Implication of Expression of Osteopontin and Its Receptor Integrin αvβ3 in the Placenta in the Development of Preeclampsia

To investigate the expression of osteopontin （OPN） and its receptor integrin αvβ3 in the placental tissue from pregnant women complicated with preeclampsia, the expression of OPN and αvβ3 in the placenta of the pregnant women with preeclampsia and healthy pregnant women was detected by immunohistochemistry, Western blotting and RT-PCR. Our results showed that OPN and αvβ3 protein were expressed in the placenta from normal pregnant woman and those with preeclampsia. OPN was located in the placental syncytiotrophoblasts and the cytoplasm of capillary endothelial cells and integrin αvβ3 was mainly expressed on the surface of trophoblast cells. Expression of OPN and integrin αvβ3 in the placental tissue from preeclampsia subjects was significantly lower than that from the control group （P〈0.05）. Compared with the control group, expression of OPN in the placental tissue from preeclampsia group was significantly lower （P〈0.05） but there was no significant difference in the expression of αv and β3 between the preeclampsia group and the controls. It is concluded that OPN and its receptor integrin αvβ3 may be involved in the pathogenesis of preeclampsia.

Effects of Integrins and Integrin αvβ3 Inhibitor on Angiogenesis in Cerebral Ischemic Stroke

Summary： Integrins such as αvβ3, α5β31 play a key role in angiogenesis regulation, invasion and metastasis, inflammation, wound healing, etc. The up-regulation of integrin αvβ3 after cerebral ischemic stroke can promote angiogenesis, which in turn improves functional recovery. In addition, the integrin αvβ3 inhibitor can block the blood-brain barrier （BBB） leakage induced by vascular endothelial growth factor （VEGF） and also can reduce inflammatory reaction, decrease the deposition of fibrinogen. Other studies showed that integrin αvβ3 is not essential in revascularization. Therefore, the effect of integrin αvβ3 in the whole process of brain function recovery merits further study.

Effect of lentivirus-mediated integrin αVβ3-sh RNA on tumor growth of mice with lung cancer xenografts

Objective: To study the effect of lentivirus-mediated integrin αVβ3-sh RNA on tumor growth of mice with lung cancer xenograft. Methods: Lung cancer tissue, paracancer tissue and normal tissue were collected and integrin αVβ3 expression was detected; BALB/c nude mice were selected, divided into integrin αVβ3 knockdown group(KD group) and negative control group(NC group), and inoculated with cells stably infected by integrin αVβ3-sh RNA lentivirus and cells stably infected by negative control-sh RNA lentivirus respectively, the growth of tumor tissue was continuously observed, and the number of apoptosis cells as well as the expression of angiogenesis, apoptosis and invasion genes in tumor tissue were detected. Results: m RNA content and protein content of integrin αVβ3 in lung cancer tissue were significantly higher than those in paracancer tissue and normal tissue; increasing trend of tumor tissue volume of KD group was weaker than that of NC group, and tumor volume at various points in time of KD group was lower than that of NC group; m RNA contents and protein contents of VEGF, FGF, EGF, Bcl-2, MMP-9, MMP-12 and MMP-13 in tumor tissue of KD group were lower than those of NC group, and apoptosis index as well as m RNA content and protein content of Bax were higher than those of NC group. Conclusions: The expression of integrin αVβ3 increases in lung cancer tissue, and lentivirus-mediated integrin αVβ3-sh RNA can inhibit tumor growth of mice with lung cancer xenografts.

Effects of controlled superovulation on the expression of HOXA10,integrin αvβ3 and LIF in mice during peri-implantation

Objective:To observe the expression of endometrial homologous box gene A10(HOXA10),Integrinαvβ3 and leukemia suppressor factor(LIF)in peri-implantation mice,and to investigate the effect of controlled superovulation(COH)on endometrial receptivity in mice.Methods:After COH model preparation and blastocyst transplantation,kunming pure-bred mature mice were randomly divided into natural blastocyst+natural pseudocyst group,COH blastocyst+natural pseudocyst group,natural blastocyst+COH pseudocyst group,and COH blastocyst+COH pseudocyst group.The pregnancy rate and average number of beds in each group were observed on day 6 of conception.Western blot and RT-PCR were used to detect the expression of HOXA10,Integrinααvβ3,LIF protein and mRNA in the endometria of mice on day 5 of conception.Results:The pregnancy rate and average number of beds in the COH group were lower than those in the natural receptor group.The expression of HOXA10,Integrinααvβ3 and LIF in endometrium was significantly lower than that of the natural receptor mice(P<0.01).Conclusion:Superovulation drugs down-regulate the expression of HOXA10,Integrinαvβ3 and LIF in endometrium during the implantation window,reduce endometrial receptivity,and lead to low clinical pregnancy rate.

Non-invasively differentiate non-alcoholic steatohepatitis by visualizing hepatic integrinαvβ3 expression with a targeted molecular imaging modality

BACKGROUND Non-invasive methods to diagnose non-alcoholic steatohepatitis(NASH),an inflammatory subtype of non-alcoholic fatty liver disease(NAFLD),are currently unavailable.AIM To develop an integrinαvβ3-targeted molecular imaging modality to differentiate NASH.METHODS Integrinαvβ3 expression was assessed in Human LO2 hepatocytes Scultured with palmitic and oleic acids(FFA).Hepatic integrinαvβ3 expression was analyzed in rabbits fed a high-fat diet(HFD)and in rats fed a high-fat,high-carbohydrate diet(HFCD).After synthesis,cyclic arginine-glycine-aspartic acid peptide(cRGD)was labeled with gadolinium(Gd)and used as a contrast agent in magnetic resonance imaging(MRI)performed on mice fed with HFCD.RESULTS Integrinαvβ3 was markedly expressed on FFA-cultured hepatocytes,unlike the control hepatocytes.Hepatic integrinαvβ3 expression significantly increased in both HFD-fed rabbits and HFCD-fed rats as simple fatty liver(FL)progressed to steatohepatitis.The distribution of integrinαvβ3 in the liver of NASH cases largely overlapped with albumin-positive staining areas.In comparison to mice with simple FL,the relative liver MRI-T1 signal value at 60 minutes post-injection of Gd-labeled cRGD was significantly increased in mice with steatohepatitis(P<0.05),showing a positive correlation with the NAFLD activity score(r=0.945;P<0.01).Hepatic integrinαvβ3 expression was significantly upregulated during NASH development,with hepatocytes being the primary cells expressing integrinαvβ3.CONCLUSION After using Gd-labeled cRGD as a tracer,NASH was successfully distinguished by visualizing hepatic integrinαvβ3 expression with MRI.

Integrin α_vβ_3靶向超声微泡的制备及体外寻靶实验

目的制备Integrinαvβ3靶向超声微泡,对其一般理化性质及体外寻靶能力进行检测。方法采用巯基-马来酰亚胺连接法制备携FITC标记iRGD肽的Integrinαvβ3靶向微泡(MBIntegrinαvβ3),并制备空白微泡(MBcontrol)。检测两组微泡的一般理化性质,在荧光显微镜下观察两组微泡荧光显像,流式细胞仪检测两组微泡FITC荧光含量。于光学显微镜及荧光显微镜下观察两组微泡与小鼠脑微血管内皮细胞(bEnd.3)的黏附情况。结果 1MBIntegrinαvβ3粒径为(0.84±0.26)μm,与MBcontrol粒径(0.79±0.19)μm差异无统计学意义(P>0.05);2荧光显微镜下观察,MBcontrol组微泡无荧光显示,MBIntegrinαvβ3组微泡表面显示绿色荧光;3流式细胞仪测得MBIntegrinαvβ3组微泡FITC荧光含量为98.4%,MBcontrol组微泡FITC荧光含量为2.5%;4体外寻靶实验显示,MBIntegrinαvβ3组微泡能聚集结合于bEnd.3细胞表面。结论成功制备了携iRGD肽的Integrinαvβ3靶向微泡,其具有良好的体外寻靶能力。

Ischemic preconditioning partially suppresses and postpones integrin α_Vβ_3 mRNA expression following transient global cerebral ischemia in C57BL/6 mice

Previous studies of integrin αvβ3 have focused on ischemic brain damage, although the role of integrin αvβ3 in ischemic preconditioning （IP） has rarely been reported. The present study analyzed the effects of IP on integrin αvβ3 mRNA expression following cerebral ischemia through the use of hematoxylin-eosin staining and real-time quantitative polymerase chain reaction techniques. Integrin avid3 mRNA expression in the ischemia group peaked at 24 hours after ischemia-reperfusion. In the IP ＋ ischemia group, integrin αvβ3 mRNA expression increased after 24 hours, but remained significantly less than the ischemia group, and expression continued to increase until 7 days after ischemiaJreperfusion. These results demonstrate that IP effectively attenuated upregulation of integrin αvβ3 mRNA expression at 24 hours after ischemia.

基于改进DeepLabv3+的轻量化作物杂草识别方法

为在存储资源与计算能力有限的设备上实现田间作物和杂草的识别,本文提出一种基于改进DeepLabv3+的轻量化语义分割网络。首先,以MobileNet v2作为DeepLabv3+的特征提取骨干网络,提出双分支残差模块替换倒残差模块,并删除后两层卷积以降低模型参数量。其次,在空洞空间金字塔池化(Atrous Spatial Pyramid Pooling,ASPP)模块中引入分组逐点卷积,使用深度扩张卷积替换标准卷积,并将卷积后的特征图进行多尺度特征融合增强对作物和杂草深层特征的提取能力。最后,将原有的非线性激活函数替换为Leaky ReLU激活函数来提升分割精度。实验结果表明:改进后网络的mIOU达到86.75%,参数量仅为0.69M,FPS达到了98,与原始DeepLabv3+以及3个典型轻量化语义分割网络的相比,参数量最小,在对比的轻量化网络中具有最高的分割精度。

罗汉果皂苷V对RSL3诱导的SH-SY5Y细胞铁死亡的抑制作用及其机制

目的:探讨罗汉果皂苷V(MV)对铁死亡诱导剂RAS选择性致死分子3(RSL3)诱导的人神经母细胞瘤SH-SY5Y细胞铁死亡的抑制作用及可能机制。方法:用RSL3诱导SH-SY5Y细胞建立铁死亡模型。MTT法检测细胞活力;倒置显微镜观察细胞形态;亚铁离子荧光探针FerroFarRed检测细胞内亚铁离子含量;线粒体红色荧光探针MitoTracker Red CMXRos检测线粒体膜电位(MMP);超氧化物阴离子荧光探针二氢乙啶和线粒体超氧化物红色荧光探针MitoSoX Red分别检测细胞内和线粒体内活性氧(ROS)。微板法检测细胞谷胱甘肽(GSH)和丙二醛(MDA)水平。Western blot检测脂酰辅酶A合成酶长链家族成员4(ACSL4)、环加氧酶2(COX-2、)谷胱甘肽过氧化物酶4(GPX4)和溶质载体家族7成员11(SLC7A11)蛋白表达水平。分子对接技术预测MV与ACSL4、COX-2、GPX4和SLC7A11的靶向关系。结果:与control组相比,RSL3组SH-SY5Y细胞活力显著降低(P<0.01),细胞内亚铁离子含量、细胞内和线粒体内ROS水平及MDA水平显著升高(P<0.05或P<0.01),MMP和GSH水平显著降低(P<0.01),ACSL4和COX-2蛋白表达水平显著升高,而GPX4和SLC7A11蛋白表达水平显著降低(P<0.01),提示成功建立了细胞铁死亡模型。MV处理使细胞活力显著升高(P<0.05),细胞内亚铁离子含量、细胞内和线粒体内ROS水平及MDA水平显著降低(P<0.01),MMP和GSH水平显著升高(P<0.05或P<0.01);ACSL4和COX-2蛋白水平显著降低,而GPX4和SLC7A11蛋白水平显著升高(P<0.05或P<0.01)。分子对接结果显示,MV与铁死亡核心蛋白ACSL4、COX-2、GPX4和SLC7A11存在结合位点。结论:MV可抑制RSL3诱导的SH-SY5Y细胞铁死亡的发生,其机制可能与激活SLC7A11/GPX4和抑制ACSL4/COX-2有关。

改进DeepLab v3+模型下的梯田遥感提取研究

[目的与意义]梯田作为农业生产的关键要素之一,其面积估算对于农业政策制定、土地规划和资源管理至关重要。为解决复杂的地形条件、种植环境导致传统遥感数据和监测方法难以开展梯田自动化提取问题,探索一种利用深度学习技术在高分辨率遥感影像中精准提取梯田面积的方法。[方法]以休耕期梯田高分六号影像构建语义分割数据集,同时提出一种改进的DeepLab v3+模型。该模型使用轻量级网络MobileNet v2作为骨干网络,为了同时兼顾局部细节和全局语境,使用多尺度特征融合(Multi-scale Feature Fusion module,MSFF)模块代替空洞空间金字塔池化(Atrous Spatial Pyramid Pooling,ASPP)模块,利用扩张率依次增大的空洞卷积级联模式改善信息丢失的问题。此外,对浅层特征和深层特征使用坐标注意力机制以加强网络对于目标的学习。[结果与讨论]利用红、绿和近红外波段组合方式在梯田提取的精度和效果上表现最佳。相比于原始DeepLab v3+网络,精确率、召回率、F_(1)评分和交并比指标分别提升4.62%、2.61%、3.81%和2.81%。此外,与UNet和原始DeepLab v3+相比,改进的DeepLab v3+在参数量上和浮点运算数有着更为优越的性能,其参数量仅为UNet的28.6%和原始DeepLab v3+的19.5%,同时浮点运算数仅为UNet和DeepLab v3+的1/5。这不仅提高了计算效率,也使得改进后的模型更适用于资源有限或计算能力较低的环境中。[结论]深度学习在高分辨率遥感影像梯田识别中具有较高的精度,有利于为梯田精细化监测和管理提供参考依据。

CXCR4通过integrinαVβ3调控头颈部鳞癌细胞迁移的机制研究

目的:研究C-X-C型趋化因子受体4(CXCR4)-整合素αVβ3(integrinαVβ3)介导头颈部鳞癌细胞迁移的分子生物学机制。方法:应用头颈部鳞癌细胞株HN-5,采用慢病毒转染方法构建CXCR4过表达的细胞株,用荧光显微镜观察基因表达情况;采用脂质体法将CXCR4-siRNA、integrinαVβ3-siRNA分别转染至HN-5细胞和CXCR4过表达的HN-5细胞中,Transwell小室实验检测HN-5细胞迁移能力,Western blot实验检测CXCR4、integrinαVβ3、基质金属蛋白酶-9(MMP-9)、基质金属蛋白酶-2(MMP-2)和黏附斑激酶(FAK)-非受体型酪氨酸蛋白激酶(SRC)-细胞外信号调节激酶(ERK)信号通路相关蛋白的表达情况。结果:CXCR4过表达可上调HN-5细胞中integrinαVβ3表达水平,而抑制CXCR4表达可下调integrinαVβ3表达;CXCR4过表达可调控MMP-2和MMP-9促进癌细胞迁移,并激活FAK-SRC-ERK信号通路;而抑制integrinαVβ3表达可明显逆转CXCR4过表达对HN-5细胞的上述作用。结论:CXCR4过表达可通过integrinαVβ3促进头颈部鳞癌细胞迁移,其作用机制可能与激活FAK-SRC-ERK通路有关。

改进Deeplabv3+的双注意力融合作物分类方法

近年来,卷积神经网络(convolutional neural networks,CNN)在农作物分类研究中不断取得新进展,但在建模长期依赖关系方面表现出一定的局限性,对农作物全局特征的捕获存在不足。针对以上问题,将Transformer引入Deeplab v3+模型,提出了一种用于无人机影像农作物分类的并行分支结构--DeepTrans(Deeplab v3+with Trans-former)模型。DeepTrans以一种并行的方式将Transformer和CNN结合在一起,利于全局特征与局部特征的有效捕获。通过引入Transformer来增强图像中信息的远距离依赖关系,提高了作物全局信息的提取能力;加入通道注意力机制和空间注意力机制加强Transformer对通道信息的敏感度及ASPP(atrous spatial pyramid pooling)对作物空间信息捕获能力。实验结果表明,DeepTrans模型在MIoU指标上可达0.812,相较于Deeplab v3+模型提高了3.9%,该模型在五类作物的分类中精度均有提升,对于容易错分的甘蔗、玉米和香蕉三种作物,其IoU分别提高了2.9%、4.7%、13%。由此可见,DeepTrans模型在农作物分类图像的内部填充和全局预测方面有着更好的分割效果,有助于更准确地监测农田作物的种植结构及规模。

基于v3洋葱域名的比特币地址威胁程度分析

比特币可以在不透露使用者身份的情况下进行交换,导致其成为不法分子在暗网上进行违法活动的主要方式。为了追踪比特币非法交易,传统方法根据比特币的伪匿名性,在整个区块链上进行启发式地址聚类,没有充分利用比特币地址在暗网上的信息。2021年Tor官方全面启用v3洋葱域名,使得以往的v2洋葱域名数据无法再作为分析的依据。设计并实现基于v3洋葱域名的比特币地址威胁程度的一体化分析框架TLAFDB。信息收集模块使用境外服务器解决地域限制并设置socks5h代理以支持暗网爬虫运行,使用洋葱种子地址在暗网中爬行收集最新的v3洋葱域名数据,信息清洗模块采用可同时覆盖Base58和Bech32编码的正则表达式以提取v3洋葱域名网页中的比特币地址,通过区块链搜索引擎Blockchain.com筛选存在真实交易的比特币地址,并建立其和所在v3洋葱域名的关联关系,信息分析模块采用人工分析和关键词匹配相结合的方法分类v3洋葱域名,赋予其关联的比特币地址类别和流行度并判定威胁程度。实验结果表明,TLAFDB收集了23627个v3洋葱域名网页,提取并分析1141个存在真实交易的比特币地址的类别、流行度和威胁程度,发现在暗网中同一个比特币地址常出现在大量的镜像洋葱域名网页上,超过95%的比特币地址被恶意使用,并且庞氏骗局交易量占高危比特币地址总交易量的99%。

基于改进MobileNet v3的苹果叶片病害识别研究

为解决移动端和嵌入式设备中苹果叶片病害识别准确率不高、效率低下的问题,提出了一种新的基于MobileNet v3网络的分类模型,以实现更加高效和准确的苹果叶片病害识别。首先通过数据增广方法增强数据集,按照9∶1的比例划分训练集和验证集;然后在MobileNet v3网络核心倒残差结构的升维部分引入全维动态卷积,以加强对不同维度注意力权重的学习,从而增强网络的拟合能力;最后在降维部分引入修改后的ConvNext Block模块,减少信息损失并增加全局感受野。采用PyTorch作为分类网络的深度学习框架,使用交叉熵损失函数作为分类任务的损失函数,Adam作为优化器,通过多组对比试验可知,MobileNet v1、MobileNet v2、ResNet34、MobileNet v3以及改进后的MobileNet v3 ODConvNext网络的准确率分别为94.5%、95.7%、97.2%、96.9%及97.5%。可见,MobileNet v3 ODConvNet网络拥有最高的Top-1准确率,相较于MobileNet v3网络和结构更为复杂的ResNet34网络分别提升了0.6、0.3百分点;在运算频率方面,相对于MobileNet v3网络仅增加了1.00×10^(6)次/s,并且仅为ResNet34网络参数量的11.84%。因此,该试验结果证明了改进后的MobileNet v3 ODConvNext模型具有更加轻量级和更高准确率的优点,满足在移动端真实场景下进行苹果叶片病害识别的要求,有助于苹果叶片病害的防治工作。

基于DeepLab v3+的综放工作面含矸率预测研究

针对综放工作面真实煤矸堆叠状态下的体积含矸率很难获取的问题,提出一种基于DeepLab v3+的综放工作面含矸率预测方法。构建了煤矸堆积体图像数据集,采用半自动的数据标注方法和限制对比度自适应直方图均衡化法对煤矸图像进行预处理。运用DeepLab v3+模型进行煤矸图像语义分割,进而计算煤矸图像的投影面积含矸率。利用PFC3D数值模拟软件,基于重建的三维煤矸块体建立数值模型,模拟顶煤放落和刮板输送机运煤过程,通过fish语言读取每个矸石或煤的体积,计算得到煤矸堆积体体积含矸率。通过分析不同顶煤厚度条件下刮板输送机上煤矸堆积体的投影面积含矸率与体积含矸率的量化关系,建立了煤流的体积含矸率预测模型。实验结果表明:DeepLab v3+模型的准确率、平均像素准确率和平均交并比分别为97.68%,97.72%,95.33%,均高于经典语义分割模型FCN8s和PSPNet,实现了煤矸堆积体投影面积含矸率的精准快速识别;体积含矸率预测模型的决定系数R^(2)为0.9828,预测效果较好。

基于改进MobileNet v3的苹果叶片病害识别方法及移动端应用

准确识别苹果叶片病害种类以进行及时防治对于苹果增量增产具有重要的意义,为实现在移动设备实时对苹果叶片进行病害识别,提高苹果的产量,减少种植者的损失。首先收集了黑星病、斑点落叶病、锈病、白粉病、混合病、褐斑病等6种苹果叶部病害和健康叶片的图像,并使用Retinex算法对图像进行数据增强,以提高数据集质量,然后将数据集按照8∶1∶1的比例划分为训练集、验证集和测试集;其次对MobileNet v3网络模型进行改进优化调整,在精简网络结构的同时减少冗余参数,并在非线性激活层后加入批归一化层,以提高网络的特征提取能力;同时,为了提升在低精度移动设备上的准确性和模型运行效率,将全连接层中的激活函数替换为ReLU6函数;最后,在模型训练时使用动量随机梯度下降优化器来进行模型权重系数的寻优,以减少训练时间和达到更高的分类准确率。试验结果表明,改进后的MobileNet v3-A3网络对苹果叶片病害图像的识别准确率为96.48%,模型权重为2.98 MB,识别速率为8.82 ms/幅图片,与其他同量级卷积神经网络相比识别精度更高、模型更小、识别速度更快。本研究使用Android Studio将权重模型封装到安卓软件中,实现了移动设备对苹果叶片病害的准确快速识别。

香烟烟雾提取物对破骨细胞体积及整合素αvβ3基因表达的影响

目的探讨香烟烟雾提取物(cigarette smoke extract,CSE)对破骨细胞体积及整合素αvβ3基因表达的影响。方法首先制作CSE溶液,实验采用7周龄C57BL/6J小鼠骨髓单个核细胞加入M-CSF和RANKL诱导生成破骨细胞,将浓度为5%的CSE加入实验组破骨细胞作用3 d,对照组不加入CSE。通过Trap染色识别对照组和实验组破骨细胞并于光镜下采图;通过甲苯胺蓝染色识别两组破骨细胞在骨片上形成的骨陷窝并于光镜下采图,均用NIH imageJ计算数量及面积。用实时荧光定量聚合酶链反应,以GAPDH作为内参,最后通过2-ΔΔCT计算,以检测Trap、CTR以及整合素αvβ3 mRNA表达水平。通过Western blot检测整合素αvβ3蛋白表达。结果破骨细胞表面积及细胞核数量的计量表明,实验组诱导形成的破骨细胞显著大于对照组(P<0.01)。实验组骨片上吸收陷窝数量及面积明显多于对照组(P<0.05),CSE处理组的TRAP mRNA水平为1.16±0.13,明显高于对照组(P<0.05),CTRmRNA水平为0.38±0.04,明显低于对照组(P<0.01)。CSE处理组的整合素αvβ3mRNA水平为1.75±0.05,显著高于对照组(P<0.01),蛋白质印迹分析结果也明显高于对照组(P<0.05)。结论CSE能显著增加破骨细胞的体积并上调整合素αvβ3的基因表达,并可能对骨吸收有促进作用。

基于改进YOLO v3的轴承端面缺陷检测算法

为提高轴承端面缺陷检测的速度以及检测精度,提出一种基于改进YOLO v3的轴承端面缺陷检测算法。首先,对图像数据集进行数据增强处理以防止产生过拟合现象;其次,通过改进K-means聚类算法重新聚类出目标检测的Anchor Boxes,并引入SKNet注意力机制模块对原网络结构以及输出层结构进行改进;最后对改进的YOLO v3算法进行实验验证,并与原YOLO v3算法进行对比分析。结果表明,改进后的YOLO v3算法相比原YOLO v3算法对轴承端面缺陷检测的mAP值提升了7.03%,检测速度提升了34.7帧/s,验证了改进算法的有效性。

基于改进MobileNet v3-Small模型的草莓病害识别方法

为了对草莓病害进行及时的诊断与治疗而提升草莓产量,将深度学习与农业生产结合以快速高效地进行病害检测。传统神经网络进行病害识别时间较长,参数量较大,难以迁移到移动端设备上,基于此提出一种改进MobileNet v3-Small模型的识别方法。首先收集了7类常见草莓病害图像样本(如角斑病、叶斑病等),通过旋转、镜像等多种数据增强方式对图像进行处理以增加图片数量,提高模型泛化能力。接着以MobileNet v3-Small模型为基础,基于原始Inception_A提出部分卷积权值共享的多尺度卷积结构,以更高效地提取草莓病害不同尺度特征。随后,在网络深层引入了ULSAM轻量级子注意力机制,形成草莓病害更高层次的抽象表示。同时,将深度可分离卷积中的第2个PW卷积替换为CondConv卷积形成PDC结构,克服了PW卷积只拥有局部感受野的缺陷,同时也降低了模型参数量。试验结果表明,改进后的MobileNet v3-Small模型准确率达到98.62%,较原模型94.91%的准确率提高了3.71百分点,并且参数量减少了0.04 M,远优于同级轻量化模型,且以远低于ResNet18的参数量取得更好的特征提取效果。综上所述,本研究所提出的改进后的MobileNet v3_Small模型能更好地在真实场景下进行草莓病害识别,为草莓生产贡献了一份力量,助力智慧农业发展。

基于Kinect V2和Unity3D的机械臂人机交互系统

针对人机交互方式存在不直观、操作复杂等问题,论文利用手部追踪技术开发出一种基于Kinect V2和Unity3D的人机交互系统,用户可以直接挥动手部就可以控制机械臂运动,交互方式更加简单、直观和灵活。该系统是通过Kinect V2来获得手部移动轨迹来实现手部追踪,并在Unity3D平台上进行交互系统的开发。利用FABRIK算法和贝塞尔曲线分别实现了逆运动学求解和运动路径的记录,通过C#编写所需的脚本功能并开发出了两种交互模式,分别为实时控制模式和示教模式,两种模式均可通过UI面板进行选择。最后对两种模式分别进行实验,实验结果证明了该系统的可行性。