Uneven Distribution Of Samples Research Articles

Attention mechanism based on deep learning for defect detection of wind turbine blade via multi-scale features

Abstract An enhanced wind turbine blade surface defect detection algorithm, CGIW-YOLOv8, has been introduced to tackle the problems of uneven distribution of defect samples, confusion between defects and background, and variations in target scales that arise during drone maintenance of wind turbine blades. This algorithm is given based on the YOLOv8 model. Initially, a data augmentation method based on geometric changes and Poisson mixing was used to enrich the dataset and address the problem of uneven sample distribution. Subsequently, the incorporation of the Coordinate Attention (CA) mechanism into the Backbone network improved the feature extraction capability in complex backgrounds. In the Neck, the Reparameterized Generalized Feature Pyramid Network (Rep-GFPN) was introduced as a path fusion strategy and multiple cross-scale connections are fused, which effectively enhances the multi-scale expression ability of the network. Finally, the original CIOU loss function was replaced with Inner-WIoU, which was created by applying the Inner-IoU loss function to the Wise-IoU loss function. It improved detection accuracy while simultaneously speeding up the model's rate of convergence. Experimental results show that the mAP of the method for defect detection reaches 92%, which is 5.5% higher than the baseline network. The detection speed is 120.5 FPS, which meets the needs of real-time detection.

Microsphere lens array embedded microfluidic chip for SERS detection with simultaneous enhancement of sensitivity and stability

Surface enhanced Raman spectroscopy (SERS) utilizes the fingerprint features of molecular vibrations to identify and detect substances. However, in traditional single focus excitation scenarios, its signal collection efficiency of the objective is restricted. Furthermore, the uneven distribution of samples on the SERS substrate would result in poor signal stability, while the excitation power is limited to avoid sample damage. SERS detection system always requires precise adjustment of focal length and spot size, making it difficult for point-of-care testing applications. Here, we proposed a SERS microfluidic chip with barium titanate microspheres array (BTMA) embedded using vacuum self-assembled hot-pressing method for SERS detection with simultaneous enhancement of sensitivity and stability. Due to photonic nano-jets and directional antenna effects, high index microspheres are perfect micro-lens for effective light focusing and signal collecting. The BTMA can not only disperse excitation beam into an array of focal points covering the target uniformly with very low signal fluctuation, but enlarge the power threshold for higher signal intensity. We conducted a proof-of-principle experiment on chip for the detection of bacteria with immuno-magnetic tags and immuno-SERS tags. Together with magnetic and ultrasonic operations, the target bacteria in the flow were evenly congregated on the focal plane of BTMA. It demonstrated a limit of detection of 5 cells/mL, excellent signal reproducibility (error∼4.84%), and excellent position tolerance of 500 μm in X–Y plane (error∼5.375%). It can be seen that BTMA-SERS microfluidic chip can effectively solve the contradiction between sensitivity and stability in SERS detection.

Transfer Learning-Based Hyperspectral Image Classification Using Residual Dense Connection Networks.

The extraction of effective classification features from high-dimensional hyperspectral images, impeded by the scarcity of labeled samples and uneven sample distribution, represents a formidable challenge within hyperspectral image classification. Traditional few-shot learning methods confront the dual dilemma of limited annotated samples and the necessity for deeper, more effective features from complex hyperspectral data, often resulting in suboptimal outcomes. The prohibitive cost of sample annotation further exacerbates the challenge, making it difficult to rely on a scant number of annotated samples for effective feature extraction. Prevailing high-accuracy algorithms require abundant annotated samples and falter in deriving deep, discriminative features from limited data, compromising classification performance for complex substances. This paper advocates for an integration of advanced spectral-spatial feature extraction with meta-transfer learning to address the classification of hyperspectral signals amidst insufficient labeled samples. Initially trained on a source domain dataset with ample labels, the model undergoes transference to a target domain with minimal samples, utilizing dense connection blocks and tree-dimensional convolutional residual connections to enhance feature extraction and maximize spatial and spectral information retrieval. This approach, validated on three diverse hyperspectral datasets-IP, UP, and Salinas-significantly surpasses existing classification algorithms and small-sample techniques in accuracy, demonstrating its applicability to high-dimensional signal classification under label constraints.

Enhancing class imbalance solutions: A projection-based fuzzy LS-TSVM approach

Class imbalance and noise present significant challenges in numerous real-world classification tasks. The prevalence of an uneven distribution of samples typically results in a bias towards the majority class in Support Vector Machine (SVM) classifiers, compounded by the often inherent noise within these samples. Addressing both class imbalance and noise, we introduce two fuzzy-based methodologies. The first method employs intuitionistic fuzzy membership, resulting in the development of the Robust Energy-based Intuitionistic Fuzzy Least Squares Twin Support Vector Machine (IF-RELSTSVM), a model specifically designed for class imbalance learning. The IF-RELSTSVM model is distinguished by its use of intuitionistic fuzzy scores for both classes, significantly attenuating the detrimental effects of noise and outliers. A distinctive attribute of IF-RELSTSVM is its proficiency in processing noisy data points, whether proximate to or distant from the hyperplane. Additionally, we introduce a novel concept of hyperplane-based fuzzy membership, calculating fuzzy memberships through a projection-based approach. This foundation supports the formulation of a Robust Energy-based Fuzzy Least Squares Twin Support Vector Machine (F-RELSTSVM), also aimed at class imbalance learning. The efficacy of the proposed IF-RELSTSVM and F-RELSTSVM algorithms is rigorously evaluated across several benchmark and synthetic datasets, employing the Area Under the ROC Curve (AUC) as a performance metric. Experimental findings indicate that these algorithms surpass baseline models in the majority of datasets tested. Statistical analyses further validate the significance of our proposed methods, demonstrating their suitability for application in environments characterized by noise and class imbalance. A case study in credit card fraud detection showcases the F-RELSTSVM algorithm achieving an impressive average AUC of 90.84%, thereby outperforming comparable algorithms and highlighting the practical applicability of our approaches in tackling challenging datasets.

Diet breadth and biodiversity in the pre-hispanic South-Central Andes (Western South America) during the Holocene: An exploratory analysis and review

This paper presents an exploratory study on the taxonomic diversity of pre-Hispanic archaeofaunas in the South-Central Andes (SCA; western South America) from the Pleistocene-Holocene boundary to the Late-Holocene. The SCA is a complex of diverse environments and has undergone distinct climate events for the last 13,000 years, such as the occurrence of warmer and drier conditions in the Middle-Holocene. The South-Central Andean area was part of the larger Andes interaction area, which was a primary center for animal and plant domestication and the emergence of agro-pastoralist economies. Since subsistence was key to these processes, the SCA provides a relevant case study on the interactions among environment, foodways and sociocultural evolution. Taxonomic diversity was used here as a proxy for diet breadth. A total of 268 archaeofaunal assemblages were sampled from the zooarchaeological literature. Reviewed variables included the cultural chronology and spatial coordinates of the assemblages, as well as the presence and abundance of taxa at the family rank. Taxonomic diversity covered two dimensions: composition (families present in each assemblage) and structure (quantitative relationships among taxa), which was measured through richness (NTAXA), ubiquity and relative abundance (NISP based rank-order). Despite the uneven distribution of samples, the analyses revealed the following trends: (1) a moderate relationship between NTAXA and distance from coastline for most of the Holocene; (2) a potential decrease in assemblage richness for coastal ecoregions during the Late-Holocene; and (3) a generalized increase in the relative abundance of Camelidae.

MALDI sample preparation methods: A mini review

Matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) has emerged as a robust analytical technique for the rapid and sensitive analysis of diverse biomolecules. MALDI-MS effectiveness relies on meticulous sample preparation, with a particular emphasis on achieving a uniform sample surface. Although the dried droplet (DD) method is widely utilized, it encounters challenges such as uneven sample distribution that impacts the ionization process and significantly varies ion signals across sample areas, resulting in poor data reproducibility. The uneven sample distribution can increase the time to finish mass analysis of samples and make automated analysis complicated. Automated sample deposition systems and alternative MALDI sample preparation methods show promise in overcoming the limitations of the DD method, resulting in improved reproducibility and mass spectral quality. These advancements are critical in improving the precision and reliability of quantitative measurements using MALDI-MS, propelling progress in MALDI imaging, quantitative analysis, and biomarker discovery across a wide range of scientific applications.

The identification of gas–liquid two-phase flow patterns in a centrifugal pump with an imbalanced sample and variable rotational speeds

Due to the effects of the working medium and external conditions, centrifugal pumps often produce complex gas–liquid two-phase flows during operation. The entrainment of gas can lead to a significant decrease in the pressure head of the centrifugal pump and a serious reduction in conveyance efficiency. Therefore, in the absence of visualization or noninvasive techniques, the question of how to predict and model the flow status inside the pump is a key issue that needs to be urgently addressed. Using experimental data for gas–liquid two-phase flows in centrifugal pumps, this paper proposes a flow pattern identification model based on one-dimensional pressure increment signals. This deep learning model consists of convolutional neural networks and bidirectional gated recurrent units. Bayesian optimization is used for automatic hyperparameter tuning, and under the premise of balanced data distribution, the model achieves an identification rate of 0.98 on the test set. Subsequently, addressing the issue of the uneven sample distribution in centrifugal pumps due to their poor gas-carrying capacity, a data augmentation model based on the conditional Wasserstein generative adversarial network with gradient penalty approach is proposed based on the predictive model. Finally, to establish a flow pattern identification method with good extrapolation capability, a transfer learning strategy that accounts for the domain adaptation problem is introduced to the dataset with variable speeds based on the previous two studies. This method achieves knowledge transfer from the source domain to the target domain with a recognition rate of over 0.988 without hyperparameter optimization.

Aspect-level sentiment classification with fused local and global context

Sentiment analysis aims to determine the sentiment orientation of a text piece (sentence or document), but many practical applications require more in-depth analysis, which makes finer-grained sentiment classification the ideal solution. Aspect-level Sentiment Classification (ALSC) is a task that identifies the emotional polarity for aspect terms in a sentence. As the mainstream Transformer framework in sentiment classification, BERT-based models apply self-attention mechanism that extracts global semantic information for a given aspect, while a certain proportion of local information is missing in the process. Although recent ALSC models have achieved good performance, they suffer from robustness issues. In addition, uneven distribution of samples greatly hurts model performance. To address these issues, we present the PConvBERT (Prompt-ConvBERT) and PConvRoBERTa (Prompt-ConvRoBERTa) models, in which local context features learned by a Local Semantic Feature Extractor (LSFE) are fused with the BERT/RoBERTa global features. To deal with the robustness problem of many deep learning models, adversarial training is applied to increase model stability. Additionally, Focal Loss is applied to alleviate the impact of unbalanced sample distribution. To fully explore the ability of the pre-training model itself, we also propose natural language prompt approaches that better solve the ALSC problem. We utilize masked vector outputs of templates for sentiment classification. Extensive experiments on public datasets demonstrate the effectiveness of our model.

Estimation of wheat kernel moisture content based on hyperspectral reflectance and satellite multispectral imagery

The wheat kernel moisture content (KMC) plays a pivotal role in determining the quality, storage viability, and economic profitability of wheat. Monitoring wheat KMC in-field before harvest provides decision-making information to ensure harvested wheat meeting trade standards. The study aims to expand the feasibility of spectroscopy to estimate wheat KMC at multi-scales. In addition to using ground hyperspectral reflectance for single-point measurement of wheat KMC, satellite multispectral images were also utilized to monitor the field distribution of wheat KMC on large-scale. For each estimation task, we extracted sensitive spectral features and compared the accuracy of three machine learning methods including Ridge, Support Vector Regression (SVR) and Random Forest Regression (RFR). We also adopted Two-stage Tradaboost.R2 to address the issue of uneven distribution of samples corresponding to satellite imagery. Experimental results showed that, based on optimized spectral features, the R2 of RFR model outperformed other machine learning models (Ridge and SVR) for wheat KMC estimation (R2 > 0.85). Among all spectral features, the B5 band and PSRI vegetation index from PlanetScope satellite, as well as the B11, and B12 bands and VARI vegetation index from Sentinel-2, proved to be effective features for wheat KMC estimation. Moreover, the incorporation of hyperspectral reflectance and multispectral imagery can improve the estimation accuracy on large-scale through the transfer learning algorithm of Two-stage Tradaboost.R2. Leveraging its superior temporal and spatial resolution, PS imagery emerges as an ideal data source for monitoring rapid fluctuations in wheat KMC during the physiological maturity period, which enabled a multi-day wheat KMC simulation. Overall, This study contributes a reliable and convenient spectroscopy tool for wheat KMC estimation for multi-scenes, offering scientific insights to support agricultural water management and wheat quality.

SPSRec: An Efficient Signal Phase Recommendation for Signalized Intersections With GAN and Decision-Tree Model

Signal phase optimization at signalized intersection is of great importance to urban traffic control and management yet is very challenging. The traditional approaches for signal phase optimization heavily rely on the traffic engineering practitioners’ experience. To tackle these challenges, a novel data-driven method is proposed to realize signal phase optimization and recommendation solely using limited amount of real signalized intersection samples. Firstly, all of discrete features related to signal phase design, encoded by one-hot representation, are sampled by the Gumbel-SoftMax distribution, which is a continuous approximation to a multinomial distribution. With this approximation distribution, the generative adversarial network (GAN) is applied to produce the most acceptable signal phase samples among all acceptable choices, dealing with the problem of insufficient samples and uneven sample distribution in real word. Thirdly, a decision-tree based classifier is established to realize signal phase recommendation automatically. We conducted extensive experiments to evaluate our proposed method on three cities in China, including Beijing, Tongxiang and Chaozhou. The experimental results showed that the proposed method could effectively improve the signalized intersection operation efficiency. Moreover, the proposed method has already been deployed in several cities, and it successfully keeps serving hundreds of signalized intersections. This confirms that SPSRec is a practical and robust solution for large-scale real-world signal control services.

Perception of Groundnuts Leaf Disease by Neural Network with Progressive Re-Sizing

India is the world's second-largest groundnut producer after Brazil. An major crop of oilseeds is groundnuts. Because of this, the crop's quality and yield have declined, which has had a detrimental effect on the agricultural economy. This is partly because the crop is more susceptible to various diseases. It is required to create more precise and reliable automated approaches to address this problem and improve the identification of groundnut leaf diseases. This article proposes a deep learning-driven approach based on a progressive scaling technique for the accurate classification and identification of groundnut leaf diseases. The five main groundnut leaf diseases that are the subject of this study are leaf spot, armyworm effect, wilts, yellow leaf, and healthy leaf. The proposed model is trained using both progressive resizing and conventional techniques, and its performance is assessed using cross-entropy loss. A fresh dataset is meticulously curated in Gujarat state, India's Saurashtra region, for training and validation. Due to the dataset's uneven sample distribution across disease categories, an extended focus loss function was used to correct this class imbalance. In order to evaluate the performance of the suggested model, a number of performance metrics are utilized, including accuracy, sensitivity, F1-score, precision, and sensitivity. Notably, the suggested model has a 96.12% success rate, which signifies a considerable increase in the disease identification accuracy. It's important to note that the model incorporating progressive resizing beats the basic neural network-based model based on cross-entropy loss, highlighting the potency of the recommended approach.

Autoencoder-Enhanced Regularized Prototypical Network for New Energy Vehicle battery fault detection

As the ownership of new energy vehicles (NEVs) is experiencing a sustained growth, the safety of NEVs has become increasingly prominent, with power battery faults emerging as the primary cause of fire accidents in NEVs. Successful detection of incipient faults can not only improve the safety and reliability but also provide optimal maintenance instructions for NEV batteries. Since NEVs operate under normal conditions most of the time, the occurrence of system faults leading to faulty samples is extremely rare, thus exhibiting the characteristics of a long-tail distribution. To resolve this problem, this paper proposes an Autoencoder-Enhanced Regularized Prototypical Network (ARPN) to overcome the limitations of class imbalance and uneven sample distribution in the data. The model consists of three key components. First, an autoencoder with an encoder and decoder structure is introduced to map the data to a low-dimensional latent space and back, effectively extracting key structural information from the original data and make full use of the limited number of faulty training samples. Secondly, a multi-layer regularized feature embedding strategy is designed to reduce the complexity of the embedding space, which makes the prototype vector get easier to be separated and distinguished and plays an important role in the proposed ARPN. Finally, a prototype-based classification method is adopted to achieve accurate classification by calculating the cosine similarity between the representation of query set samples in the embedding space and the prototype vectors of each class. Experiments on several battery datasets, obtained from the operational data of NEVs, demonstrate the effectiveness of the proposed method.

SPAW-SMOTE: Space Partitioning Adaptive Weighted Synthetic Minority Oversampling Technique For Imbalanced Data Set Learning

Abstract The problem of data imbalance is common in reality, which greatly affects the performance of classifiers. Most of the solutions are to balance the data set by generating new minority class samples, which are faced with the problems of selecting the appropriate area for generating samples, fuzzy classification boundary and uneven distribution of samples. To solve these problems, we propose a novel oversampling algorithm named space partitioning adaptive weighted synthetic minority oversampling technique (SPAW-SMOTE). We first divide the data space into boundary space and non-boundary space based on spatial partitioning techniques. The number of samples to be generated is assigned to different spaces by the designed adaptive weighting algorithm, which is used to solve the problems of uneven distribution of samples and easy to blur the classification boundary. Finally, we also endeavor to develop a new generation algorithm to reduce the probability of overlapping samples generated when synthesizing new samples and to ensure the diversity of new samples. Experimental results on 18 real-world data sets show that the average performance (G-mean, F1-measure and Area Under Curve) of SPAW-SMOTE is significantly better than other existing oversampling techniques.

Named Entity Recognition in Power Marketing Domain Based on Whole Word Masking and Dual Feature Extraction

With the aim of solving the current problems of low utilization of entity features, multiple meanings of a word, and poor recognition of specialized terms in the Chinese power marketing domain named entity recognition (PMDNER), this study proposes a Chinese power marketing named entity recognition method based on whole word masking and joint extraction of dual features. Firstly, word vectorization of the electricity text data is performed using the RoBERTa pre-training model; then, it is fed into the constructed dual feature extraction neural network (DFENN) to acquire the local and global features of text in a parallel manner and fuse them. The output of the RoBERTa layer is used as the auxiliary classification layer, the output of the DFENN layer is used as the master classification layer, and the output of the two layers is dynamically combined through the attention mechanism to weight the outputs of the two layers so as to fuse new features, which are input into the conditional random field (CRF) layer to obtain the most reasonable label sequence. A focal loss function is used in the training process to alleviate the problem of uneven sample distribution. The experimental results show that the method achieved an F1 value of 88.58% on the constructed named entity recognition dataset in the power marketing domain, which is a significant improvement in performance compared with the existing methods.

Mixed-former: multi-fusion remote sensing change detection

ABSTRACT Although remote sensing change detection methods based on deep learning have achieved relatively excellent results, they still face many challenges, such as many difficult samples and uneven sample distribution. The current CNN method is limited by the lack of effective global spatial-temporal information, while the transformer method is unable to accomplish the task of grounded application due to the complex network structure. Therefore, we propose mixedformer, a multi-fusion siamese network based on the enhancement of spatial-temporal information fusion. In this regard, the spatial-temporal symmetry module extracts valid potential contributing feature information based on a global self-attentive mechanism over the enhanced image information sequence. Because each embedding contains other embedding information, feature information can be generated at different locations to obtain denoised local feature information and global spatial-temporal information. At the same time, we argue that features in different layers, whether shallow or deep, contribute to the final detected change results with some weight. We also believe that both shallow and bottom features contribute some weight to the final detected change results. To achieve this, we use global cross-attention in the middle layer of decoding, linking valid information from deep and shallow layers, filtering noise from upsampling of the bottom features, while refining the edge information of the changes. Extensive experiments have shown that our model achieves excellent results on all three CD datasets with an appropriate amount of computation.

Deep learning method for risk identification of autonomous bus operation considering image data augmentation strategies

Objective The autonomous bus is a key application scenario for autonomous driving technology. Identifying the risk of autonomous bus operation is of great significant to improve road traffic safety and promote the large-scale application of autonomous driving technology. Methods For the purpose of risk identification, the actual operation data for autonomous buses in Shanghai were converted into 3 kinds of grayscale images and 1 kind of radar image from a temporal–spatial perspective, and a deep learning convolutional neural network, AlexNet, was applied for image recognition. This article uses several image data augmentation strategies to address the problem of uneven distribution of samples and compares the effectiveness of different strategies. Results The optimal accuracy (ACC) of the risk identification was 90.4%, the optimal true positive rate (TPR) was 83.7%, and the optimal false negative rate (FPR) was 94.58%. The accuracy of risk identification using AlexNet was higher based on the sample containing Fourier images. In addition, risk identification accuracy based on grayscale images was higher than that based on radar images. Conclusions Autonomous buses were found to be vulnerable to risks in areas such as turning sections and intersections. In addition, the results show that the Fourier transform, an image augmentation strategy, can effectively solve the uneven sample distribution problem, and the length of the input time series has an impact on the accuracy of the risk identification.

Eres-UNet++: Liver CT image segmentation based on high-efficiency channel attention and Res-UNet++

Computerized tomography (CT) is of great significance for the localization and diagnosis of liver cancer. Many scholars have recently applied deep learning methods to segment CT images of liver and liver tumors. Unlike natural images, medical image segmentation is usually more challenging due to its nature. Aiming at the problem of blurry boundaries and complex gradients of liver tumor images, a deep supervision network based on the combination of high-efficiency channel attention and Res-UNet++ (ECA residual UNet++) is proposed for liver CT image segmentation, enabling fully automated end-to-end segmentation of the network. In this paper, the UNet++ structure is selected as the baseline. The residual block feature encoder based on context awareness enhances the feature extraction ability and solves the problem of deep network degradation. The introduction of an efficient attention module combines the depth of the feature map with spatial information to alleviate the uneven sample distribution impact; Use DiceLoss to replace the cross-entropy loss function to optimize network parameters. The liver and liver tumor segmentation accuracy on the LITS dataset was 95.8% and 89.3%, respectively. The results show that compared with other algorithms, the method proposed in this paper achieves a good segmentation performance, which has specific reference significance for computer-assisted diagnosis and treatment to attain fine segmentation of liver and liver tumors.

PV-YOLO: Lightweight YOLO for Photovoltaic Panel Fault Detection

The rapid development of the photovoltaic industry in recent years has made the efficient and accurate completion of photovoltaic operation and maintenance a major focus in recent studies. The key to photovoltaic operation and maintenance is the accurate multifault identification of photovoltaic panel images collected using drones. In this paper, PV-YOLO is proposed to replace YOLOX’s backbone network, CSPDarknet53, with a transformer-based PVTv2 network to obtain local connections between images and feature maps to extract more edge-detail features of similar faults. The CBAM attention mechanism is added to enhance the effective features and improve the detection accuracy of small objects. The label assignment mechanism is optimized, and the SIoU loss functionis used to improve the uneven distribution of samples and accelerate network convergence. Experiments on the dataset prove that this method is superior to the existing technology, as the highest mAP value is 92.56%. This value is 10.46% higher than that of YOLOX, and the mAP is optimal under the same parameter magnitude,proving the model’s effectiveness.Moreover, mAP is increased by over 10%, especially for small targets. In this paper, we implemented a lightweight design for the model, and proposes four models of different sizes to be-sized models that are suitable for different detection scenarios.

Electrochemically assisted wide area Raman with standard curved surface quantification method

Reproducibility is still a great challenge for surface-enhanced Raman scattering (SERS), because the uncontrollable fabrication of SERS substrates or the uneven distribution of samples on the substrate result in the signal fluctuation with or between the substrates. Herein, a novel SERS quantitative method with good reproducibility was proposed. It is based on the basic principle that the SERS signal intensity is not only related to electromagnetic enhancement and the concentration of sample, but also related to the specific surface area of the substrate. The surface area information of the substrate is obtained through electrochemical technology, and then introduced into the standard curve with the linear relationship of concentration of sample and SERS intensity as a new variable to obtain a 3D standard curved surface, which effectively corrects the signal difference between the substrates, and combines the wide area Raman method to reduce the difference within the substrate, thereby improving the reproducibility of SERS quantitative detection. Using malachite green (MG) as the probe molecule and using cyclic voltammetry to calculate the substrate area fitted plane model (CV-standard curved surface), the root mean square error (RMSE) of the predicted result is 0.26 and the relative error (RE) is 0.25. It shows that the detection error significantly reduces comparing with the traditional standard curve method. Also, the proposed method can be used in other SERS quantitative detection and has potential application prospects.

Learning latent local manifold embeddings for hyperspectral feature extraction

Hyperspectral images with an immense number of spectral bands provide abundant discriminant information for accurate land-cover classification in the remote sensing field. However, these narrow and adjacent bands contain a large amount of redundant information. Analyzing these images always requires a huge storage space with expensive computational costs. Furthermore, their high correlation coefficient would lead to the Hughes phenomenon, hindering the improvement of classification performance. We propose a linear semi-supervised hyperspectral feature extraction method L3ME to learn latent local manifold embeddings. Although labeled samples are beneficial to construct learning models, their number is always limited in real-world tasks. The motivation of this paper is to jointly enhance the contributions of labeled and unlabeled samples for learning local manifold structures of hyperspectral images. Features in labeled samples are extracted by two procedures, the adaptive patch alignment framework and integrated intraclass-interclass relationships, from different perspectives. The former aims to solve the problem of the uneven distribution of classes by introducing spectral angle based adaptive parameters. The latter aims to solve the problem of the uneven distribution of samples by constructing several adjacency graphs. The locality preserving projection is capable of preserving the local neighborhood structure of samples. A penalty for sparse regularization is cleverly integrated into the proposed linear discriminant objective function, which is optimized using a novel updating strategy. The convergence of L3ME is proved in detail and analyzed in this paper. Experiments on three typical hyperspectral datasets illustrate the effectiveness of the proposed method over some state-of-the-art techniques. The implementation of L3ME is available at https://github.com/biowby/L3ME.