The taste and quality of kiwifruit are key factors affecting consumers' purchase intention and satisfaction. As an important indicator for measuring kiwifruit quality, sugar content is crucial for quality grading. Accurate and rapid kiwifruit grading based on sugar content is of great significance for ensuring product quality and enhancing market competitiveness. Traditional grading methods mostly adopt destructive sampling, which are cumbersome, low in efficiency, and difficult to meet the needs of modern large-scale production. Therefore, this paper proposes a kiwifruit classification method based on the Hierarchical 3D Convolution and Attention Mechanism Network (H3DAMNet). This method performs 3D convolution operations on multiple dimensions of hyperspectral data blocks simultaneously to deeply extract spatial-spectral features. It assigns weights to each channel through the channel attention mechanism to weaken attention to irrelevant information, and introduces the bottleneck self-attention mechanism to capture the positional dependence in input features, thereby effectively modeling global information. Referring to industry standards, kiwifruit are classified into three grades based on sugar content: first-grade (≥14.5 °Brix), second-grade (13.5-14.5 °Brix), and third-grade (≤13.5 °Brix). On the test set containing 280 kiwifruit samples, the overall accuracy (OA) of this method reaches 97.5% and the average accuracy (AA) is 97.3%, successfully realizing the accurate classification of kiwifruit according to sugar content and setting a reference example for the classification of other similar fruits.

Objective With the rapid development of deep learning, feature matching has advanced considerably,particularly in computer vision. This progress has led to improved performance in tasks such as 3D reconstruction, motion tracking, and image registration, all of which depend heavily on accurate feature matching. Nevertheless, current techniques often face a trade-off between accuracy and computational efficiency. Some methods achieve high matching accuracy and robustness but suffer from slow processing due to algorithmio complexity. Others offer faster processing but compromise matching accuracy, especially under challenging conditions such as dynamic scenes, low-texture environments, or large view-angle variations. The key challenge is to provide a balanced solution that ensures both accuracy and efficiency. To address this, this paper proposea a Lightweight Feature exTraction and matching Algorithm (LFTA), which integrates an additive attention mechanism within a lightweight architecture. LFTA enhances the robustness and accuracy of feature matching while maintaining the computational efficiency required for real-time applications. Methods LFTA utilizes a multi-scale feature extraction network designed to capture information from imagea at different levels of detail. A triple-exchange fusion attention mechanism mergea information across multiple dimensions, including spatial and channel features, allowing the network to learn more robust feature representations. This mechanism improves matching accuracy, particularly in scenarios with sparse textures or large viewpoint variations. LFTA further integrates an adaptive Gaussian kernel to dynamically generate keypoint heatmaps. The kernel adjusts according to local feature strength, enabling accurate keypoint extraction in both high-response and low-response regions. To improve keypoint precision, a dynamic Non-Maximum Suppression (NMS) strategy is applied, which adapts to varying keypoint densities across different image regions. This approach reduces redundancy and improves detection accuracy. In the final stage, LFTA employs a lightweight module with an additive Transformer attention mechanism to refine feature matching. This module strengthens feature fusion while reducing computational complexity through depthwise separable bonvolutions. These operations substantially lower parameter count and computational cost without affecting performance. Through this combination of techniques, LFTA achieves accurate pixel-level matching with fast inference times, making it suitable for real-time applications. Results and Discussions The performance of LFTA is assessed through extensive experiments conducted on two widely used and challenging datasets: MegaDepth and ScanNet. These dataseta offer diverse scenarios for evaluating the robustness and efficiency of feature matching methods, including variations in texture, environmental complexity, and viewpoint changes. The results indicate that LFTA achieves higher accuracy and computational efficiency than conventional feature matching approaches. On the MegaDepth dataset, an AUC 20 degrees of 79.77% is attained, which is comparable to or exceeds state-of-the-art methods such as LoFTR. Notably, this level of performance is achieved while reducing inference time by approximately 70%, supporting the suitability of LFTA for practical, time-sensitive applications. When compared with other efficient methods, Including Xfeat and Alike, LFTA demonstrates superior matching accuracy with only a marginal increase in Inference time, proving ita competitive performance in both accuracy and speed. The improvement in accuracy js particularly apparent in scenarios characterized by sparse textures or large viewpoint variations, where traditional methods often fail to maintain robustness. Ablation studies confirm the contribution of each LFTA pomponent. Exclusion of the triple-exchange fusion attention mechanism results in a significant reduction in bocuracy, indicating its function in managing complex feature interactions. Similarly, both the adaptive Gaussian kernel and dynamic NMS are found to improve keypoint extraction, emphasizing their roles in enhancing overall matching precision. Conclusions The LFTA algorithm addresses the long-standing trade-off between feature extraction accuracy and computational efficiency in feature matching. By integrating the triple-exchange fusion attention mechanism, adaptive Gaussian kernels, and lightweight fine-tuning strategies, LFTA achieves high matching bocuracy in dynamic and complex environments while maintaining low computational requirements. Experimental results on the MegaDepth and ScanNet datasets demonstrate that LFTA performs well under typical feature matching conditions and showa clear advantages in more challenging scenarios, including low-texture regions and large viewpoint variations. Given its efficiency and robustness, LFTA is well suited for real-time applications such as Augmented Reality (AR), autonomous driving, and robotic vision, where fast and boourate feature matching is essential. Future work will focus on further optimizing the algorithm for high-resolution images and more complex scenes, with the potential integration of hardware acceleration to reduce computational overhead. The method could also be extended to other computer vision tasks, including image segmentation and object detection, where reliable feature matching is required.

Few-shot fine-grained image classification (FS-FGIC) is particularly challenging because a model must distinguish subtle inter-class differences from only a few labeled examples. However, many existing methods still rely on single-layer features that struggle to directly localize truly discriminative regions, and on static class prototypes estimated from very limited supports, which cannot accurately capture class characteristics and often mismatch the query distribution. To address these issues, we propose Multi-scale Prototype Reconstruction (MPRe), which couples a Multi-scale Feature Interaction Module (MFIM) that explicitly aligns and fuses multi-level features via global–context–local attention with a Prototype Reconstruction Module (PRM) that stabilizes and reconstructs query-adaptive prototypes through similarity weighting and query-guided ridge regression. Extensive experiments on CUB-200-2011 and Stanford Dogs demonstrate that MPRe consistently outperforms state-of-the-art few-shot fine-grained image classification methods.

With the rapid expansion of large data centers and storage systems, the prediction and diagnosis of disk failures has become a critical task to ensure system reliability and reduce downtime. In this study, we propose a novel deep learning model for hard disk failure prediction, which integrates multilayer convolutional blocks, a multi-head self-attention mechanism, and a hybrid TSMixer layer. The model first extracts multilevel features from hard disk operational data through multilayer convolution, and then focuses on critical time series information using the multi-head attention mechanism to capture long-term dependencies. Finally, the model is enhanced with a TSMixer layer to learn complex time series dynamics. Experimental results on publicly available datasets show that the proposed model outperforms existing methods in terms of prediction accuracy. This research offers an efficient and accurate solution for proactive disk failure management in data centers, facilitating the implementation of preventive maintenance strategies and improving overall system reliability.

In this study, we employ near infrared spectroscopy with Stacking ensemble learning to perform non destructive sugar content analysis in kiwifruit. Our research focuses on the "Yunhai No. 1" kiwifruit variety from Hubei. Using an infrared analyzer, we gathered spectral data from 280 samples, spanning 1 557 wavelengths in the 1000-10.000 cm range. and measured sugar content with a refractometer, Outliers were identified and excluded using a singular sample identification algorithm that combines Monte Carlo random sampling with a T test. The SPXY algorithm was then employed to split the data into training and testing sets in a 4:1 ratio. Data preprocessing involved multiple scattering corrections (MSC). Savitzky Golay smoothing (SG), de trending (DT), vector normalization (VN), and standard normal variable (SNV) transformations. Feature wavelengths were initially selected using uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), and interval variable iterative space shrinkage approach (IVISSA), followed by a secondary selection with the successive projections algorithm (SPA) to remove collinear variables. To address the limitations of single models in generalization, we designed an integrated learning model using the Stacking algorithm. This model incorporated Bayesian ridge regression (BRR), artial least squares regression (PLSR), su, ort vector re ression (SVR), and artificial neural networks (ANN) as base learners, with linear regression (LR) serving as the meta learner. We assessed the performance of various ensemble model combinations and analyzed the influence of base learners coefficient, Experimental results indicated that vector normalization methods. The VN-CARS PLSR model demonstrated superior performance with Ri of 0. 805 and RMSEP of 0. 498. identifying 177 feature wavelengths and reducing data volume by 88.6% compared to the original spectrum. Comparisons of different base learner combinations in the Stacking algorithm revealed that the PLS SVR ANN integrated model achieved the highest dat the stacking ensemble model offers more comprehensive modeling capabilities and superior generalization than single models, providing valuable technical support for non destructive sugar quality detection in kiwifruit.

Kiwifruit soft rot is the most serious fungal disease in the kiwifruit postharvest storage and sales process. It has a long incubation period, and it is difficult to classify it by manual screening when it does not show obvious symptoms in the early stage of infection. Therefore, hyperspectral imaging technology (470 similar to 900 nm) was used to study the early detection and identification of kiwifruit soft rot. In the experiment, 295 hyperspectral images of healthy kiwifruit and early and late kiwifruit infected with soft rot were collected, and the samples were divided into training set and test set samples according to 7: 3 by Kennard stone algorithm. Firstly, the region of interest of the samples was selected, and then the average spectrum of the region was taken as the original spectral curve of the sample. Principal component analysis (PCA), successive projections algorithm (SPA) and competitive adaptive reweighting sampling algorithm (CARS) were used to extract spectral features from original spectral curves. Secondly, non subsampled contourlet transform (NSCT) was used for band fusion of the 8 feature bands in the SPA solution process to obtain the fusion image, and then gray level co-occurrence matrix method (GLCM) was used to extract the texture features of the fusion image. Finally, the spectral features and texture features were fused, and the nearest neighbor algorithm (KNN), random forest (RF) and support vector machine (SVM) classification models were established respectively for the early classification and detection of kiwifruit soft rot. In addition, this paper also compares with the texture features extracted from principal component images or feature bands in other literatures. The main innovation of this paper is using NSCT to fuse the feature band images and then extract the texture features, which not only reduces the feature dimension and feature redundancy, but also integrates the complementary information of different band images to improve the classification accuracy. The experimental results show that SVM is the most suitable classifier for this study, and the classification results using spectral features or texture features alone are not satisfactory. However, the classification accuracy can reach 92.05% after the fusion of the two features. Most of the early samples of kiwifruit soft rot have been correctly identified, which indicated that the fusion of the two features obtained the different information of spectrum and image in hyperspectral images. It embodies the "spatial spectral unity" of hyperspectral images. In this study, a rapid and accurate non-destructive test was carried out on kiwifruit at the early stage of soft rot, which could provide some reference and guiding significance for the quality classification of kiwifruit after harvest.

Kiwifruit soft rot is highly contagious and causes serious economic loss. Therefore, early detection and elimination of soft rot are important for postharvest treatment and storage of kiwifruit. This study aims to accurately detect kiwifruit soft rot based on hyperspectral images by using a deep learning approach for image classification. A dual-branch selective attention capsule network (DBSACaps) was proposed to improve the classification accuracy. The network uses two branches to separately extract the spectral and spatial features so as to reduce their mutual interference, followed by fusion of the two features through the attention mechanism. Capsule network was used instead of convolutional neural networks to extract the features and complete the classification. Compared with existing methods, the proposed method exhibited the best classification performance on the kiwifruit soft rot dataset, with an overall accuracy of 97.08% and a 97.83% accuracy for soft rot. Our results confirm that potential soft rot of kiwifruit can be detected using hyperspectral images, which may contribute to the construction of smart agriculture.

There is a growing demand for ready-to-eat kiwifruit in the world. However, ready-to-eat kiwifruit has a rather narrow range of firmness (e.g. 10-30 N), and it remains challenging to predict this firmness in a non-destructive manner. Here, we report a strategy for non-destructive prediction of kiwifruit firmness based on Fourier transform near-infrared (FT-NIR) spectroscopy. The radial basis function (RBF) model displayed superior performance, with a coefficient of determination (R2c ) of 0.83, a cross-validation coefficient of determination (R2p) of 0.73, a root mean square error of calibration (RMSEC) of 0.58, a root mean square error of prediction (RMSEP) of 0.72, and a ratio of performance to deviation (RPD) of 1.92. To enhance the accuracy of kiwifruit firmness prediction, we optimized the FT-NIR algorithm through data preprocessing, feature selection, and dimensionality reduction. The results showed that the FD-CARS-SVR (RBF) algorithm exhibited the best performance in predicting kiwifruit firmness during the shelf life with impressive values of R2 c (0.99), R2p (0.92), RMSEC (0.15), RMSEP (0.40), and RPD (3.48). To further evaluate the applicability of the FT-NIR model, we compared the data predicted by the model and acquired from the KiwifirmTM and penetrometer GY-4. The results revealed pronounced superiority of the FT-NIR model for the firmness ranging from 10 to 40 N to replace KiwifirmTM, providing a new non-destructive model for the prediction of the firmness of ready-to-eat kiwifruit.

As a crucial component of many natural language processing tasks, extracting entities and relations transforms unstructured text information into structured data, providing essential support for constructing knowledge graphs (KGs). However, current entity relation extraction models often prioritize the extraction of richer semantic features or the optimization of relation extraction methods, overlooking the significance of positional information and subject characteristics in this task. To solve this problem, we introduce the subject position-based complex exponential embedding for entity relation extraction model (SPECE). The encoder module of this model ingeniously combines a randomly initialized dilated convolutional network with a BERT encoder. Notably, it determines the initial position of the predicted subject based on semantic cues. Furthermore, it achieves a harmonious integration of positional encoding features and textual features through the adoption of the complex exponential embedding method. The experimental outcomes on both the NYT and WebNLG datasets reveal that, when compared to other baseline models, our proposed SPECE model demonstrates significant improvements in the F1 score on both datasets. This further validates its efficacy and superiority.

提升电子信息工程专业学生分析问题和解决问题的能力,是建设课程思政的重要任务.嵌入式操作系统专业课程内核要素众多且相互耦合,导致学生较难理解这门专业课程的知识点.针对此,从内核要素、运行时序和案例分析的角度,研究嵌入式操作系统教学方法,通过教学案例和学生实验成绩验证了该教学方法有助于提升该门课程的教学效果.

The safety hazards caused by elevator faults are increasing day by day, and it is urgent to study the problem of elevator fault prediction. In this paper, a combined model elevator fault prediction method based on CNN, LSTM and self-attentive mechanism is proposed. The operating parameters and high frequency faults of elevators are statistically determined, and a large amount of real-time elevator data is collected to build a data set. The advanced features are extracted by CNN and fed to LSTM for training, and then fed to softmax classifier for classification prediction after further feature extraction by the self-attention mechanism. The experimental results show that the method can effectively predict elevator faults.

To address challenges in the non- destructive inspection and classification of kiwifruit hardness quality, we propose a classification model that incorporates hyperspectral imaging technology and a convolution neural network. This network combines the spatial feature information extracted by the Haar wavelet and the space-spectrum joint information extracted by the three-dimensional (3D) convolution kernel. In this network, the data decomposition of channel connections is executed to ensure that all features can be utilized by the model, which improves the ability of network feature learning. Experiments on the acquired hyperspectral image-based, self-made kiwifruit hardness quality dataset (named Kiwi_seed) demonstrate that the Haar wavelet transform module can significantly improve the feature extraction ability of the network. Ablation experiments reveal that the classification accuracy of the model incorporating the Haar wavelet transform module is increased by 7. 4% and reaches the optimum level at 97. 3%, which is better than the classical image classification network. The proposed classification model can be effectively used for the non-destructive inspection and classification of kiwifruit quality.

Extracting entities and relations, as a crucial part of many tasks in natural language processing, transforms the unstructured text information into structured information and provides corresponding data support for knowledge graph (KG) and knowledge vault (KV) construction. Nevertheless, the mainstream relation-extraction methods, the pipeline method and the joint method, ignore the dependency between the subject entity and the object entity. This work introduces a pre-trained BERT model and a dilated gated convolutional neural network (DGCNN) as an encoder to distinguish the long-range semantics representation from the input sequence. In addition, we propose a cross-attention neural network as a decoder to learn the importance of each subject word for each word of the input sequence. Experiments were undertaken with two extensive datasets, the New York Times Corpus (NYT) and WebNLG Corpus, and showed that our model performs significantly better than the CasRel model, outperforming the baseline by 1.9% and 0.7% absolute gain in terms of F1-score.

The purpose of grafting is to improve the ability of plants to resist soil-borne diseases and abiotic stresses. The early detection of the grafting healing state of melon is an important demand for the current industrial development of nursery plants. Based on the standard normal variable transformation savitzky Golay smoothing second derivative (SNV-SG-SD) preprocessing, this paper proposes a competitive adaptive reweighting (DIS-CARS-SPA) feature extraction algorithm fusing grafting difference information. Establishes a radial basis function support vector machine ( GS-RBF-SVM) classification model based on grid optimization, The early classification detection of melon grafting healing state based on hyperspectral imaging was realized. Firstly, hyperspectral images of grafted survival seedlings and non-survival seedlings with pumpkin as rootstock and melon as scion were collected within 1 7 days of the healing period. Nine spectral preprocessing methods, two feature extraction algorithms and five optimization algorithms, and four kernel function support vector machine (SVM) classification models were used for analysis. The results show that the best is SNV-SG-SD spectral preprocessing, DIS-CARS-SPA feature extraction and GS-RBF-SVM classification model. Further analysis using the model shows that the classification accuracy of different types of binary classification on the same day can reach more than 99% on any day within 1 similar to 7 days of the healing period. More than 90.17% of the grafted seedlings survived on different days; More than 97.03% of the grafted non-survival seedlings could be classified on different days. On different days and types of 14 classifications, it can reach 96.85% which is 0.59% higher than the cars-spa feature extraction method without fusion of grafting difference information and 3.37% higher than the method without only preprocessing feature extraction. The results show that the proposed method can not only realize the two classifications of grafted survival seedlings and non-survival seedlings on the same day but also the two classifications of the same type on different days and the multi-classification of different types on different days. In practical application, it can advance the classification time to the first day after grafting (3 4 days for naked-eye observation and 1 2 days for machine vision technology). At the same time, the third day is the difference between mutation days of grafted survival seedlings and non-survival seedlings. The state of grafted survival seedlings can be divided into three stages: weak, medium strong, and the state of non-survival seedlings can be divided into two stages: weak weaker. This conclusion can provide effective guidance for the production of grafted melon seedlings and has a certain theoretical and practical value.

With the development of the Internet of Things technology, higher requirements are put forward for the positioning accuracy of objects. This paper presents an indoor fusion positioning algorithm based on the 5th Generation Mobile Communication Technology (5G), which effectively solves two problems. The first is that fingerprint positioning is susceptible to environmental changes and results in inaccurate fingerprint matching. The second is the problem of the hyperbolic positioning algorithm based on the line-of-sight fluctuating too much in complex indoor environments. This paper uses a 5G flexible subcarrier interval of Orthogonal Frequency Division Multiplexing (OFDM) to significantly reduce the time delay error of Time Of Arrival (TOA), and an improved genetic algorithm and the weighted hyperbolic algorithm are used to estimate the optimal position coordinates. In the offline database establishment stage of fingerprint positioning, the Channel State Information-reference signal (CSI-RS) of multiple-beam sets provides high-dimensional information for subsequent training and prediction. The online stage cooperates with the improved residual network model to make predictions. Finally, the positioning information and the error distribution function generated by the two positioning processes are simultaneously used as the input of the Kalman filter to obtain the precise position coordinates. The simulation results show that in complex indoor scenes where line-of-sight propagation and non-line-of-sight propagation paths are mixed, the accuracy of this method can reach below 0.82 m. Thus, the positioning accuracy is significantly improved compared with other methods, which can meet most indoor scene positioning needs.

A residual network based on dual-channel attention mechanism is designed to address the difficulty in extracting single and segmented seedling dataset images from the channel attention mechanism feature of the SENet network, which integrates the channel attention mechanism and spatial attention. The mechanism module can obtain the channel and spatial dimension feature weights simultaneously to enhance the feature learning ability of the network. To address the problem of missing the target in the segmented sample data, a random erasure method is proposed. Experiments on the self-made plug seedling Plant _seed dataset demonstrate that the improved network ResNet34+CBAM_basic_conv , which introduces the attention mechanism module between the ResNet34 network residual module and the conv*_x module, reaches the optimal accuracy of 93. 8%. The error rate of the model classification drops after some images in the dataset are randomly erased, demonstrating the excellent performance of the proposed method.

Vegetables are one of the most essential foods in human's daily diet. They not only provide various vitamins required by the human body but also supplement nutrients, such as dietary fiber. Detecting vegetable traits is critical during the growth and development precesses. Hyperspectral imaging technology is a new type of non-destructive testing technology that combines traditional spectroscopy with machine vision technology. It can not only obtain image dimension information but also delve deeper into the spectral dimension information within vegetables and investigate the changes of vegetable traits at the same time, based on the image dimension level and spectral dimension level of vegetable images. This article reviewed the research results of hyperspectral imaging on the nondestructive detection of vegetable traits from three aspects: the internal quality detection of vegetables, nutrient element monitoring, and disease diagnosis. The future development direction is proposed combined with the existing problems.

Early diagnosis and accurate identification of apple diseases play a major role in reducing growing costs and curbing economic losses. The diagnosis and identification of apple diseases are more difficult in the natural farming environment. A large amount of background noise in complex natural environments makes apple disease features relatively inconspicuous and makes the features of different diseases less distinguishable. A single-scale feature extraction network will be more difficult to extract effective information. In order to solve this problem, this paper proposes an apple leaf classification network based on attention mechanism and multi-scale feature fusion. First, the residual unit of ResNet50 is improved by replacing the second convolution in the residual unit with a pyramidal convolution modified by using dilated convolution to obtain multi-scale fused features. Then a channel attention module is added to the residual bypass to enhance the weighting of the disease features and improve the classification accuracy. The experiments in this paper first validate the role of the attention mechanism and pyramidal convolution separately and find that both improve the model performance. Then the combination of attention mechanism and pyramidal convolution is validated, and the optimized model has stronger noise immunity and the classification accuracy on the validation set is 94.96%. The results show that the optimized model has a better classification effect and higher robustness for apple leaf disease pictures in the natural environment.

The Angle-of-Arrival (AoA) based indoor positioning technology is widely used in indoor localization. The accuracy of AoA mostly depends on the estimation of the signal. However, the traditional smoothing Multiple-Signal Classification (MUSIC) algorithm can't handle coherent signal dealing with low signal-to-noise ratio (SNR) scenarios, leading to a biased AoA measurement. We present a weighted smoothing MUSIC method to estimate the signal in low SNR scenarios. This method weights the forward and backward smoothing matrix generated by carrier smoothing, hence using an AoA-Time-of-Flight (ToF) joint estimation algorithm. The proposed method tests by experiments based on two datasets, simulation Wi-Fi Channel State Information (CSI) data, and real Wi-Fi CSI data. The result state that the proposed method effectively improves the performance of the traditional smoothing MUSIC algorithm, holding great promise against the AoA measurement of the coherent signal under complex indoor circumstances.

In view of the challenge in autonomous navigation of unmanned ships where environmental conditions are complicated, this paper proposes a global trajectory planning model with local risk collision avoidance. The model establishes MAKLINK global connectivity map from original sea area, and provides global trajectory planning strategy based on ACO algorithm, and then introduces Q-learning algorithm to realise local risk collision avoidance, thus achieving real-time trajectory planning for unmanned ships. Compared to traditional models, our proposed one reduces level of complexity in environmental modelling, without bringing path uncertainty due to the presence of reinforcement learning and also has a faster trajectory convergence rate and shorter path length. This work would bring meaningful insights to future autonomous navigation research.