In this paper, an electrical resistance tomography(ERT) imaging method is used as a classifier, and then the Dempster-Shafer's evidence theory with fuzzy clustering is integrated to improve the ERT image quality. ...In this paper, an electrical resistance tomography(ERT) imaging method is used as a classifier, and then the Dempster-Shafer's evidence theory with fuzzy clustering is integrated to improve the ERT image quality. The fuzzy clustering is applied to determining the key mass function, and dealing with the uncertain, incomplete and inconsistent measured imaging data in ERT. The proposed method was applied to images with the same investigated object under eight typical current drive patterns. Experiments were performed on a group of simulations using COMSOL Multiphysics tool and measurements with a piece of porcine lung and a pair of porcine kidneys as test materials. Compared with any single drive pattern, the proposed method can provide images with a spatial resolution of about 10% higher, while the time resolution was almost the same.展开更多
针对疲劳驾驶检测问题,提出一种以softmax损失与中心损失相结合的深度卷积神经网络算法。首先,利用含有方向的梯度直方图(histogram of oriented gridients,HOG)和级联分类器(support vector machine,SVM)算法的Dlib库中预训练的人脸检...针对疲劳驾驶检测问题,提出一种以softmax损失与中心损失相结合的深度卷积神经网络算法。首先,利用含有方向的梯度直方图(histogram of oriented gridients,HOG)和级联分类器(support vector machine,SVM)算法的Dlib库中预训练的人脸检测器,来检测驾驶员的脸部区域。其次,使用级联回归(ensemble of regression trees,ERT)算法实现脸部68个关键点标定及眼睛和嘴巴的定位。最后,为了优化softmax损失在深度卷积网络分类中出现的类内间距大的问题,加入中心损失函数,提高类间差异性、类内紧密性以及驾驶员脸部疲劳状态识别准确率。在自建测试集和YawDD哈欠数据集中的实验结果显示,该方法能够准确地识别检测驾驶员疲劳表情,平均识别准确率达到98.81%。与传统的疲劳驾驶检测识别方法相比,该方法可以自动进行疲劳特征提取,并且训练准确率、检测识别率及鲁棒性得到提高;与未改进的深度卷积网络相比,检测识别的概率平均提高了约5.09%。展开更多
基金Supported by National Natural Science Foundation of China(No.61774014 and No.60772080)
文摘In this paper, an electrical resistance tomography(ERT) imaging method is used as a classifier, and then the Dempster-Shafer's evidence theory with fuzzy clustering is integrated to improve the ERT image quality. The fuzzy clustering is applied to determining the key mass function, and dealing with the uncertain, incomplete and inconsistent measured imaging data in ERT. The proposed method was applied to images with the same investigated object under eight typical current drive patterns. Experiments were performed on a group of simulations using COMSOL Multiphysics tool and measurements with a piece of porcine lung and a pair of porcine kidneys as test materials. Compared with any single drive pattern, the proposed method can provide images with a spatial resolution of about 10% higher, while the time resolution was almost the same.
文摘针对疲劳驾驶检测问题,提出一种以softmax损失与中心损失相结合的深度卷积神经网络算法。首先,利用含有方向的梯度直方图(histogram of oriented gridients,HOG)和级联分类器(support vector machine,SVM)算法的Dlib库中预训练的人脸检测器,来检测驾驶员的脸部区域。其次,使用级联回归(ensemble of regression trees,ERT)算法实现脸部68个关键点标定及眼睛和嘴巴的定位。最后,为了优化softmax损失在深度卷积网络分类中出现的类内间距大的问题,加入中心损失函数,提高类间差异性、类内紧密性以及驾驶员脸部疲劳状态识别准确率。在自建测试集和YawDD哈欠数据集中的实验结果显示,该方法能够准确地识别检测驾驶员疲劳表情,平均识别准确率达到98.81%。与传统的疲劳驾驶检测识别方法相比,该方法可以自动进行疲劳特征提取,并且训练准确率、检测识别率及鲁棒性得到提高;与未改进的深度卷积网络相比,检测识别的概率平均提高了约5.09%。
