目前在铝材表面缺陷检测领域,常用的检测模型存在检测精度不高、实时性不强和参数量大等问题。针对上述问题,对目标检测模型YOLOv8做了改进。首先,该模型使用自研的动态可变形卷积模块取代原有最后一层C2f模块;其次,将RT-DETR(real-time...目前在铝材表面缺陷检测领域,常用的检测模型存在检测精度不高、实时性不强和参数量大等问题。针对上述问题,对目标检测模型YOLOv8做了改进。首先,该模型使用自研的动态可变形卷积模块取代原有最后一层C2f模块;其次,将RT-DETR(real-time detection transformer)模型中的检测头移植到新模型中,利用Transformer中解码器的思想,省去了模型后处理的步骤;最后,将NWD(normalized Wasserstein distance)loss和WIoUv3(wise-intersection over union version 3)loss结合作为改进后模型的回归损失函数,使模型动态筛选锚框,解决IoU对于不同尺寸缺陷类型敏感度不同导致的标签分配不准确的问题。改进之后的模型相比于基线模型YOLOv8,平均精度提高了3.8%,每秒处理帧数提高至92 f/s,同时在钢材表面缺陷检测数据集和太阳能电池板表面缺陷检测数据集上也具有很强的鲁棒性,新模型在实时检测、实际部署中具有很大优势。展开更多
燃气机组在新型电力系统中将长期发挥重要的支撑和调节作用。为推动燃气机组碳排放的精准监测,在常规碳排放计量方法的基础上,提出了基于机器学习与在线监测的数字化监测方法,以及基于电厂厂级实时信息监控系统(supervisory information...燃气机组在新型电力系统中将长期发挥重要的支撑和调节作用。为推动燃气机组碳排放的精准监测,在常规碳排放计量方法的基础上,提出了基于机器学习与在线监测的数字化监测方法,以及基于电厂厂级实时信息监控系统(supervisory information system,SIS)进行在线数据与排放因子的核算方法。以某燃气机组为例,对比分析了统计期内通过4类计量方法(包括基于指南的人工核算、在线核算、基于冷干法和稀释法的在线监测以及数字化监测)所获得二氧化碳排放量的准确性。结论如下:碳排放在线核算与人工核算之间相对误差控制在1%以内,两者可互为参考,以实现厂内数据校验;稀释法和冷干法所采用的在线监测技术均能获取准确的碳排放量,但稀释法需定期校准以维持其测量精度;通过大量机组运行数据建立起来的数字化监测模型能够实现碳排放量的实时预测,其准确度较高。在缺失二氧化碳在线监测数据时,该模型仍可为机组运行提供指导,并有望推广至同类型机组。展开更多
针对目前热电池内部装配缺陷检测效率低、准确度不高的问题,研究了一种可精准分割内部电池堆图像并能够准确识别缺陷种类的方法。首先采用水平、垂直积分投影法对目标电池堆边缘特征进行提取,利用局部自适应对比度增强算法对局部不清晰...针对目前热电池内部装配缺陷检测效率低、准确度不高的问题,研究了一种可精准分割内部电池堆图像并能够准确识别缺陷种类的方法。首先采用水平、垂直积分投影法对目标电池堆边缘特征进行提取,利用局部自适应对比度增强算法对局部不清晰部分进行细节纹理增强;然后研究了缺陷结构的灰度特性,计算提取出缺陷特征参数;最后使用BP(back propagation)神经网络和CART(classification and regression tree)决策树对特征参数分类识别,并根据分类准确度进行权重分配,将加权融合后的结果作为检测的最终判据。实验结果表明:该方法对2000个样本的检测准确度达98.9%,为热电池的X射线缺陷检测提供了有效的途径。展开更多
目的探讨高血压合并腔隙性脑梗死患者临床患病特点及其相关危险因素。方法随机选取2012年2月-2015年2月就诊于新疆维吾尔自治区人民医院米东医院内一科的高血压病患者作为调查对象,测定所有研究对象胆固醇(total cholesterol,TC)、甘油...目的探讨高血压合并腔隙性脑梗死患者临床患病特点及其相关危险因素。方法随机选取2012年2月-2015年2月就诊于新疆维吾尔自治区人民医院米东医院内一科的高血压病患者作为调查对象,测定所有研究对象胆固醇(total cholesterol,TC)、甘油三酯(triglyceride,TG)、高密度脂蛋白胆固醇(high density lipoprotein-cholesterol,HDL-C)、低密度脂蛋白胆固醇(low density lipoprotein-cholesterol,LDL-C)和空腹血糖(fasting blood glucos,FBG)等生化指标。检测患者同型半胱氨酸(Hcy)水平,以血Hcy浓度≥15μmol/L为高同型半胱氨酸血症。对颈动脉内膜厚度(IMT)进行超声检测。结果高血压不合并腔隙性脑梗死与高血压合并腔隙性脑梗死比较,左、右颈总动脉内中膜的厚度、Hcy、吸烟、饮酒及糖尿病之间差异有统计学意义(P <0.05)。将高血压合并脑梗死的危险因素进行二分类logistic回归分析,在多因素分析中,高血压患者合并糖尿病(OR=4.988,95%CI=1.966～12.655)、吸烟(OR=3.629,95%CI=1.186～11.100)、高龄(OR=1.127,95%CI=1.073～1.183)、高TC(OR=0.453,95%CI=0.216～0.952)是腔隙性脑梗死的危险因素,差异有统计学意义(P<0.05)。结论合并有糖尿病、吸烟、年龄是高血压合并脑梗死的独立危险因素。在腔隙性脑梗死的防控过程中,应根据患者的具体情况,给予积极的个性化的干预措施。展开更多
In recent years, Lamb wave and piezoelectric transducers(PZTs) array based wavenumber filtering technique for damage estimation has been gradually studied. Compared with the time domain and frequency domain analysis o...In recent years, Lamb wave and piezoelectric transducers(PZTs) array based wavenumber filtering technique for damage estimation has been gradually studied. Compared with the time domain and frequency domain analysis of the Lamb wave signals, the wavenumber domain analysis is an effective approach to distinguish wave propagating direction and wave modes. However, the spatial resolution sampled by the PZTs is lower than that sampled by scanning laser Doppler vibrometer. As for the diameter of the PZT, it cannot be very small. In this paper, a new Lamb wave spatial sampling signal optimization method based on Morlet wavelet is proposed.Firstly, the frequency band parameter of the Morlet mother wavelet function is calculated by the Lamb wave excitation signal. Then, the sum of squared errors between the Lamb wave spatial sampling signal and the Morlet wavelet function fitting waveform at each scale factor and time factor is calculated. Finally, the scale factor and time factor corresponding to the least sum of squared errors can be judged to be the best match scale factor and time factor respectively, and the Morlet wavelet function fitting waveform in that scale factor and time factor can be seen as the optimized Lamb wave spatial sampling signal. The validation experiment performed on a glass fiber epoxy composite plate shows that the proposed method can improve the spatial resolution and length of the Lamb wave spatial sampling signal, and the sum of squared errors of this method is no more than 0.2.展开更多
Objective: to comprehensively improve the effect of nursing services for prevention of delirium in emergency critically ill patients. Methods: in order to effectively improve the clinical nursing effect of critically ...Objective: to comprehensively improve the effect of nursing services for prevention of delirium in emergency critically ill patients. Methods: in order to effectively improve the clinical nursing effect of critically ill patients with delirium, combined with the assessment and management process of delirium for critically ill patients treated in our hospital from January 2019 to January 2020, and selected the patients without delirium and delirium. The differences between the two groups were compared, the causes of delirium were analyzed and targeted nursing measures were taken, and the APACHE-II scores of the delirium group before and after nursing were compared. Results: there were significant statistical differences in age, respiratory failure, liver diseases, hypotensive shock, electrolyte disturbance and APACHE-II score between the delirium group and the delirium group: P < 0.05. Compared with before nursing, the APACHE-II score of patients in delirium group was significantly lower after nursing, P < 0.05. Conclusion: the incidence rate of delirium in emergency critically ill patients is relatively high. The early control of the risk factors of delirium patients and the provision of high-quality nursing services can prevent and reduce the incidence rate of delirium, promote the benign outcome of patients, and have positive significance for the treatment of patients.展开更多
文摘目前在铝材表面缺陷检测领域,常用的检测模型存在检测精度不高、实时性不强和参数量大等问题。针对上述问题,对目标检测模型YOLOv8做了改进。首先,该模型使用自研的动态可变形卷积模块取代原有最后一层C2f模块;其次,将RT-DETR(real-time detection transformer)模型中的检测头移植到新模型中,利用Transformer中解码器的思想,省去了模型后处理的步骤;最后,将NWD(normalized Wasserstein distance)loss和WIoUv3(wise-intersection over union version 3)loss结合作为改进后模型的回归损失函数,使模型动态筛选锚框,解决IoU对于不同尺寸缺陷类型敏感度不同导致的标签分配不准确的问题。改进之后的模型相比于基线模型YOLOv8,平均精度提高了3.8%,每秒处理帧数提高至92 f/s,同时在钢材表面缺陷检测数据集和太阳能电池板表面缺陷检测数据集上也具有很强的鲁棒性,新模型在实时检测、实际部署中具有很大优势。
文摘燃气机组在新型电力系统中将长期发挥重要的支撑和调节作用。为推动燃气机组碳排放的精准监测,在常规碳排放计量方法的基础上,提出了基于机器学习与在线监测的数字化监测方法,以及基于电厂厂级实时信息监控系统(supervisory information system,SIS)进行在线数据与排放因子的核算方法。以某燃气机组为例,对比分析了统计期内通过4类计量方法(包括基于指南的人工核算、在线核算、基于冷干法和稀释法的在线监测以及数字化监测)所获得二氧化碳排放量的准确性。结论如下:碳排放在线核算与人工核算之间相对误差控制在1%以内,两者可互为参考,以实现厂内数据校验;稀释法和冷干法所采用的在线监测技术均能获取准确的碳排放量,但稀释法需定期校准以维持其测量精度;通过大量机组运行数据建立起来的数字化监测模型能够实现碳排放量的实时预测,其准确度较高。在缺失二氧化碳在线监测数据时,该模型仍可为机组运行提供指导,并有望推广至同类型机组。
文摘针对目前热电池内部装配缺陷检测效率低、准确度不高的问题,研究了一种可精准分割内部电池堆图像并能够准确识别缺陷种类的方法。首先采用水平、垂直积分投影法对目标电池堆边缘特征进行提取,利用局部自适应对比度增强算法对局部不清晰部分进行细节纹理增强;然后研究了缺陷结构的灰度特性,计算提取出缺陷特征参数;最后使用BP(back propagation)神经网络和CART(classification and regression tree)决策树对特征参数分类识别,并根据分类准确度进行权重分配,将加权融合后的结果作为检测的最终判据。实验结果表明:该方法对2000个样本的检测准确度达98.9%,为热电池的X射线缺陷检测提供了有效的途径。
文摘目的探讨高血压合并腔隙性脑梗死患者临床患病特点及其相关危险因素。方法随机选取2012年2月-2015年2月就诊于新疆维吾尔自治区人民医院米东医院内一科的高血压病患者作为调查对象,测定所有研究对象胆固醇(total cholesterol,TC)、甘油三酯(triglyceride,TG)、高密度脂蛋白胆固醇(high density lipoprotein-cholesterol,HDL-C)、低密度脂蛋白胆固醇(low density lipoprotein-cholesterol,LDL-C)和空腹血糖(fasting blood glucos,FBG)等生化指标。检测患者同型半胱氨酸(Hcy)水平,以血Hcy浓度≥15μmol/L为高同型半胱氨酸血症。对颈动脉内膜厚度(IMT)进行超声检测。结果高血压不合并腔隙性脑梗死与高血压合并腔隙性脑梗死比较,左、右颈总动脉内中膜的厚度、Hcy、吸烟、饮酒及糖尿病之间差异有统计学意义(P <0.05)。将高血压合并脑梗死的危险因素进行二分类logistic回归分析,在多因素分析中,高血压患者合并糖尿病(OR=4.988,95%CI=1.966～12.655)、吸烟(OR=3.629,95%CI=1.186～11.100)、高龄(OR=1.127,95%CI=1.073～1.183)、高TC(OR=0.453,95%CI=0.216～0.952)是腔隙性脑梗死的危险因素,差异有统计学意义(P<0.05)。结论合并有糖尿病、吸烟、年龄是高血压合并脑梗死的独立危险因素。在腔隙性脑梗死的防控过程中,应根据患者的具体情况,给予积极的个性化的干预措施。
基金the National Natural Science Foundation of China(No.51705530)the Xuzhou Science and Technology Plan Project(No.KH17010)the Air Force Service Academy Youth Research Fund Project(No.KY2018D002A)
文摘In recent years, Lamb wave and piezoelectric transducers(PZTs) array based wavenumber filtering technique for damage estimation has been gradually studied. Compared with the time domain and frequency domain analysis of the Lamb wave signals, the wavenumber domain analysis is an effective approach to distinguish wave propagating direction and wave modes. However, the spatial resolution sampled by the PZTs is lower than that sampled by scanning laser Doppler vibrometer. As for the diameter of the PZT, it cannot be very small. In this paper, a new Lamb wave spatial sampling signal optimization method based on Morlet wavelet is proposed.Firstly, the frequency band parameter of the Morlet mother wavelet function is calculated by the Lamb wave excitation signal. Then, the sum of squared errors between the Lamb wave spatial sampling signal and the Morlet wavelet function fitting waveform at each scale factor and time factor is calculated. Finally, the scale factor and time factor corresponding to the least sum of squared errors can be judged to be the best match scale factor and time factor respectively, and the Morlet wavelet function fitting waveform in that scale factor and time factor can be seen as the optimized Lamb wave spatial sampling signal. The validation experiment performed on a glass fiber epoxy composite plate shows that the proposed method can improve the spatial resolution and length of the Lamb wave spatial sampling signal, and the sum of squared errors of this method is no more than 0.2.
文摘Objective: to comprehensively improve the effect of nursing services for prevention of delirium in emergency critically ill patients. Methods: in order to effectively improve the clinical nursing effect of critically ill patients with delirium, combined with the assessment and management process of delirium for critically ill patients treated in our hospital from January 2019 to January 2020, and selected the patients without delirium and delirium. The differences between the two groups were compared, the causes of delirium were analyzed and targeted nursing measures were taken, and the APACHE-II scores of the delirium group before and after nursing were compared. Results: there were significant statistical differences in age, respiratory failure, liver diseases, hypotensive shock, electrolyte disturbance and APACHE-II score between the delirium group and the delirium group: P < 0.05. Compared with before nursing, the APACHE-II score of patients in delirium group was significantly lower after nursing, P < 0.05. Conclusion: the incidence rate of delirium in emergency critically ill patients is relatively high. The early control of the risk factors of delirium patients and the provision of high-quality nursing services can prevent and reduce the incidence rate of delirium, promote the benign outcome of patients, and have positive significance for the treatment of patients.
