Local and Global Feature-Aware Dual-Branch Networks for Plant Disease Recognition.

Accurate identification of plant diseases is important for ensuring the safety of agricultural production. Convolutional neural networks (CNNs) and visual transformers (VTs) can extract effective representations of images and have been widely used for the intelligent recognition of plant disease images. However, CNNs have excellent local perception with poor global perception, and VTs have excellent global perception with poor local perception. This makes it difficult to further improve the performance of both CNNs and VTs on plant disease recognition tasks. In this paper, we propose a local and global feature-aware dual-branch network, named LGNet, for the identification of plant diseases. More specifically, we first design a dual-branch structure based on CNNs and VTs to extract the local and global features. Then, an adaptive feature fusion (AFF) module is designed to fuse the local and global features, thus driving the model to dynamically perceive the weights of different features. Finally, we design a hierarchical mixed-scale unit-guided feature fusion (HMUFF) module to mine the key information in the features at different levels and fuse the differentiated information among them, thereby enhancing the model's multiscale perception capability. Subsequently, extensive experiments were conducted on the AI Challenger 2018 dataset and the self-collected corn disease (SCD) dataset. The experimental results demonstrate that our proposed LGNet achieves state-of-the-art recognition performance on both the AI Challenger 2018 dataset and the SCD dataset, with accuracies of 88.74% and 99.08%, respectively.

Learning entity-oriented representation for biomedical relation extraction.

Biomedical Relation Extraction (BioRE) aims to automatically extract semantic relations for given entity pairs and is of great significance in biomedical research. Current popular methods often utilize pretrained language models to extract semantic features from individual input instances, which frequently suffer from overlapping semantics. Overlapping semantics refers to the situation in which a sentence contains multiple entity pairs that share the same context, leading to highly similar information between these entity pairs. In this study, we propose a model for learning Entity-oriented Representation (EoR) that aims to improve the performance of the model by enhancing the discriminability between entity pairs. It contains three modules: sentence representation, entity-oriented representation, and output. The first module learns the global semantic information of the input instance; the second module focuses on extracting the semantic information of the sentence from the target entities; and the third module enhances distinguishability among entity pairs and classifies the relation type. We evaluated our approach on four BioRE tasks with eight datasets, and the experiments showed that our EoR achieved state-of-the-art performance for PPI, DDI, CPI, and DPI tasks. Further analysis demonstrated the benefits of entity-oriented semantic information in handling multiple entity pairs in the BioRE task.

GIFTdb: a useful gene database for plant fruit traits improving.

Fruit traits are critical determinants of plant fitness, resource diversity, productive and quality. Gene regulatory networks in plants play an essential role in determining fruit traits, such as fruit size, yield, firmness, aroma and other important features. Many research studies have focused on elucidating the associated signaling pathways and gene interaction mechanism to better utilize gene resources for regulating fruit traits. However, the availability of specific database of genes related to fruit traits for use by the plant research community remains limited. To address this limitation, we developed the Gene Improvements for Fruit Trait Database (GIFTdb, http://giftdb.agroda.cn). GIFTdb contains 35 365 genes, including 896 derived from the FR database 1.0, 305 derived from 30 882 articles from 2014 to 2021, 236 derived from the Universal Protein Resource (UniProt) database, and 33 928 identified through homology analysis. The database supports several aided analysis tools, including signal transduction pathways, gene ontology terms, protein-protein interactions, DNAWorks, Basic Local Alignment Search Tool (BLAST), and Protein Subcellular Localization Prediction (WoLF PSORT). To provide information about genes currently unsupported in GIFTdb, potential fruit trait-related genes can be searched based on homology with the supported genes. GIFTdb can provide valuable assistance in determining the function of fruit trait-related genes, such as MYB306-like, by conducting a straightforward search. We believe that GIFTdb will be a valuable resource for researchers working on gene function annotation and molecular breeding to improve fruit traits.

Knowledge Distillation Facilitates the Lightweight and Efficient Plant Diseases Detection Model.

Plant disease diagnosis in time can inhibit the spread of the disease and prevent a large-scale drop in production, which benefits food production. Object detection-based plant disease diagnosis methods have attracted widespread attention due to their accuracy in classifying and locating diseases. However, existing methods are still limited to single crop disease diagnosis. More importantly, the existing model has a large number of parameters, which is not conducive to deploying it to agricultural mobile devices. Nonetheless, reducing the number of model parameters tends to cause a decrease in model accuracy. To solve these problems, we propose a plant disease detection method based on knowledge distillation to achieve a lightweight and efficient diagnosis of multiple diseases across multiple crops. In detail, we design 2 strategies to build 4 different lightweight models as student models: the YOLOR-Light-v1, YOLOR-Light-v2, Mobile-YOLOR-v1, and Mobile-YOLOR-v2 models, and adopt the YOLOR model as the teacher model. We develop a multistage knowledge distillation method to improve lightweight model performance, achieving 60.4% mAP@ .5 in the PlantDoc dataset with small model parameters, outperforming existing methods. Overall, the multistage knowledge distillation technique can make the model lighter while maintaining high accuracy. Not only that, the technique can be extended to other tasks, such as image classification and image segmentation, to obtain automated plant disease diagnostic models with a wider range of lightweight applicability in smart agriculture. Our code is available at https://github.com/QDH/MSKD.

Structure enhanced deep clustering network via a weighted neighbourhood auto-encoder.

Structural deep clustering involves the use of neural networks for fusing semantic and structural representations for clustering tasks, and it has been receiving increasing attention. In some pioneering works, auto-encoder (AE)-specific representations were integrated with a graph convolutional network (GCN)-specific representation by delivering semantic information to the GCN module layer-by-layer. Although promising performance has been achieved in various applications, we observed that a vital aspect was overlooked in these works: the structural information may vanish in the learning process because of the over-smoothing problem of the GCN module, leading to non-representative features and, thus, deteriorating clustering performance. In this study, we address this issue by proposing a structure enhanced deep clustering network. The GCN-specific structural data representation is enhanced and supervised by its structural information. Specifically, the GCN-specific structural data representation is strengthened during the learning process by combining it with a structure enhanced semantic (SES) representation. A novel structure enhanced AE, named the weighted neighbourhood AE (wNAE), is employed to learn the SES representation for each data sample. Finally, we design a joint supervision strategy to uniformly guide the simultaneous learning of the wNAE and GCN modules and the clustering assignment. Experimental results for different datasets empirically validate the importance of semantic and neighbour-wise structure learning.

Study the biomechanical performance of the membranous semicircular canal based on bionic models.

A BA (bionic ampulla) was designed and fabricated using an SMPF (Symmetric electrodes Metal core PVDF Fiber) sensor, which could imitate the sensory hair cells to sense the deformation of the cupula of the BA. Based on the BA, a bionic semicircular canal with membrane semicircular canal (MBSC) and a bionic semicircular canal without membrane semicircular canal (NBSC) were designed and fabricated. The biomechanical models of the MBSC and NBSC were established. The biomechanical models were verified through the perception experiments of the MBSC and the NBSC. The results showed that the SMPF could sense the deformation of the cupula. The MBSC and NBSC could sense the angular velocity and accelerations. What's more, it was speculated that in a human body, the endolymph probably had a function of liquid mass while the membranous semicircular canal and the cupula had a function similar to a spring in the human semicircular canal.

Outcomes of the bionic semicircular canals support the "density hypothesis" and "circular hypothesis".

To date, there are three main hypotheses explaining why the human semicircular canals (HSCCs) cannot sense linear accelerations. To further study this issue, we designed a bionic ampulla (BA) instrumented with a symmetrical metal core polyvinylidene fluoride fiber as a bionic sensor, which imitates the structure and function of the human ampulla. The BA was confirmed to have a good sensing ability in experiments with a straight tube. Additionally, we designed a bionic semicircular canal model, a blocking model, and a square model. We compared the perception performance of these three models to test the "density hypothesis," the "closed loop hypothesis," and the "circular hypothesis." The outcomes of these experiments verified the "density hypothesis" and "circular hypothesis," but did not support the "closed loop hypothesis," shedding light on why the HSCC is sensitive to angular acceleration, but not to linear acceleration.

Study on the Perception Mechanism of Utricles Based on Bionic Models.

BACKGROUND: The relationship between utricle diseases and structural lesions is not very clear in the clinic due to the complexity and delicacy of the utricle structure. Therefore, it is necessary to study the perception mechanism of the utricle. METHODS: Imitating the sensory cells in the macula of the utricle, a symmetrical metal core PVDF fiber (SMPF) was designed as a bionic hair sensor to fabricate a bionic macula (BM), a bionic macula with sand (BMS) and a bionic utricle (BU). Then experiments were carried out on them. RESULTS: This indicated the SMPF sensor can sense its bending deformation, which was similar to the sensory cell. The amplitude of the output charges of the SMPF in BMS and BU were significantly improved. The SMPF, whose electrode boundary was perpendicular to the impact direction, exhibited the largest output charges. CONCLUSION: The presence of otoliths and endolymph can improve the sensing ability of the utricle. The human brain can judge the direction of head linear accelerations based on the location of the sensory cell in the macula that produces the largest nerve signals. This provides a possibility of studying utricle abnormal functions in vitro in the future.

Connecting macroscopic diffusion metrics of cardiac diffusion tensor imaging and microscopic myocardial structures based on simulation.

To investigate the relationship between microscopic myocardial structures and macroscopic measurements of diffusion tensor imaging (DTI), we proposed a cardiac DTI simulation method using the Bloch equation and the Monte Carlo random walk in a realistic myocardium model reconstructed from polarized light imaging (PLI) data of the entire human heart. To obtain a realistic simulation, with the constraints of prior knowledge pertaining to the maturational change of the myocardium structure, appropriate microstructure modeling parameters were iteratively determined by matching DTI simulations and real acquisitions of the same hearts in terms of helix angle, fractional anisotropy (FA) and mean diffusivity (MD) maps. Once a realistic simulation was obtained, we varied the extra-cellular volume (ECV) ratio, myocyte orientation heterogeneity and myocyte size, and explored the effects of microscopic changes in tissue structure on macroscopic diffusion metrics. The experimental results demonstrated the feasibility of simulating the DTI of the whole heart using PLI measurements. When varying ECV from 15% to 55%, mean FA decreased from 0.55 to 0.26, axial diffusivity increased by 0.6 µm2/ms, and radial diffusivity increased by 0.7 µm2/ms. When orientation heterogeneity was varied from 0 to 20∘, mean FA decreased from 0.4 to 0.3, axial diffusivity decreased by 0.08 µm2/ms, and radial diffusivity increased by 0.03 µm2/ms. When mean diameter of myocytes was varied from 6 µm to 10 µm, FA decreased from 0.67 to 0.46, axial and radial diffusivities increased by 0.05 and 0.2 µm2/ms, respectively.

Protein-protein interaction relation extraction based on multigranularity semantic fusion.

Extracting semantic relationships about biomedical entities in a sentence is a typical task in biomedical information extraction. Because a sentence usually contains several named entities, it is important to learn global semantics of a sentence to support relation extraction. In related works, many strategies have been proposed to encode a sentence representation relevant to considered named entities. Despite the current success, according to the characteristic of languages, semantics of words are expressed on multigranular levels which also heavily depends on local semantic of a sentence. In this paper, we propose a multigranularity semantic fusion method to support biomedical relation extraction. In this method, Transformer is adopted for embedding words of a sentence into distributed representations, which is effective to encode global semantic of a sentence. Meanwhile, a multichannel strategy is applied to encode local semantics of words, which enables the same word to have different representations in a sentence. Both global and local semantic representations are fused to enhance the discriminability of the neural network. To evaluate our method, experiments are conducted on five standard PPI corpora (AImed, BioInfer, IEPA, HPRD50, and LLL), which achieve F1-scores of 83.4%, 89.9%, 81.2%, 84.5%, and 92.5%, respectively. The results show that multigranular semantic fusion is helpful to support the protein-protein interaction relationship extraction.

A neuralized feature engineering method for entity relation extraction.

Making full use of semantic and structure information in a sentence is critical to support entity relation extraction. Neural networks use stacked neural layers to perform designated feature transformations and can automatically extract high-order abstract feature representations from raw inputs. However, because a sentence usually contains several pairs of named entities, the networks are weak when encoding semantic and structure information of a relation instance. In this paper, we propose a neuralized feature engineering approach for entity relation extraction. This approach enhances the neural network by manually designed features, which have the advantage of using prior knowledge and experience developed in feature-based models. Neuralized feature engineering encodes manually designed features into distributed representations to increase the discriminability of a neural network. Experiments show that this approach considerably improves the performance compared to that of neural networks or feature-based models alone, exceeding state-of-the-art performance by more than 8% and 16.5% in terms of F1-score on the ACE corpus and the Chinese literature text corpus, respectively.

Design and fabrication of an E-whisker using a PVDF ring.

Mammalian whiskers can perceive obstacles and airflows. In this study, an electronic whisker (E-whisker) sensor was designed and fabricated by setting a PVDF ring with symmetrical electrodes on the root of a fiber beam. Vibration displacements with different waveforms were applied at the free end of the E-whisker beam to study the relationship between the vibration displacements and the output signals. The E-whisker protrusion sensing ability was investigated by driving it to sweep through the surface of a base platform. A static E-whisker beam and a swinging E-whisker were then separately placed in a wind tunnel to detect the airflow perception of the sensor. The experimental results suggested that the E-whisker could sense the frequencies and amplitudes of displacements at its free end, the height and width of a platform or the heights of other irregular protrusions; the static E-whisker could sense the magnitude or direction of an impact airflow, while the swinging E-whisker could sense the magnitude of a constant airflow. Thus, this kind of E-whisker could perceive the environment and airflow through touch sensation and could be used as a physical model to study the principles and abilities of animal whiskers to perceive obstacles and airflows.

Accurate intravoxel incoherent motion parameter estimation using Bayesian fitting and reduced number of low b-values.

PURPOSE: Intravoxel incoherent motion (IVIM) magnetic resonance imaging is a potential noninvasive technique for the diagnosis of brain tumors. However, perfusion-related parameter mapping is a persistent problem. The purpose of this paper is to investigate the IVIM parameter mapping of brain tumors using Bayesian fitting and low b-values. METHODS: Bayesian shrinkage prior (BSP) fitting method and different low b-value distributions were used to estimate IVIM parameters (diffusion D, pseudo-diffusion D*, and perfusion fraction F). The results were compared to those obtained by least squares (LSQ) on both simulated and in vivo brain data. Relative error (RE) and reproducibility were used to evaluate the results. The differences of IVIM parameters between brain tumor and normal regions were compared and used to assess the performance of Bayesian fitting in the IVIM application of brain tumor. RESULTS: In tumor regions, the value of D* tended to be decreased when the number of low b-values was insufficient, especially with LSQ. BSP required less low b-values than LSQ for the correct estimation of perfusion parameters of brain tumors. The IVIM parameter maps of brain tumors yielded by BSP had smaller variability, lower RE, and higher reproducibility with respect to those obtained by LSQ. Obvious differences were observed between tumor and normal regions in parameters D (P < 0.05) and F (P < 0.001), especially F. BSP generated fewer outliers than LSQ, and distinguished better tumors from normal regions in parameter F. CONCLUSIONS: Intravoxel incoherent motion parameters clearly allow brain tumors to be differentiated from normal regions. Bayesian fitting yields robust IVIM parameter mapping with fewer outliers and requires less low b-values than LSQ for the parameter estimation.

Estimation of intravoxel incoherent motion parameters using low b-values.

Intravoxel incoherent motion (IVIM) imaging is a magnetic resonance imaging (MRI) technique widely used in clinical applications for various organs. However, IVIM imaging at low b-values is a persistent problem. This paper aims to investigate in a systematic and detailed manner how the number of low b-values influences the estimation of IVIM parameters. To this end, diffusion-weighted (DW) data with different low b-values were simulated to get insight into the distributions of subsequent IVIM parameters. Then, in vivo DW data with different numbers of low b-values and different number of excitations (NEX) were acquired. Finally, least-squares (LSQ) and Bayesian shrinkage prior (BSP) fitting methods were implemented to estimate IVIM parameters. The influence of the number of low b-values on IVIM parameters was analyzed in terms of relative error (RE) and structural similarity (SSIM). The results showed that the influence of the number of low b-values on IVIM parameters is variable. LSQ is more dependent on the number of low b-values than BSP, but the latter is more sensitive to noise.