目的利用深度学习技术,建立临床常见的侵袭性真菌图像辅助分类模型。方法回顾性收集2020年9月—2021年4月解放军总医院第八医学中心曲霉菌属、酵母菌属和新型隐球菌属真菌感染者的显微镜图像,按7∶1.5∶1.5的比例随机分为训练集、验证...目的利用深度学习技术,建立临床常见的侵袭性真菌图像辅助分类模型。方法回顾性收集2020年9月—2021年4月解放军总医院第八医学中心曲霉菌属、酵母菌属和新型隐球菌属真菌感染者的显微镜图像,按7∶1.5∶1.5的比例随机分为训练集、验证集和测试集。使用训练集和验证集图像对改进的MobileNetV2网络结构进行训练和参数调试,构建基于多尺度注意力机制的卷积神经网络(convolutional neural network,CNN)真菌图像11分类模型。以机器鉴定结果为金标准,以查准率、召回率和F1值为指标评价该模型对测试集真菌图像的分类效果。将该模型与5种经典CNN模型进行比较,评价指标包括模型参数量、内存占用量、网络每秒处理的图像数量(frames per second,FPS)、准确率及受试者操作特征曲线下面积(area under the curve,AUC)。结果共纳入真菌显微镜图像7666张,分别包括曲霉菌属、酵母菌属和新型隐球菌属图像2781张、4115张、770张。其中训练集5366张、验证集1150张、测试集1150张。改进的MobileNetV2模型对测试集11种真菌图像具有较高的分类性能,查准率为96.36%~100%,召回率为96.53%~100%,F1值为97.01%~100%。该模型的参数量、内存占用量分别为4.22 M、356.89 M,FPS为573,准确率为(99.09±0.18)%,AUC为0.9944±0.0018,综合性能优于5种经典网络模型。结论本研究提出的真菌图像分类模型,在保持低运算成本的情况下,可获得较高的真菌图像识别能力,其整体性能优于常见的经典模型。展开更多
A fiber Bragg grating (FBG) geophone and a surface seismic wave-based algorithm for detecting the direction of arrival (DOA) are described. The operational principle of FBG geophone is introduced and illustrated with ...A fiber Bragg grating (FBG) geophone and a surface seismic wave-based algorithm for detecting the direction of arrival (DOA) are described. The operational principle of FBG geophone is introduced and illustrated with systematic experimental data, demonstrating an improved FBG geophone with many advantages over the conventional geophones. An innovative, robust, and simple algorithm is developed for obtaining the bearing information on the seismic events, such as people walking, or vehicles moving. Such DOA estimate is based on the interactions and projections of surface-propagating seismic waves generated by the moving personnel or vehicles with a single tri-axial seismic sensor based on FBGs. Of particular interest is the case when the distance between the source of the seismic wave and the detector is less than or comparable to one wavelength (less than 100 m), corresponding to near-field detection, where an effective method of DOA finding lacks.展开更多
文摘目的利用深度学习技术,建立临床常见的侵袭性真菌图像辅助分类模型。方法回顾性收集2020年9月—2021年4月解放军总医院第八医学中心曲霉菌属、酵母菌属和新型隐球菌属真菌感染者的显微镜图像,按7∶1.5∶1.5的比例随机分为训练集、验证集和测试集。使用训练集和验证集图像对改进的MobileNetV2网络结构进行训练和参数调试,构建基于多尺度注意力机制的卷积神经网络(convolutional neural network,CNN)真菌图像11分类模型。以机器鉴定结果为金标准,以查准率、召回率和F1值为指标评价该模型对测试集真菌图像的分类效果。将该模型与5种经典CNN模型进行比较,评价指标包括模型参数量、内存占用量、网络每秒处理的图像数量(frames per second,FPS)、准确率及受试者操作特征曲线下面积(area under the curve,AUC)。结果共纳入真菌显微镜图像7666张,分别包括曲霉菌属、酵母菌属和新型隐球菌属图像2781张、4115张、770张。其中训练集5366张、验证集1150张、测试集1150张。改进的MobileNetV2模型对测试集11种真菌图像具有较高的分类性能,查准率为96.36%~100%,召回率为96.53%~100%,F1值为97.01%~100%。该模型的参数量、内存占用量分别为4.22 M、356.89 M,FPS为573,准确率为(99.09±0.18)%,AUC为0.9944±0.0018,综合性能优于5种经典网络模型。结论本研究提出的真菌图像分类模型,在保持低运算成本的情况下,可获得较高的真菌图像识别能力,其整体性能优于常见的经典模型。
基金This project was funded in part bythe U . S . Army
文摘A fiber Bragg grating (FBG) geophone and a surface seismic wave-based algorithm for detecting the direction of arrival (DOA) are described. The operational principle of FBG geophone is introduced and illustrated with systematic experimental data, demonstrating an improved FBG geophone with many advantages over the conventional geophones. An innovative, robust, and simple algorithm is developed for obtaining the bearing information on the seismic events, such as people walking, or vehicles moving. Such DOA estimate is based on the interactions and projections of surface-propagating seismic waves generated by the moving personnel or vehicles with a single tri-axial seismic sensor based on FBGs. Of particular interest is the case when the distance between the source of the seismic wave and the detector is less than or comparable to one wavelength (less than 100 m), corresponding to near-field detection, where an effective method of DOA finding lacks.