Search

Zero-shot learning casts light on lacking unseen class data by transferring knowledge from seen classes via a joint semantic space. However, the distributions of samples from seen and unseen classes are usually imbalanced. Many zero-shot learning methods fail to obtain satisfactory results in the generalized zero-shot learning task, where seen and unseen classes are all used for the test. Also, irregular structures of some classes may result in inappropriate mapping from visual features space to semantic attribute space. A novel generative mixup networks with semantic graph alignment is proposed in this article to mitigate such problems. To be specific, our model first attempts to synthesize samples conditioned with class-level semantic information as the prototype to recover the class-based feature distribution from the given semantic description. Second, the proposed model explores a mixup mechanism to augment training samples and improve the generalization ability of the model. Third, triplet gradient matching loss is developed to guarantee the class invariance to be more continuous in the latent space, and it can help the discriminator distinguish the real and fake samples. Finally, a similarity graph is constructed from semantic attributes to capture the intrinsic correlations and guides the feature generation process. Extensive experiments conducted on several zero-shot learning benchmarks from different tasks prove that the proposed model can achieve superior performance over the state-of-the-art generalized zero-shot learning.

Unsupervised domain adaptation aims to learn a classification model for the target domain without any labeled samples by transferring the knowledge from the source domain with sufficient labeled samples. The source and the target domains usually share the same label space but are with different data distributions. In this paper, we consider a more difficult but insufficient-explored problem named as few-shot domain adaptation, where a classifier should generalize well to the target domain given only a small number of examples in the source domain. In such a problem, we recast the link between the source and target samples by a mixup optimal transport model. The mixup mechanism is integrated into optimal transport to perform the few-shot adaptation by learning the cross-domain alignment matrix and domain-invariant classifier simultaneously to augment the source distribution and align the two probability distributions. Moreover, spectral shrinkage regularization is deployed to improve the transferability and discriminability of the mixup optimal transport model by utilizing all singular eigenvectors. Experiments conducted on several domain adaptation tasks demonstrate the effectiveness of our proposed model dealing with the few-shot domain adaptation problem compared with state-of-the-art methods.

Rapid diagnosis based on naked-eye colorimetric detection remains challenging, but it could build new capacities for molecular point-of-care testing (POCT). In this study, we evaluated the performance of 16 types of single-stranded DNA-fluorophore-quencher (ssDNA-FQ) reporters for use with clusters of regularly spaced short palindrome repeats (CRISPR)/Cas12a-based visual colorimetric assays. Among them, nine ssDNA-FQ reporters were found to be suitable for direct visual colorimetric detection, with especially very strong performance using ROX-labeled reporters. We optimized the reaction concentrations of these ssDNA-FQ reporters for a naked-eye read-out of assay results (no transducing component required for visualization). In particular, we developed a convolutional neural network algorithm to standardize and automate the analytical colorimetric assessment of images and integrated this into the MagicEye mobile phone software. A field-deployable assay platform named RApid VIsual CRISPR (RAVI-CRISPR) based on a ROX-labeled reporter with isothermal amplification and CRISPR/Cas12a targeting was established. We deployed RAVI-CRISPR in a single tube toward an instrument-less colorimetric POCT format that required only a portable rechargeable hand warmer for incubation. The RAVI-CRISPR was successfully used for the high-sensitivity detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and African swine fever virus (ASFV). Our study demonstrates this RAVI-CRISPR/MagicEye system to be suitable for distinguishing different pathogenic nucleic acid targets with high specificity and sensitivity as the simplest-to-date platform for rapid pen- or bed-side testing.

The research and development of bio-degumming technology is under a slow progress due to the shortage of proper efficient bacterial strains and processes. A degumming bacterial strain-Pectobacterium wasabiae (PW)-with broad-spectrum degumming abilities was screened out in this study. After the fermentation for 12 h, the residual gum contents of kenaf bast, ramie bast, hemp bast, flax bast, and Apocynum venetum bast were all lower than 15%. This bacterial strain could realize the simultaneous extracellular secretion of pectinase, mannase, and xylanase with the maximum enzyme activity levels of 130.25, 157.58, and 115.24 U/mL, respectively. The optimal degumming conditions of this bacterial strain were as follows: degumming time of 12 h, bath ratio of 1:10, temperature of 33 °C, and inoculum size of 2%. After the bio-degumming through this bacterial strain, the COD in wastewater was below 4000 mg/L, which was over 60% lower than that in boiling-off wastewater generated by chemical degumming. This technology achieves higher efficiency, higher quality, and lower pollution.

Because it is very expensive to collect a large number of labeled samples to train deep neural networks in certain fields, semi-supervised learning (SSL) researcher has become increasingly important in recent years. There are many consistency regularization-based methods for solving SSL tasks, such as the $$\Pi $$ model and mean teacher. In this paper, we first show through an experiment that the traditional consistency-based methods exist the following two problems: (1) as the size of unlabeled samples increases, the accuracy of these methods increases very slowly, which means they cannot make full use of unlabeled samples. (2) When the number of labeled samples is vary small, the performance of these methods will be very low. Based on these two findings, we propose two methods, metric learning clustering (MLC) and auxiliary fake samples, to alleviate these problems. The proposed methods achieve state-of-the-art results on SSL benchmarks. The error rates are 10.20%, 38.44% and 4.24% for CIFAR-10 with 4000 labels, CIFAR-100 with 10,000 labels and SVHN with 1000 labels by using MLC. For MNIST, the auxiliary fake samples method shows great results in cases with the very few labels.

Early and rapid detection of Japanese encephalitis virus (JEV) is necessary for timely preventive and control measures. However, JEV RNA detection remains challenging due to the low level of viremia. In this study, a RApid VIsual CRISPR (RAVI-CRISPR) assay based on reverse transcription loop-mediated isothermal amplification (RT-LAMP) and CRISPR/Cas12a targeting was developed for easy detection of JEV in the field. We showed successful detection of 8.97 or more copies of the C gene sequence of JEV RNA within approximately 60 min. This assay also displayed no cross-reactivity with other porcine pathogens. We applied our one-tube RAVI-CRISPR assay to 18 brain tissue sample for JE diagnosis. The results from both fluorescence intensity measurements and directly naked-eye visualization were consistent with those from real-time PCR analysis. Taken together, our results showed that one-tube RAVI-CRISPR assay is robust, convenient, sensitive, specific, affordable, and potentially adaptable to on-site detection or surveillance of JEV in clinical and vector samples.

The growing demand for and supply of meat and meat products has led to a proportional increase in cases of meat adulteration. Adulterated meat poses serious economic and health consequences globally. Current laboratory methods for meat species identification require specialized equipment with limited field applications. This study developed an inexpensive, point-of-care Loop-Mediated Isothermal Amplification (LAMP)-CRISPR/Cas12a colorimetric assay to detect meat species using a Texas Red-labelled single-strand (ssDNA) reporter. As low as 1.0 pg/µL of the porcine NADH4, the chicken NADH dehydrogenase subunit 2 (ND2) and the duck D-loop genes was detectable under white, blue and ultraviolet light. The test turnaround time from DNA extraction to visualization was approximately 40 min. The assay accurately detected pure and mixed-meat products in the laboratory (n = 15) and during a pilot point-of-care test (n = 8) in a food processing factory. The results are 100% reproducible using lateral flow detection strips and the real-time PCR detection instrument. This technology is fully deployable and usable in any standard room. Thus, our study demonstrates that this method is a straightforward, specific, sensitive, point-of-care test (POCT) adaptable to various outlets such as customs, quarantine units and meat import/export departments.

Lumpy skin disease (LSD) is a severe and highly infectious pox disease of cattle caused by the lumpy skin disease virus (LSDV). To facilitate early control of LSD, this study aimed to develop a new rapid on-site LSDV detection method using an orf068 gene-based recombinase polymerase amplification assay (RPA) coupled with a CRISPR-Cas12a-based fluorescence assay (RPA-Cas12a-fluorescence assay). The results showed that the sensitivity of our RPA-Cas12a-fluorescence assay for detecting LSDV orf068 gene reached 5 copies/μL with plasmid as a template, and 102 TCID50/mL with viral genomic DNA as a template. No cross-reaction with other common bovine viruses was observed. Further, an on-site RPA-Cas12a-fluorescence assay of 40 clinical samples from cattle with or without LSD showed a diagnostic sensitivity of 96.3% (95% CI: 81.0–99.9%) and specificity of 92.31% (95% CI: 62.1–99.6%), which was close to those of the quantitative PCR assay. Therefore, our RPA-Cas12a-fluorescence assay has promising prospects in on-site rapid LSDV detection.

Sex selection technologies have immensely impacted swine production globally. Conventional earlier embryo sex identification methods require professional technicians and sophisticated laboratory instruments. Rapid on-site gender identification of porcine embryos and pork products remains challenging. In this study, we developed a CRISPR/Cas12a-based fluorescence visualization point-of-care sex determination test that is rapid, accurate and easy to implement on-site. The CRISPR/Cas12a assay coupled with either the polymerase chain reaction (PCR) or loop-mediated isothermal amplification (LAMP) employs precisely designed primers and single-guide RNAs targeting the sex-determining region Y (SRY) and the zinc finger protein X-linked (ZFX) genes. PCR and LAMP amplicons were cleaved with the subsequent generation of fluorescing products detectable with portable blue and ultraviolet light transilluminators. Approximately two copies per microliter of the ZFX and SRY genes were detected using the RApid VIsual CRISPR (RAVI-CRISPR) assay. This method is a sensitive, inexpensive, versatile, and point-of-care test. The technology has other potential applications like determining the sex of diverse livestock species, detecting livestock disease-causing pathogens and evaluating the quality of meat products.

Beyond its powerful genome-editing capabilities, the CRISPR/Cas system has opened up a new era of molecular diagnostics due to its highly specific base recognition and trans-cleavage activity. However, most CRISPR/Cas detection systems are mainly used to detect nucleic acids of bacteria or viruses, while the application of single nucleotide polymorphism (SNP) detection is limited. The MC1R SNPs were investigated by CRISPR/enAsCas12a and are not limited to the protospacer adjacent motif (PAM) sequence in vitro. Specifically, we optimized the reaction conditions, which proved that the enAsCas12a has a preference for divalent magnesium ion (Mg2+) and can effectively distinguish the genes with a single base difference in the presence of Mg2+, and the Melanocortin l receptor (MC1R) gene with three kinds of SNP sites (T305C, T363C, and G727A) was quantitatively detected. Since the enAsCas12a is not limited by PAM sequence in vitro, the method shown here can extend this extraordinary CRISPR/enAsCas12a detection system to other SNP targets, thus providing a general SNP detection toolbox.

Rapid diagnosis based on naked-eye colorimetric detection remains challenging, but it could build new capacities for molecular point-of-care testing (POCT). In this study, we evaluated the performance of 16 types of single-stranded DNA-fluorophore-quencher (ssDNA-FQ) reporters for use with CRISPR/Cas12a based visual colorimetric assays. Among them, 9 ssDNA-FQ reporters were found to be suitable for direct visual colorimetric detection, with especially very strong performance using ROX-labeled reporters. We optimized the reaction concentrations of these ssDNA-FQ reporters for naked-eye read-out of assay results (no transducing component required for visualization). Subsequently, we developed a convolutional neural network algorithm standardize and to automate the analytical colorimetric assessment of images and integrated this into the MagicEye mobile phone software. A field-deployable assay platform named RApid VIsual CRISPR (RAVI-CRISPR) based on a ROX-labeled reporter with isothermal amplification and CRISPR/Cas12a targeting was established. We deployed RAVI-CRISPR in a single tube towards an instrument-less colorimetric POCT format that requires only a portable rechargeable hand warmer for incubation. The RAVI-CRISPR was successfully used for the single-copy detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and African swine fever virus (ASFV). Our study demonstrates this novel RAVI-CRISPR system for distinguishing different pathogenic nucleic acid targets with high specificity and sensitivity as the simplest-to-date platform for rapid pen-side testing.

The R&D of bio-degumming technology is under a slow progress due to the shortage of proper efficient bacterial strains and processes. A degumming bacterial strain—Pectobacterium wasabiae (PW)—with broad-spectrum degumming abilities was screened out in this study. After the fermentation for 12 h, the residual gum contents of kenaf bast, ramie bast, hemp bast, flax bast, and Apocynum venetum bast were all lower than 15%. This bacterial strain could realize the simultaneous extracellular secretion of pectinase, mannase, and xylanase with the maximum enzyme activity levels of 130.25, 157.58, and 115.24 IU/mL, respectively. The optimal degumming conditions of this bacterial strain were as follows: degumming time of 12 h, bath ratio of 1:10, temperature of 33 ℃, and inoculum size of 2%. After the bio-degumming through this bacterial strain, the COD in wastewater was below 4,000 mg/L, which was over 60% lower than that in boiling-off wastewater generated by chemical degumming. This technology achieves higher efficiency, higher quality, and lower pollution.

Zero-shot learning has received great interest in visual recognition community. It aims to classify new unobserved classes based on the model learned from observed classes. Most zero-shot learning methods require pre-provided semantic attributes as the mid-level information to discover the intrinsic relationship between observed and unobserved categories. However, it is impractical to annotate the enriched label information of the observed objects in real-world applications, which would extremely hurt the performance of zero-shot learning with limited labeled seen data. To overcome this obstacle, we develop a Low-rank Semantics Grouping (LSG) method for zero-shot learning in a semi-supervised fashion, which attempts to jointly uncover the intrinsic relationship across visual and semantic information and recover the missing label information from seen classes. Specifically, the visual-semantic encoder is utilized as projection model, low-rank semantic grouping scheme is explored to capture the intrinsic attributes correlations and a Laplacian graph is constructed from the visual features to guide the label propagation from labeled instances to unlabeled ones. Experiments have been conducted on several standard zero-shot learning benchmarks, which demonstrate the efficiency of the proposed method by comparing with state-of-the-art methods. Our model is robust to different levels of missing label settings. Also visualized results prove that the LSG can distinguish the test unseen classes more discriminative.

African swine fever (ASF) is one of the most severe diseases of pigs. In this study, a CRISPR-Cas12a (also known as Cpf1) system coupled with nucleic acid amplification was optimized for the detection of ASF virus (ASFV). Two novel single-stranded DNA-fluorophore-quencher (ssDNA-FQ) reporters were developed to increase the brightness of the fluorescent signal for the visualization of nucleic acid detection. The CRISPR-Cas12a system was used to simultaneously cleave the polymerase chain reaction (PCR) or loop-mediated isothermal amplification (LAMP) amplicons and the newly developed ssDNA-FQ reporter, resulting in fluorescence that could be easily detected in multiple platforms, especially on cheap and portable blue or UV light transilluminators. This specific cleavage with fluorescence reveals the presence of the amplicon and confirms its identity, thereby preventing false-positive test results from nonspecific amplicons. This method is also uninterfered by the presence of large amounts of irrelevant background DNA and displays no cross-reactivity with other porcine DNA or RNA viruses. When coupled with LAMP, the Cas12a platform can detect a plasmid containing p72 with as few as 2 copies/μL reaction. Our results indicate that the CRISPR-Cas12a enhanced fluorescence assay coupled with nucleic acid amplification is robust, convenient, specific, confirmatory, affordable, and potentially adaptable for ASF diagnosis.

This paper presents a projection based iterative method (PIM) for solving the L1-minimization problem with its application to sparse representation and reconstruction. First, the unconstrained basis pursuit denoising (BPDN) problem is transformed into the cross-and-bouquet (CAB) form with a variable λ, and an iterative algorithm is proposed based on the projection method with the gradient of ‖x‖1 being transformed into a piecewise-linear function, which enhances the convergence of the algorithm. The global convergence of the algorithm is proved by Lyapunov method. Then, experiments conducted on random Gaussian sparse signals reconstruction and five well-known face data sets present the effectiveness and robustness of the proposed algorithm. It is also shown that the algorithm is robust to different sparsity levels and amplitude of signals, and has higher convergence rate and recognition accuracy compared with other L1-minimization algorithms especially in the case of noise interference.

This paper deals with sparse signal reconstruction by designing a discrete-time projection neural network. Sparse signal reconstruction can be converted into an $L_{1}$ -minimization problem, which can also be changed into the unconstrained basis pursuit denoising problem. To solve the $L_{1}$ -minimization problem, an iterative algorithm is proposed based on the discrete-time projection neural network, and the global convergence of the algorithm is analyzed by using Lyapunov method. Experiments on sparse signal reconstruction and several popular face data sets are organized to illustrate the effectiveness and performance of the proposed algorithm. The experimental results show that the proposed algorithm is not only robust to different levels of sparsity and amplitude of signals and the noise pixels but also insensitive to the diverse values of scalar weight. Moreover, the value of the step size of the proposed algorithm is close to 1/2, thus a fast convergence rate is potentially possible. Furthermore, the proposed algorithm achieves better classification performance compared with some other algorithms for face recognition.

In this paper, a projection-base algorithm is proposed for solving the constrained L 1 -minimization problem. Furthermore, the algorithm is utilized to sparse signal reconstruction. The L 1 -minimization is first converted into some equations which are described by the projections onto a hyer box set and the nonnegative quadrant. Then a iterative algorithm is proposed for solving the L 1 -minimization problem. Next, the algorithm is applied to sparse signal reconstruction described as an L 1 - minimization problem subject to L ∞ -norm noise constraint, or equivalently bound constraint. Finally, several experiments are presented to show the performance of the proposed algorithm.

In this paper, a novel particle swarm optimizer is developed by introducing projection operators described by projection matrices into the algorithm. Under the projection operators, the particles will oscillate along the directions determined by the projection operators to enhance global explorations. At the same time, the particles explore locally the optimal solutions when they are close to the global best solution. Moreover, two oscillation factors are introduced into the standard particle swarm optimizer to depict the particles' oscillation amplitudes. Simulations on several benchmark functions and system approximation illustrate the significant and effective impact of the two projection parts on the particle swarm optimizer.

This paper presents a weighted elastic net constrained sparse coding method for solving the sparse representation (SR) based image classification problem. First, the original model is transformed into the unconstrained basis pursuit denoising (BPDN) problem under the assumption that the residual error of the SR follows Gaussian distribution. The probability density function (PDF) of the residual error is approximated by the first order Taylor expansion and a weighted diagonal matrix is calculated according to the Taylor expansion formulation, which aims to eliminate the sparse representation error. Second, the weighted BPDN problem is solved by dual augmented Lagrangian method (DALM). The consistency and convergence of the proposed method are analysed. Finally, experiments conducted on three well-known face data sets present the effectiveness of the proposed method compared with some other algorithms for face recognition, especially with noisy and disguise situations.

Long-term variations of the sea surface salinity (SSS), air temperature (AT) and sea surface temperature (SST) of the Bohai Sea during 1960–1997 were analyzed. They all showed positive trends. The trends of the annual mean SSS, AT and SST of the Bohai Sea were, respectively, 0.074 y −1 , 0.024°C y −1 and 0.011°C y −1 . The increases of AT and SST were consistent with, the recent warming in northern China, in the Huanghai Sea (Yellow Sea) and in the East China Sea. The rise of SSS can be attributed to the rapid reduction of the total river discharge into the Bohai Sea, as well as to the increase inflow of high salinity water from the Huanghai Sea. It may also be attributed to increasing human use of river water and increases in evaporation from the sea surface. These changes in the marine environment seemed to have important influence on the Bohai Sea ecosystem.