)Sino-UK Industrial Fund(RP202G0289)Data Science Enhancement Fund(P202RE237)LIAS(P202ED10&P202RE969)Fight for Sight(24NN201).
文摘Artificial intelligence(AI)[1,2]allows computers to think and behave like humans,so it is now becoming more and more influential in almost every field[3].Hence,users in businesses,industries,hospitals[4],etc.,need to understand how these AI models work[5]and the potential impact of using them.
基金National Key R&D Program of China,Grant/Award Number:2018YFB1702503Open Foundation of the State Key Laboratory of Fluid Power and Mechatronic Systems,Grant/Award Number:GZKF-202108+2 种基金Open Foundation of the Guangdong Provincial Key Laboratory of Electronic Information Products Reliability TechnologyChina National Postdoctoral Program for Innovative Talents,Grant/Award Number:BX20200210China Postdoctoral Science Foundation,Grant/Award Number:2019M660086。
文摘The cavitation in axial piston pumps threatens the reliability and safety of the overall hydraulic system.Vibration signal can reflect the cavitation conditions in axial piston pumps and it has been combined with machine learning to detect the pump cavitation.However,the vibration signal usually contains noise in real working conditions,which raises concerns about accurate recognition of cavitation in noisy environment.This paper presents an intelligent method to recognise the cavitation in axial piston pumps in noisy environment.First,we train a convolutional neural network(CNN)using the spectrogram images transformed from raw vibration data under different cavitation conditions.Second,we employ the technique of gradient-weighted class activation mapping(Grad-CAM)to visualise class-discriminative regions in the spectrogram image.Finally,we propose a novel image processing method based on Grad-CAM heatmap to automatically remove entrained noise and enhance class features in the spectrogram image.The experimental results show that the proposed method greatly improves the diagnostic performance of the CNN model in noisy environments.The classification accuracy of cavitation conditions increases from 0.50 to 0.89 and from 0.80 to 0.92 at signal-to-noise ratios of 4 and 6 dB,respectively.
文摘Corona Virus(COVID-19)is a novel virus that crossed an animal-human barrier and emerged in Wuhan,China.Until now it has affected more than 119 million people.Detection of COVID-19 is a critical task and due to a large number of patients,a shortage of doctors has occurred for its detection.In this paper,a model has been suggested that not only detects the COVID-19 using X-ray and CT-Scan images but also shows the affected areas.Three classes have been defined;COVID-19,normal,and Pneumonia for X-ray images.For CT-Scan images,2 classes have been defined COVID-19 and non-COVID-19.For classi-fication purposes,pretrained models like ResNet50,VGG-16,and VGG19 have been used with some tuning.For detecting the affected areas Gradient-weighted Class Activation Mapping(GradCam)has been used.As the X-rays and ct images are taken at different intensities,so the contrast limited adaptive histogram equalization(CLAHE)has been applied to see the effect on the training of the models.As a result of these experiments,we achieved a maximum validation accuracy of 88.10%with a training accuracy of 88.48%for CT-Scan images using the ResNet50 model.While for X-ray images we achieved a maximum validation accuracy of 97.31%with a training accuracy of 95.64%using the VGG16 model.
文摘鉴于Single Shot Multibox Detector(SSD)算法对中小目标检测时会出现漏检甚至错检的情况,提出一种改进的SSD目标检测算法,以提高中小目标检测的准确性.运用Gradient-weighted Class Activation Mapping(Grad-CAM)技术对检测过程中的细节作可视化处理,并以类激活图的形式呈现各检测层细节,分析各检测层的类激活图发现SSD算法中待检测目标的错检以及中小目标的漏检现象与回归损失函数相关.据此,采用Kullback-Leibler(KL)边框回归损失策略,利用Non Maximum Suppression(NMS)算法输出最终预测框.实验结果表明,改进算法相较于已有检测算法具有更高的准确率以及稳定性.
文摘采用2016—2020年福建台网所记录的爆破和天然地震事件以及背景噪声数据集,使用CNN模型、Inception10模型、ResNet18模型和Vgg16模型4种深度学习网络模型进行分类研究。针对深度学习网络模型的“黑盒”问题,将梯度类激活映射(Gradient-weighted Class Activation Mapping,Grad-CAM)算法引入这4种分类模型中,得到每个模型的可视化图。通过可视化图可以直观地看出模型在做出分类决策时对于不同波形特征的依赖权重,为模型的可解释性提供依据,进而提高模型的可信度。通过对模型的可视化图分析得出,分类效果更好的CNN模型和Vgg16模型在做出决策时更依赖于地震波形的震相特征,对于震前和震后的波段关注较小;而ResNet18模型和Inception10模型对于震相特征的关注不够敏锐。通过Grad-CAM算法对模型进行可视化分析得到的结果能够很好地反映模型的分类效果,对于改进和选择合适的分类模型具有重要意义。
