针对无人机航拍图像目标检测中视野变化大、时空信息复杂等问题,文中基于YOLOv5(You Only Look Once Version5)架构,提出基于图像低维特征融合的航拍小目标检测模型.引入CA(Coordinate Attention),改进MobileNetV3的反转残差块,增加图...针对无人机航拍图像目标检测中视野变化大、时空信息复杂等问题,文中基于YOLOv5(You Only Look Once Version5)架构,提出基于图像低维特征融合的航拍小目标检测模型.引入CA(Coordinate Attention),改进MobileNetV3的反转残差块,增加图像空间维度信息的同时降低模型参数量.改进YOLOv5特征金字塔网络结构,融合浅层网络中的特征图,增加模型对图像低维有效信息的表达能力,进而提升小目标检测精度.同时为了降低航拍图像中复杂背景带来的干扰,引入无参平均注意力模块,同时关注图像的空间注意力与通道注意力;引入VariFocal Loss,降低负样本在训练过程中的权重占比.在VisDrone数据集上的实验验证文中模型的有效性,该模型在有效提升检测精度的同时明显降低复杂度.展开更多
目的为精神分裂症患者攻击行为评估提供一种以攻击行为发生次数作为评分依据的纵向评估工具。方法对一生攻击史量表进行翻译并修订,形成一生攻击史量表中文版(Life History of Aggression-Chinese Version,LHA-CV)。基于成都市社区及强...目的为精神分裂症患者攻击行为评估提供一种以攻击行为发生次数作为评分依据的纵向评估工具。方法对一生攻击史量表进行翻译并修订,形成一生攻击史量表中文版(Life History of Aggression-Chinese Version,LHA-CV)。基于成都市社区及强制医疗所共369例精神分裂症患者,采用分半信度、重测信度及评估者间一致性对LHA-CV的信度进行分析,采用项目分析、结构效度及效标效度进行效度分析。结果项目分析发现,LHA-CV整体的同质性与区分度较好。探索性因素分析发现,Kaiser-Meyer-Olkin(KMO)检验值为0.80,Bartlett球形检验χ2=1203.46(P<0.05),析出非躯体攻击、躯体攻击、自我攻击和反社会行为/结果4个因子,11个条目的因子载荷均大于0.40。对该因子模型进行验证性因素分析,卡方自由度比(χ2/df)为3.61,近似误差均方根为0.07,拟合优度指数为0.92,比较拟合指数为0.90,增值拟合指数为0.90,各因子区分效度较好。效标效度检验结果显示,LHA-CV总分与MacArthur社区暴力工具评估的攻击行为等级、Buss-Perry攻击性量表总分、人格障碍诊断问卷(第4版)反社会型人格障碍分量表得分呈正相关(P<0.05)。LHA-CV的非躯体攻击、躯体攻击、自我攻击、反社会行为/结果以及总分的Cronbach’sα系数分别为0.82、0.73、0.74、0.56和0.79。LHA-CV总分的重测信度为0.82(P<0.05),Spearman-Brown分半信度为0.66,组内相关系数为0.99。结论LHA-CV具有良好的信度与效度,可作为纵向测评精神分裂症患者攻击行为的评估工具。展开更多
Background: Non-uniformity in signal intensity occurs commonly in magnetic resonance (MR) imaging, which may pose substantial problems when using a 3T scanner. Therefore, image non-uniformity correction is usually app...Background: Non-uniformity in signal intensity occurs commonly in magnetic resonance (MR) imaging, which may pose substantial problems when using a 3T scanner. Therefore, image non-uniformity correction is usually applied. Purpose: To compare the correction effects of the phased-array uniformity enhancement (PURE), a calibration-based image non-uniformity correction method, among three different software versions in 3T Gd-EOB-DTPA-enhanced MR imaging. Material and Methods: Hepatobiliary-phase images of a total of 120 patients who underwent Gd-EOB-DTPA-enhanced MR imaging on the same 3T scanner were analyzed retrospectively. Forty patients each were examined using three software versions (DV25, DV25.1, and DV26). The effects of PURE were compared by visual assessment, histogram analysis of liver signal intensity, evaluation of the spatial distribution of correction effects, and evaluation of quantitative indices of liver parenchymal enhancement. Results: The visual assessment indicated the highest uniformity of PURE-corrected images for DV26, followed by DV25 and DV25.1. Histogram analysis of corrected images demonstrated significantly larger variations in liver signal for DV25.1 than for the other two versions. Although PURE caused a relative increase in pixel values for central and lateral regions, such effects were weaker for DV25.1 than for the other two versions. In the evaluation of quantitative indices of liver parenchymal enhancement, the liver-to-muscle ratio (LMR) was significantly higher for the corrected images than for the uncorrected images, but the liver-to-spleen ratio (LSR) showed no significant differences. For corrected images, the LMR was significantly higher for DV25 and DV26 than for DV25.1, but the LSR showed no significant differences among the three versions. Conclusion: There were differences in the effects of PURE among the three software versions in 3T Gd-EOB-DTPA-enhanced MR imaging. Even if the non-uniformity correction method has the same brand name, correction effects may differ depending on the software version, and these differences may affect visual and quantitative evaluations.展开更多
文摘针对无人机航拍图像目标检测中视野变化大、时空信息复杂等问题,文中基于YOLOv5(You Only Look Once Version5)架构,提出基于图像低维特征融合的航拍小目标检测模型.引入CA(Coordinate Attention),改进MobileNetV3的反转残差块,增加图像空间维度信息的同时降低模型参数量.改进YOLOv5特征金字塔网络结构,融合浅层网络中的特征图,增加模型对图像低维有效信息的表达能力,进而提升小目标检测精度.同时为了降低航拍图像中复杂背景带来的干扰,引入无参平均注意力模块,同时关注图像的空间注意力与通道注意力;引入VariFocal Loss,降低负样本在训练过程中的权重占比.在VisDrone数据集上的实验验证文中模型的有效性,该模型在有效提升检测精度的同时明显降低复杂度.
文摘目的为精神分裂症患者攻击行为评估提供一种以攻击行为发生次数作为评分依据的纵向评估工具。方法对一生攻击史量表进行翻译并修订,形成一生攻击史量表中文版(Life History of Aggression-Chinese Version,LHA-CV)。基于成都市社区及强制医疗所共369例精神分裂症患者,采用分半信度、重测信度及评估者间一致性对LHA-CV的信度进行分析,采用项目分析、结构效度及效标效度进行效度分析。结果项目分析发现,LHA-CV整体的同质性与区分度较好。探索性因素分析发现,Kaiser-Meyer-Olkin(KMO)检验值为0.80,Bartlett球形检验χ2=1203.46(P<0.05),析出非躯体攻击、躯体攻击、自我攻击和反社会行为/结果4个因子,11个条目的因子载荷均大于0.40。对该因子模型进行验证性因素分析,卡方自由度比(χ2/df)为3.61,近似误差均方根为0.07,拟合优度指数为0.92,比较拟合指数为0.90,增值拟合指数为0.90,各因子区分效度较好。效标效度检验结果显示,LHA-CV总分与MacArthur社区暴力工具评估的攻击行为等级、Buss-Perry攻击性量表总分、人格障碍诊断问卷(第4版)反社会型人格障碍分量表得分呈正相关(P<0.05)。LHA-CV的非躯体攻击、躯体攻击、自我攻击、反社会行为/结果以及总分的Cronbach’sα系数分别为0.82、0.73、0.74、0.56和0.79。LHA-CV总分的重测信度为0.82(P<0.05),Spearman-Brown分半信度为0.66,组内相关系数为0.99。结论LHA-CV具有良好的信度与效度,可作为纵向测评精神分裂症患者攻击行为的评估工具。
文摘Background: Non-uniformity in signal intensity occurs commonly in magnetic resonance (MR) imaging, which may pose substantial problems when using a 3T scanner. Therefore, image non-uniformity correction is usually applied. Purpose: To compare the correction effects of the phased-array uniformity enhancement (PURE), a calibration-based image non-uniformity correction method, among three different software versions in 3T Gd-EOB-DTPA-enhanced MR imaging. Material and Methods: Hepatobiliary-phase images of a total of 120 patients who underwent Gd-EOB-DTPA-enhanced MR imaging on the same 3T scanner were analyzed retrospectively. Forty patients each were examined using three software versions (DV25, DV25.1, and DV26). The effects of PURE were compared by visual assessment, histogram analysis of liver signal intensity, evaluation of the spatial distribution of correction effects, and evaluation of quantitative indices of liver parenchymal enhancement. Results: The visual assessment indicated the highest uniformity of PURE-corrected images for DV26, followed by DV25 and DV25.1. Histogram analysis of corrected images demonstrated significantly larger variations in liver signal for DV25.1 than for the other two versions. Although PURE caused a relative increase in pixel values for central and lateral regions, such effects were weaker for DV25.1 than for the other two versions. In the evaluation of quantitative indices of liver parenchymal enhancement, the liver-to-muscle ratio (LMR) was significantly higher for the corrected images than for the uncorrected images, but the liver-to-spleen ratio (LSR) showed no significant differences. For corrected images, the LMR was significantly higher for DV25 and DV26 than for DV25.1, but the LSR showed no significant differences among the three versions. Conclusion: There were differences in the effects of PURE among the three software versions in 3T Gd-EOB-DTPA-enhanced MR imaging. Even if the non-uniformity correction method has the same brand name, correction effects may differ depending on the software version, and these differences may affect visual and quantitative evaluations.
