With the impact of the COVID‐19 epidemic, the demand for masked face recognition technology has increased. In the process of masked face recognition, some problems such as less feature information ...and poor robustness to the environment are obvious. The current masked face recognition model is not quantified enough for feature extraction, there are large errors for faces with high similarity, and the categories cannot be clustered during the detection process, resulting in poor classification of masks, which cannot be well adapted to changes in multiple environments. To solve current problems, this paper designs a new masked face recognition model, taking improved Single Shot Multibox Detector (SSD) model as a face detector, and replaces the input layer VGG16 of SSD with Deep Residual Network (ResNet) to increase the receptive field. In order to better adapt to the network, we adjust the convolution kernel size of ResNet. In addition, we fine‐tune the Xception network by designing a new fully connected layer, and reduce the training cycle. The weights of the three input samples including anchor, positive and negative are shared and clustered together with triplet network to improve recognition accuracy. Meanwhile, this paper adjusts alpha parameter in triplet loss. A higher value of alpha can improve the accuracy of model recognition. We further adopt a small trick to classify and predict face feature vectors using multi‐layer perceptron (MLP), and a total of 60 neural nodes are set in the three neural layers of MLP to get higher classification accuracy. Moreover, three datasets of MFDD, RMFRD and SMFRD are fused to obtain high‐quality images in different scenes, and we also add data augmentation and face alignment methods for processing, effectively reducing the interference of the external environment in the process of model recognition. According to the experimental results, the accuracy of masked face recognition reaches 98.3%, it achieves better results compared with other mainstream models. In addition, the hyper‐parameters tuning experiment is carried out to improve the utilization of computing resources, which shows better results than the indicators of different networks.
State of Washington is the one of key producers of grapevine (ranked number 2) and blueberry (ranked number 1) crops in the U.S. Abiotic and biotic crop stressors cause series of morphological, ...physiological and biochemical changes to these berry crop plants. In abiotic crop stress management, growers face critical challenge to detect and manage viral diseases, such as grapevine leafroll disease (GLD) and is one of key hindrance for sustainability of the state’s grape and wine industry. Similarly, freeze damage to the blueberry buds is considered as the major hindering factor for its production in the central Washington. Conventionally, laboratory-based laborious and destructive methods are used for detecting of the GLD virus symptoms and blueberry buds freeze damage in respective berry crops. The critical need is thus to have rapid and nondestructive method for resolving such issues. Recently, hyperspectral imaging (HSI) has been explored to detect crop stressors as rapid, noncontact and often nondestructive method. Ideally, HSI data can help identify the most sensitive wavelengths which then can be used to build the miniaturized and portable optical sensing module for field level detection of crop stressors. The overall goal of this research was to find important spectral signatures for detecting the GLD symptoms in a red-fruited wine grape cultivar and freeze damage of the blueberry buds. Collectively, findings indicated that the individual feature wavelength was not sensitive sufficiently to detect the GLD infected samples, which was leading to find the combination of few wavelengths. Analytics aided in finalizing six salient common wavelengths (690, 715, 731, 1409, 1425 and 1582 nm) for identification of GLD infected leaves at early stage. Furthermore, the individual feature wavelength was not sensitive to detect the bud injury. The combination of nine salient common wavelengths (566, 599, 698, 715, 896, 1012, 1384, 1442 and 1599 nm) were found reliable to detect bud freeze injury levels (low, medium, and severe mortality). The results achieved in this study laid a foundation for future studies to develop customized multispectral sensor for identifying GLD-infected grapevines and freeze injured blueberry buds in field.
Depression, a common psychiatric illness and global public health problem, remains poorly understood across different life stages, which hampers the development of novel treatments.
To identify new ...candidate genes for therapeutic development, we performed differential gene expression analysis of single-nucleus RNA sequencing data from the dorsolateral prefrontal cortex of older adults (n = 424) in relation to antemortem depressive symptoms. Additionally, we integrated genome-wide association study results for depression (n = 500,199) along with genetic tools for inferring the expression of 14,048 unique genes in 7 cell types and 52 cell subtypes to perform a transcriptome-wide association study of depression followed by Mendelian randomization.
Our single-nucleus transcriptome-wide association study analysis identified 68 candidate genes for depression and showed the greatest number being in excitatory and inhibitory neurons. Of the 68 genes, 53 were novel compared to previous studies. Notably, gene expression in different neuronal subtypes had varying effects on depression risk. Traits with high genetic correlations with depression, such as neuroticism, shared more transcriptome-wide association study genes than traits that were not highly correlated with depression. Complementing these analyses, differential gene expression analysis across 52 neocortical cell subtypes showed that genes such as KCNN2, SCAI, WASF3, and SOCS6 were associated with late-life depressive symptoms in specific cell subtypes.
These 2 sets of analyses illustrate the utility of large single-nucleus RNA sequencing data both to uncover genes whose expression is altered in specific cell subtypes in the context of depressive symptoms and to enhance the interpretation of well-powered genome-wide association studies so that we can prioritize specific susceptibility genes for further analysis and therapeutic development.
•The measured locations of watermelons did not affect the second resonant frequency f2.•The repeatability of f2 measured by the LDV system was high.•The stiffness coefficient S2 had better linear ...relationships with firmness variables.
Firmness is an important factor in describing the quality of agricultural products and is correlated with the vibrational characteristics of the object. In this study, the vibration response of ‘Qilin’ watermelons at postharvest was measured with an experimental system based on a laser Doppler vibrometer (LDV) for firmness detection. The vibration excitation applied by an electrodynamic shaker was monitored simultaneously with an accelerometer. After the excitation and response signals were transformed to the same dimension and converted from time-domain into frequency-domain by fast Fourier transform (FFT) processing, the ratio of response to excitation was calculated to determine the second resonance frequency (f2). Subsequently, three widely used stiffness coefficients (S1=f22m, S2=f22m2/3ρ1/3 and S3=f22m2/3, where m is the sample mass and ρ is the sample density) were calculated. These coefficients were selected as vibration parameters in addition to f2. Moreover, a puncture test was conducted to obtain reliable firmness variables from force/deformation curves, including maximum force (Fmax) and mean force at a 3–10mm distance (Fave). The effect of the measured locations of watermelons on f2 was not significant, and a relatively stronger linear relationship was observed between S2 and Fmax (r=0.410 with P<0.01). However, no strong relations could be established between the vibration parameters and the firmness variables. This was most likely because of the firmness reference method, the watermelon variety or the small distributions of weight and density of the test samples. Further efforts are needed to identify the reasons for the weak relations.
•Leaf level VIS–NIR sensing was assessed for grapevine GLRaV–3 detection.•Salient feature wavelengths were 1001, 1027 and 1052 nm.•Features were robust to detect GLRaV–3 symptoms at asymptomatic ...stage.•QDA performed better than Naïve Bayes in classifying infected samples.
Grapevine leafroll disease (GLD) is one of the major threats to wine grapes (Vitis vinifera) causing substantial economic losses to the growers. This study was undertaken to evaluate the applicability of visible and near infrared (VIS-NIR) spectroradiometery as a rapid, robust and non–destructive optical sensing method for the detection of Grapevine leafroll-associated virus 3 (GLRaV-3) at different phenological stages in a red-berried wine grape cultivar. Using VIS-NIR spectroradiometer, data was collected from the healthy and GLRaV-3-infected leaf samples from cv. Cabernet Sauvignon for two seasons at specific intervals during asymptomatic and symptomatic stages of the disease. Fiber optic leaf clip was used to collect spectral responses from grapevine leaves under field conditions. Salient feature extraction using stepwise multilinear regression and partial least square regression methods showed significant differences between healthy and virus–infected leaves in the visible (351, 377, 501, 526, 626 and 676 nm) and near infrared (701, 726, 826, 901, 951, 976, 1001, 1027, 1052 and 1101 nm) regions. Spectral wavelengths from near infrared region (1001, 1027 and 1052 nm) were validated at different phenological stages spanning both asymptomatic and symptomatic stages of the disease. Selected spectral wavelengths demonstrated robustness in virus detection with overall classification accuracies in the range of 75–99% using quadratic discriminant analysis (QDA) classifier. QDA based classification accuracies for healthy, infected and overall classes were significantly higher compared to Naïve Bayes classifier. The accuracy for virus detection during asymptomatic stages was not significantly different from the symptomatic phase, indicating reliability of the selected features for early detection of GLRaV–3–infected grapevines.
The relationship between genetic variation and gene expression in brain cell types and subtypes remains understudied. Here, we generated single-nucleus RNA sequencing data from the neocortex of 424 ...individuals of advanced age; we assessed the effect of genetic variants on RNA expression in cis (cis-expression quantitative trait loci) for seven cell types and 64 cell subtypes using 1.5 million transcriptomes. This effort identified 10,004 eGenes at the cell type level and 8,099 eGenes at the cell subtype level. Many eGenes are only detected within cell subtypes. A new variant influences APOE expression only in microglia and is associated with greater cerebral amyloid angiopathy but not Alzheimer's disease pathology, after adjusting for APOEε4, providing mechanistic insights into both pathologies. Furthermore, only a TMEM106B variant affects the proportion of cell subtypes. Integration of these results with genome-wide association studies highlighted the targeted cell type and probable causal gene within Alzheimer's disease, schizophrenia, educational attainment and Parkinson's disease loci.
Due to the illumination, complex background, and occlusion of the litchi fruits, the accurate detection of litchi in the field is extremely challenging. In order to solve the problem of the low ...recognition rate of litchi-picking robots in field conditions, this study was inspired by the ideas of ResNet and dense convolution and proposed an improved feature-extraction network model named YOLOv3_Litchi, combining dense connections and residuals for the detection of litchis. Firstly, based on the traditional YOLOv3 deep convolution neural network and regression detection, the idea of residuals was to be put into the feature-extraction network to effectively avoid the problem of decreasing detection accuracy due to the excessive depths of the network layers. Secondly, under the premise of a good receptive field and high detection accuracy, the large convolution kernel was replaced by a small convolution kernel in the shallow layer of the network, thereby effectively reducing the model parameters. Finally, the idea of feature pyramid was used to design the network to identify the small target litchi to ensure that the shallow features were not lost and simultaneously reduced the model parameters. Experimental results show that the improved YOLOv3_Litchi model achieved better results than the classic YOLOv3_DarkNet-53 model and the YOLOv3_Tiny model. The mean average precision (mAP) score was 97.07%, which was higher than the 95.18% mAP of the YOLOv3_DarkNet-53 model and the 94.48% mAP of the YOLOv3_Tiny model. The frame frequency was 58 fps, which was higher than 29 fps of the YOLOv3_DarkNet-53 model. Compared with the classic Faster R-CNN model with the feature-extraction network VGG16, the mAP was increased by 1%, and the FPS advantage was obvious. Compared with the classic single shot multibox detector (SSD) model, both the accuracy and the running efficiency were improved. The results show that the improved YOLOv3_Litchi model had stronger robustness, higher detection accuracy, and less computational complexity for the identification of litchi in the field conditions, which should be helpful for litchi orchard precision management.
Maturity is a key attribute to evaluate the quality and acceptability of fruit products. In this study, the impact method was used for nondestructive measurement of kiwifruit maturity. The fruit was ...vertically dropped onto an impact plate, and an accelerometer was used to measure the response signal. Then, fruit firmness, soluble solid content (SSC), titratable acidity (TA), and sensory scores were measured to determine the kiwifruit maturity. In addition, different modeling methods were proposed for data analysis. The results showed that the optimized prediction results were obtained by the principal component analysis-back-propagation neural network (PCA-BPNN) method for both quantitative and qualitative analysis. The optimized correlation coefficient between prediction and actual values (rsub.p) and root mean square error of prediction (RESEP) for firmness, SSC, TA, and sensory score were 0.881 (2.359N), 0.641 (1.511 Brix), 0.568 (0.023%), and 0.935 (0.693), respectively. The optimized discriminant accuracy for immature, mature, and overmature kiwifruits was 94.2% and 92.1% for calibration and validation, respectively. Such results indicated the feasibility of the proposed impact method for kiwifruit maturity evaluation.
BACKGROUND: Highbush blueberry (Vaccinium corymbosum), the species primarily grown in the state of Washington, U.S., is relatively cold hardy. However, low temperatures in winter and early spring can ...still cause freeze damage to the buds. OBJECTIVE: This study explored hyperspectral imaging (HSI) for detecting freeze induced bud damage. Blueberry buds (c.v. Duke) were collected over two seasons and tested in the laboratory to detect damage at four typical phenological stages. METHODS: The HSI data was acquired via line scan HSI system with spectral wavelength ranging from 517 to 1729 nm for buds grouped into either normal or injured mortalities. The successive projection algorithm was employed for pertinent feature wavelength selection. Analysis of variance and linear regression were then applied for evaluating sensitivity of feature wavelengths. RESULTS: Overall, five salient wavelengths (706, 723, 872, 1384, and 1591 nm) were selected to detect bud freeze injury. A quadratic discriminant analysis method-based analysis verified reliability of these five wavelengths in bud damage detection with overall accuracy in the ranges of 64 to 82%for the test datasets of each stage in two seasons. CONCLUSIONS: This study indicated potential of optical sensing to identify the injured buds using five salient wavelengths.