目的散焦模糊检测致力于区分图像中的清晰与模糊像素,广泛应用于诸多领域,是计算机视觉中的重要研究方向。待检测图像含复杂场景时,现有的散焦模糊检测方法存在精度不够高、检测结果边界不完整等问题。本文提出一种由粗到精的多尺度散...目的散焦模糊检测致力于区分图像中的清晰与模糊像素,广泛应用于诸多领域,是计算机视觉中的重要研究方向。待检测图像含复杂场景时,现有的散焦模糊检测方法存在精度不够高、检测结果边界不完整等问题。本文提出一种由粗到精的多尺度散焦模糊检测网络,通过融合不同尺度下图像的多层卷积特征提高散焦模糊的检测精度。方法将图像缩放至不同尺度,使用卷积神经网络从每个尺度下的图像中提取多层卷积特征,并使用卷积层融合不同尺度图像对应层的特征;使用卷积长短时记忆(convolutional long-short term memory,Conv-LSTM)层自顶向下地整合不同尺度的模糊特征,同时生成对应尺度的模糊检测图,以这种方式将深层的语义信息逐步传递至浅层网络;在此过程中,将深浅层特征联合,利用浅层特征细化深一层的模糊检测结果;使用卷积层将多尺度检测结果融合得到最终结果。本文在网络训练过程中使用了多层监督策略确保每个Conv-LSTM层都能达到最优。结果在DUT(Dalian University of Technology)和CUHK(The Chinese University of Hong Kong)两个公共的模糊检测数据集上进行训练和测试,对比了包括当前最好的模糊检测算法BTBCRL(bottom-top-bottom network with cascaded defocus blur detection map residual learning),De Fusion Net(defocus blur detection network via recurrently fusing and refining multi-scale deep features)和DHDE(multi-scale deep and hand-crafted features for defocus estimation)等10种算法。实验结果表明:在DUT数据集上,本文模型相比于De Fusion Net模型,MAE(mean absolute error)值降低了38.8%,F0.3值提高了5.4%;在CUHK数据集上,相比于LBP(local binary pattern)算法,MAE值降低了36.7%,F0.3值提高了9.7%。通过实验对比,充分验证了本文提出的散焦模糊检测模型的有效性。结论本文提出的由粗到精的多尺度散焦模糊检测方法,通过融合不同尺度图像的特征,以及使用卷积长短时记忆层自顶向下地整合深层的语义信息和浅层的细节信息,使得模型在不同的图像场景中能得到更加准确的散焦模糊检测结果。展开更多
OBJECTIVE: To analyze the component law of Chinese medicines in fuming-washing therapy for knee osteoarthritis(KOA), and develop new fuming-washing prescriptions for KOA through unsupervised data mining methods.METHOD...OBJECTIVE: To analyze the component law of Chinese medicines in fuming-washing therapy for knee osteoarthritis(KOA), and develop new fuming-washing prescriptions for KOA through unsupervised data mining methods.METHODS: Chinese medicine recipes for fuming-washing therapy for KOA were collected and recorded in a database. The correlation coefficient among herbs, core combinations of herbs, andnew prescriptions were analyzed using modified mutual information, complex system entropy cluster, and unsupervised hierarchical clustering, respectively.RESULTS: Based on analysis of 345 Chinese medicine recipes for fuming-washing therapy, 68 herbs occurred frequently, 33 herb pairs occurred frequently, and 12 core combinations were found.Five new fuming-washing recipes for KOA were developed.CONCLUSION: Chinese medicines for fuming-washing therapy of KOA mainly consist of wind-dampness-dispelling and cold-dispersing herbs, blood-activating and stasis-resolving herbs,and wind-dampness-dispelling and heat-clearing herbs. The treatment of fuming-washing therapy for KOA also includes dispelling wind-dampness and dispersing cold, activating blood and resolving stasis, and dispelling wind-dampness and clearing heat. Zhenzhutougucao(Herba Speranskiae Tuberculatae), Honghua(Flos Carthami), Niuxi(Radix Achyranthis Bidentatae), Shenjincao(Herba Lycopodii Japonici), Weilingxian(Radix et Rhizoma Clematidis Chinensis), Chuanwu(Radix Aconiti), Haitongpi(Cortex Erythrinae Variegatae), Ruxiang(Olibanum),Danggui(Radix Angelicae Sinensis), Caowu(Radix Aconiti Kusnezoffii), Moyao(Myrrha), and Aiye(Folium Artemisiae Argyi) are the main herbs used in the fuming-washing treatment for KOA.展开更多
文摘目的散焦模糊检测致力于区分图像中的清晰与模糊像素,广泛应用于诸多领域,是计算机视觉中的重要研究方向。待检测图像含复杂场景时,现有的散焦模糊检测方法存在精度不够高、检测结果边界不完整等问题。本文提出一种由粗到精的多尺度散焦模糊检测网络,通过融合不同尺度下图像的多层卷积特征提高散焦模糊的检测精度。方法将图像缩放至不同尺度,使用卷积神经网络从每个尺度下的图像中提取多层卷积特征,并使用卷积层融合不同尺度图像对应层的特征;使用卷积长短时记忆(convolutional long-short term memory,Conv-LSTM)层自顶向下地整合不同尺度的模糊特征,同时生成对应尺度的模糊检测图,以这种方式将深层的语义信息逐步传递至浅层网络;在此过程中,将深浅层特征联合,利用浅层特征细化深一层的模糊检测结果;使用卷积层将多尺度检测结果融合得到最终结果。本文在网络训练过程中使用了多层监督策略确保每个Conv-LSTM层都能达到最优。结果在DUT(Dalian University of Technology)和CUHK(The Chinese University of Hong Kong)两个公共的模糊检测数据集上进行训练和测试,对比了包括当前最好的模糊检测算法BTBCRL(bottom-top-bottom network with cascaded defocus blur detection map residual learning),De Fusion Net(defocus blur detection network via recurrently fusing and refining multi-scale deep features)和DHDE(multi-scale deep and hand-crafted features for defocus estimation)等10种算法。实验结果表明:在DUT数据集上,本文模型相比于De Fusion Net模型,MAE(mean absolute error)值降低了38.8%,F0.3值提高了5.4%;在CUHK数据集上,相比于LBP(local binary pattern)算法,MAE值降低了36.7%,F0.3值提高了9.7%。通过实验对比,充分验证了本文提出的散焦模糊检测模型的有效性。结论本文提出的由粗到精的多尺度散焦模糊检测方法,通过融合不同尺度图像的特征,以及使用卷积长短时记忆层自顶向下地整合深层的语义信息和浅层的细节信息,使得模型在不同的图像场景中能得到更加准确的散焦模糊检测结果。
基金Supported by Grant from the Administration of Traditional Chinese Medicine of Guangdong Province in China(No.20131161)the Specialized Research Fund for the Doctoral Program of Higher Education of China(No.20124425110004)
文摘OBJECTIVE: To analyze the component law of Chinese medicines in fuming-washing therapy for knee osteoarthritis(KOA), and develop new fuming-washing prescriptions for KOA through unsupervised data mining methods.METHODS: Chinese medicine recipes for fuming-washing therapy for KOA were collected and recorded in a database. The correlation coefficient among herbs, core combinations of herbs, andnew prescriptions were analyzed using modified mutual information, complex system entropy cluster, and unsupervised hierarchical clustering, respectively.RESULTS: Based on analysis of 345 Chinese medicine recipes for fuming-washing therapy, 68 herbs occurred frequently, 33 herb pairs occurred frequently, and 12 core combinations were found.Five new fuming-washing recipes for KOA were developed.CONCLUSION: Chinese medicines for fuming-washing therapy of KOA mainly consist of wind-dampness-dispelling and cold-dispersing herbs, blood-activating and stasis-resolving herbs,and wind-dampness-dispelling and heat-clearing herbs. The treatment of fuming-washing therapy for KOA also includes dispelling wind-dampness and dispersing cold, activating blood and resolving stasis, and dispelling wind-dampness and clearing heat. Zhenzhutougucao(Herba Speranskiae Tuberculatae), Honghua(Flos Carthami), Niuxi(Radix Achyranthis Bidentatae), Shenjincao(Herba Lycopodii Japonici), Weilingxian(Radix et Rhizoma Clematidis Chinensis), Chuanwu(Radix Aconiti), Haitongpi(Cortex Erythrinae Variegatae), Ruxiang(Olibanum),Danggui(Radix Angelicae Sinensis), Caowu(Radix Aconiti Kusnezoffii), Moyao(Myrrha), and Aiye(Folium Artemisiae Argyi) are the main herbs used in the fuming-washing treatment for KOA.
