Accurate Feature Extraction Research Articles

Adaptive Ramanujan package decomposition method: a novel precise health indicator of rolling bearings

ABSTRACT In this paper, an adaptive Ramanujan packet decomposition (ARPD) method based on precise envelope segmentation is proposed to improve the accuracy of weak fault feature extraction and realise early non-destructive testing and condition evaluation of rolling bearings. Firstly, the power spectral density of the original signal is envelope-processed by accurate envelope segmentation method, so as to reduce the extreme points produced by interference frequency, and then realise the optimal frequency band division and reduce the influence of noise components. Secondly, the Ramanujan spectrum signal-to-noise ratio (RSSNR) index is introduced to quantify the fault characteristic information of the corresponding component in each frequency band. The largest component of the index is selected as the optimal mode, and the current number of layers is defined as the optimal number of modes. Finally, through the idea of iterative decomposition and RSSNR index, the problem that the optimal mode number is difficult to determine accurately is eliminated, and the optimal mode number is determined adaptively. The results of simulation and experimental signal analysis show that ARPD method can accurately divide the frequency bands and has excellent weak fault feature extraction ability, so it is an effective early non-destructive testing and condition evaluation method for rolling bearings.

Interaction of Asymmetric Adaptive Network Structures and Parameter Balance in Image Feature Extraction and Recognition

To better process irregular sample images for their image feature extraction and recognition, this essay proposes asymmetric adaptive neural network (AACNN) structures, including dual structures of an adaptive image feature extraction network (AT-CNN) and adaptive image recognition network (AT-ACNN). They both comprise an Adaptive Transform (AT) module and a deep learning network, but the ACNN comprises pixel-adaptive convolutional (PAC) kernels that CNN does not have, reflecting the asymmetry of these network structures. Structural analysis and comparative testing experiments indicated that the proposed method is more appropriate and effective for dealing with irregular sample images with different sizes and views, mainly focusing on their feature extraction accuracy and image recognition efficiency. The proposed method constructs the interaction between asymmetric dual network structures, essential in improving model performance and efficiency. It specifically manifests that the PAC kernels in an ACNN resolves the problem of content-agnostic convolution in image recognition by learning image features from a pre-trained CNN. On the other hand, it improves image recognition efficiency by using feature maps extracted from the pre-trained CNN to train the classifiers in the ACNN. We also found that parameter balance was essential in adaptive neural network structure for better performance in further testing experiments. When setting the Dropout layer parameter at 0.5 and the iteration number was 32, the proposed model achieved adequate recognition accuracy and efficiency. Smaller parameters affect model performance, but more extensive parameters significantly increase computational burden and loss. Comparative testing experiments fully validated its superiority compared with traditional methods based on CNNs. Using traditional carving patterns from Anhui Province as an example, we conducted image recognition and feature graphic application under ideal parameter balance conditions and thereby demonstrated the practicality and value of the proposed method.

Using Deep Learning Techniques to Enhance Blood Cell Detection in Patients with Leukemia

Medical diagnosis plays a critical role in the early detection and treatment of diseases by examining symptoms and supporting findings through advanced laboratory testing. Early and accurate diagnosis is essential for detecting medical problems and then prescribing the most effective treatment strategies, especially in life-threatening diseases such as leukemia. Leukemia, a blood malignancy, is one of the most prevalent cancer types affecting both adults and children. It is caused by the rapid and uncontrolled growth of abnormal white blood cells in the bone marrow. This accumulation interferes with the production of normal blood cells, leading to a weakened immune deficiency, anemia, and bleeding disorders. Conventional leukemia diagnostic methods are time-consuming, manually intensive, and inefficient. This research study proposes an automatic diagnostics prediction of leukemia by analyzing blood images according to the shape of the blast cells using digital image processing and machine learning. The purpose of blood cell detection is to precisely identify and classify diverse blood cells, detecting anomalies associated with blood cancers like leukemia. This supports early diagnosis and monitoring, which leads to more effective treatments and improved results for cancer patients. To accomplish this task, we use digital image processing techniques and then apply the convolutional neural network (CNN) deep learning algorithm to blood sample images. This research employs a multi-stage methodology, including data preparation, data preprocessing, feature extraction, and then classification. While our model is built on a typical CNN architecture, we make significant advances by using preprocessing techniques and hyperparameter tuning. We have modified its layers combination to include convolutional, pooling, and fully connected layers that are optimized for image characteristics. These layers are fine-tuned for better feature extraction and classification accuracy. This study showed that blood cell detection for diagnosing acute leukemia based on images had 99% accuracy and outperformed other advanced models, including DenseNet121, ResNet-50, Incep-tionv3, MobileNet, and EfficientNet. The comprehensive analysis of the results reveals the highest accuracy of leukemia detection as compared to existing studies in the relevant literature.

A Comparative Study of Deep Learning and Transfer Learning Approaches for Image Forgery Detection

The rapid growth of online activities such as commerce, education, research, and virtual conferences has led to a greater reliance on digital images as primary information sources on social media and other platforms. The extensive use, combined with the ease of modification via image-editing software, highlights the crucial need for effective image forgery detection tools. Traditional detection methods based on handcrafted features have grown less efficient, prompting the introduction of deep learning-based approaches, many of which combine transfer learning with pre-trained models to improve detection efficiency and shorten training time. This research presents a comprehensive evaluation of image forgery detection algorithms, categorizing them as classical, deep learning, and transfer learning frameworks. The study compares deep learning with transfer learning methods, assessing their strengths in feature extraction, classification, and detection accuracy. The findings indicate that, while transfer learning models are particularly effective at feature extraction using pre-trained architectures, deep learning remains superior for classification tasks. This insight intends to help academics construct high-accuracy, efficient models for detecting various forms of forgeries. Combining pre-trained models for feature extraction and deep learning for classification is the best option for real-time digital forensics, increasing detection accuracy and processing speed.

Malaria Cell Image Classification Using Compact Deep Learning Architectures on Jetson TX2

Malaria is a significant global health issue, especially in tropical regions. Accurate and rapid diagnosis is critical for effective treatment and reducing mortality rates. Traditional diagnostic methods, like blood smear microscopy, are time-intensive and prone to error. This study introduces a deep learning approach for classifying malaria-infected cells in blood smear images using convolutional neural networks (CNNs); Six CNN models were designed and trained using a large labeled dataset of malaria cell images, both infected and uninfected, and were implemented on the Jetson TX2 board to evaluate them. The model was optimized for feature extraction and classification accuracy, achieving 97.72% accuracy, and evaluated using precision, recall, and F1-score metrics and execution time. Results indicate deep learning significantly improves diagnostic time efficiency on embedded systems. This scalable, automated solution is particularly useful in resource-limited areas without access to expert microscopic analysis. Future work will focus on clinical validation.

Single-Scene SAR Image Data Augmentation Based on SBR and GAN for Target Recognition

High-performance neural networks for synthetic aperture radar (SAR) automatic target recognition (ATR) often encounter the challenge of data scarcity. The lack of sufficient labeled SAR image datasets leads to the consideration of using simulated data to supplement the dataset. On the one hand, electromagnetic computation simulations provide high amplitude accuracy but are inefficient for large-scale datasets due to their complex computations and physical models. On the other hand, ray tracing simulations offer high geometric accuracy and computational efficiency but struggle with low amplitude correctness, hindering accurate numerical feature extraction. Furthermore, the emergence of generative adversarial networks (GANs) provides a way to generate simulated datasets, trying to balance computational efficiency with image quality. Nevertheless, the simulated SAR images generated based on random noise lack constraints, and it is also difficult to generate images that exceed the parameter conditions of the real image’s training set. Hence, it is essential to integrate physics-based simulation techniques into GANs to enhance the generalization ability of the imaging parameters. In this paper, we present the SingleScene-SAR Simulator, an efficient framework for SAR image simulation that operates under limited real SAR data. This simulator integrates rasterized shooting and bouncing rays (SBR) with cycle GAN, effectively achieving both amplitude correctness and geometric accuracy. The simulated images are appropriate for augmenting datasets in target recognition networks. Firstly, the SingleScene-SAR Simulator employs a rasterized SBR algorithm to generate radar cross section (RCS) images of target models. Secondly, a specific training pattern for cycle GAN is established to translate noisy RCS images into simulated SAR images that closely resemble real ones. Finally, these simulated images are utilized for data augmentation. Experimental results based on the constructed dataset show that with only one scene SAR image containing 30 target chips, the SingleScene-SAR Simulator can efficiently produce simulated SAR images that exhibit high similarity in both spatial and statistical distributions compared with real images. By employing simulated SAR images for data augmentation, the accuracy of target recognition networks can be consistently and significantly enhanced.

Enhancing Facial Recognition: A Comprehensive Review of Deep Learning Approaches and Future Perspectives

This paper presents a comprehensive review of the application of deep learning in facial recognition, covering fundamental aspects from neural network architectures to advanced training methods. It starts with an introduction to the basic architectures such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), including variations like Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks. The review extends to sophisticated architectures like AlexNet, GoogleNet, VGGNet, and ResNet, which have significantly pushed the boundaries of accuracy in face recognition tasks. The paper details the process of data preprocessing in face recognition, which involves critical steps such as face detection, alignment, and normalization to ensure uniformity in the input data, enhancing the accuracy of feature extraction. Various feature extraction methods are discussed, including CNN-based and Generative Adversarial Network (GAN)-based techniques, which have shown considerable promise in deriving complex facial features from raw images. Loss functions such as Euclidean distance loss, angular/cosine margin loss, and softmax loss variants are explored to understand their impact on enhancing the discriminative power of the facial recognition systems. The paper highlights the evolution from traditional models using eigenfaces and feature descriptors like Local Binary Patterns (LBP) to cutting-edge deep learning models that utilize deep identity features (DeepID) and triplet loss functions to improve recognition accuracy.

Swin-Caption: Swin Transformer-Based Image Captioning with Feature Enhancement and Multi-Stage Fusion

The objective of image captioning is to empower computers to generate human-like sentences autonomously, describing a provided image. To tackle the challenges of insufficient accuracy in image feature extraction and underutilization of visual information, we present a Swin Transformer-based model for image captioning with feature enhancement and multi-stage fusion (Swin-Caption). Initially, the Swin Transformer is employed in the capacity of an encoder for extracting images, while feature enhancement is adopted to gather additional image feature information. Subsequently, a multi-stage image and semantic fusion module is constructed to utilize the semantic information from past time steps. Lastly, a two-layer LSTM is utilized to decode semantic and image data, generating captions. The proposed model outperforms the baseline model in experimental tests and instance analysis on the public datasets Flickr8K, Flickr30K, and MS-COCO.

An efficient personalized federated learning approach in heterogeneous environments: a reinforcement learning perspective

In order to address the problem of data heterogeneity, in recent years, personalized federated learning has tailored models to individual user data to enhance model performance on clients with diverse data distributions. However, the existing personalized federated learning methods do not adequately address the problem of data heterogeneity, and lack the processing of system heterogeneity. Consequently, these issues lead to diminished training efficiency and suboptimal model performance of personalized federated learning in heterogeneous environments. In response to these challenges, we propose FedPRL, a novel approach to personalized federated learning designed specifically for heterogeneous environments. Our method tackles data heterogeneity by implementing a personalized strategy centered on local data storage, enabling the accurate extraction of features tailored to the data distribution of individual clients. This personalized approach enhances the performance of federated learning models when dealing with non-IID data. To overcome system heterogeneity, we design a client selection mechanism grounded in reinforcement learning and user quality evaluation. This mechanism optimizes the selection of clients based on data quality and training time, thereby boosting the efficiency of the training process and elevating the overall performance of personalized models. Moreover, we devise a local training method that utilizes global knowledge distillation of non-target classes, which combined with traditional federated learning can effectively address the issue of catastrophic forgetting during global model updates. This approach enhances the generalization capability of the global model and further improves the performance of personalized models. Extensive experiments on both standard and real-world datasets demonstrate that FedPRL effectively resolves the challenges of data and system heterogeneity, enhancing the efficiency and model performance of personalized federated learning methods in heterogeneous environments, and outperforming state-of-the-art methods in terms of model accuracy and training efficiency.

Improved Visual SLAM Algorithm Based on Dynamic Scenes

This work presents a novel RGB-D dynamic simultaneous localization and mapping (SLAM) method that improves accuracy, stability, and efficiency of localization while relying on deep learning in a dynamic environment, in contrast to traditional static scene-based visual SLAM methods. Based on the classic framework of traditional visual SLAM, we propose a method that replaces the traditional feature extraction method with a convolutional neural network approach, aiming to enhance the accuracy of feature extraction and localization, as well as to improve the algorithm’s ability to capture and represent the characteristics of the entire scene. Subsequently, the semantic segmentation thread was utilized in a target detection network combined with geometric methods to identify potential dynamic areas in the image and generate masks for dynamic objects. Finally, the standard deviation of the depth information of potential dynamic points was calculated to identify true dynamic feature points, to guarantee that static feature points were used for position estimation. We performed experiments based on the public datasets to validate the feasibility of the proposed algorithm. The experimental results indicate that the improved SLAM algorithm, which boasts a reduction in absolute trajectory error (ATE) by approximately 97% compared to traditional static visual SLAM and about 20% compared to traditional dynamic visual SLAM, also exhibited a 68% decrease in computation time compared to well-known dynamic visual SLAM, thereby possessing absolute advantages in both positioning accuracy and operational efficiency.

LB-LIOSAM: an improved mapping and localization method with loop detection

Purpose In large-scale environments, LIO-SAM (Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping) exhibits poor robustness due to the accumulation of errors caused by factors such as the prevalence of similar surroundings and the lack of features in certain open areas. Therefore, the purpose of this study is to optimize the loop detection module of LIO-SAM to reduce error accumulation and enhance mapping and localization performance. Design/methodology/approach Based on the LIO-SAM framework, the LinK3D (Linear Keypoints Representation for 3D LiDAR Point Cloud) feature extraction algorithm is integrated in the front end, while the BoW3D (Bag of Words for Real-Time Loop Closing in 3D LiDAR SLAM) loop detection algorithm is integrated in the back end. The features extracted by LinK3D serve as the range factors for the LiDAR, the BoW3D generates loop closure factors and these, along with inertial measurement unit (IMU) preintegration factors and global positioning system (GPS) factors, are added to the factor graph of LIO-SAM. This addition of constraints enhances the mapping and localization effects, optimizing the overall mapping and localization performance. Findings Based on the electrically controlled car, experiments were conducted in the experimental scenario proposed in this paper. Compared to LIO-SAM, the method presented in this paper significantly reduces cumulative errors. While ensuring real-time performance, it demonstrates superior mapping and localization effects. Originality/value This paper proposes and validates a method that integrates LinK3D, BoW3D and LIO-SAM, named LB-LIOSAM, which enhances the accuracy of feature extraction, optimizes the loop detection module of LIO-SAM and improves its mapping and localization performance in specific environmental scenarios.

Flame Combustion State Detection Method of Cement Rotary Furnace Based on Improved RE-DDPM and DAF-FasterNet

It is of great significance to effectively identify the flame-burning state of cement rotary kilns to optimize the calcination process and ensure the quality of cement. However, high-temperature and smoke-filled environments bring about difficulties with respect to accurate feature extraction and data acquisition. To address these challenges, this paper proposes a novel approach. First, an improved denoising diffusion probability model (RE-DDPM) is proposed. By applying a mask to the burning area and mixing it with the actual image in the denoising process, local diversity generation in the image was realized, and the problem of limited and uneven data was solved. Secondly, this article proposes the DAF-FasterNet model, which incorporates a deformable attention mechanism (DAS) and replaces the ReLU activation function with FReLU so that it can better focus on key flame features and extract finer spatial details. The RE-DDPM method exhibits faster convergence and lower FID scores, indicating that the generated images are more realistic. DAF-FasterNet achieves 98.9% training accuracy, 98.1% test accuracy, and a 22.3 ms delay, making it superior to existing methods in flame state recognition.

Accurate Feature Extraction from Historical Geologic Maps Using Open-Set Segmentation and Detection

This study presents a novel AI method for extracting polygon and point features from historical geologic maps, representing a pivotal step for assessing the mineral resources needed for energy transition. Our innovative method involves using map units in the legends as prompts for one-shot segmentation and detection in geological feature extraction. The model, integrated with a human-in-the-loop system, enables geologists to refine results efficiently, combining the power of AI with expert oversight. Tested on geologic maps annotated by USGS and DARPA for the AI4CMA DARPA Challenge, our approach achieved a median F1 score of 0.91 for polygon feature segmentation and 0.73 for point feature detection when such features had abundant annotated data, outperforming current benchmarks. By efficiently and accurately digitizing historical geologic map, our method promises to provide crucial insights for responsible policymaking and effective resource management in the global energy transition.

Abstract 4139198: A Systematic Approach to Prompting Large Language Models for Automated Feature Extraction from Cardiovascular Imaging Reports

Introduction: Cardiovascular radiology reports contain valuable diagnostic information linked to images, but the unstructured text format makes feature extraction difficult on a large scale. Large language models (LLMs) allow for feature extraction where string parsing alone is insufficient, but require careful prompting for accurate results. Hypothesis: We hypothesize that a systematic prompting approach using LLMs can expedite the extraction of features from unstructured text in transesophageal echocardiography (TEE) reports. Methods: The data consisted of 7106 intraoperative TEE reports, 600 of which were manually reviewed to obtain pre- and post-intervention ground truth values for left ventricular ejection fraction (LVEF), right ventricular systolic function (RVSF), and tricuspid regurgitation (TR). Reports are paired with an imaging study consisting of 50-200 clips. For each feature considered, 100 of the 600 labeled reports were used to engineer a prompt in Llama-2 that maximized feature extraction accuracy. Results: We found that using multiple, shorter prompts yielded higher accuracy than did fewer, longer prompts. Additionally, when imposing semantic information onto a numerical scale, prompt engineering in combination with string parsing (Figure 1) gave the best results. When evaluated on the 500 labeled reports withheld for testing, the finalized prompts had accuracies of 94.1%, 94.8%, and 91.3% for LVEF, RVSF, and TR, respectively. Using this strategy, 5000 intraoperative TEE reports were analyzed and used to train and evaluate a regression model for LVEF estimation from TEE clips (Figure 2). Conclusion: We have shown that performing prompt engineering on Llama-2 can be used to extract features from unstructured TEE reports in an accurate manner. As an extension of these methods, automated feature prediction from echocardiograms can be used to create rapid, low-cost, and accessible cardiac assessments.

Exploring EEG-Based biomarkers for improved early Alzheimer's disease detection: A feature-based approach utilizing machine learning

This paper presents a comprehensive investigation into Electroencephalogram (EEG) signal processing and analysis techniques aimed at enhancing early diagnosis methods for Alzheimer's Disease (AD). Leveraging a dataset that has EEG data of individuals diagnosed with Mild Cognitive Impairment (MCI), AD, Healthy Controls, and the study explores Preprocessing Methods and Feature Extraction Techniques, with machine learning model notably Support Vector Machines (SVM).In the preprocessing phase, a combination of high pass, lowpass, Savitzky–Golay, and median filters are applied, informed by a comprehensive review of filter comparison literature. Feature extraction encompasses three primary categories: ‘Statistical, ‘Frequency Domain’ and ‘Time Domain’. The scope of this work is to explore features in all these three domains and build SVM based model for efficient classification. In our investigation, we achieved a categorization accuracy of 92 % through the utilization of statistical features. Employing time domain features resulted in an accuracy of 87 %, while frequency domain features also yielded an 87 % accuracy rate in our study. The primary objective of this study is that it aims to enhance early AD diagnosis through advanced EEG signal processing and machine learning techniques, focusing on preprocessing methods, feature extraction, and classification accuracy.

Lightweight Vehicle Detection Based on Mamba_ViT.

Vehicle detection algorithms are essential for intelligent traffic management and autonomous driving systems. Current vehicle detection algorithms largely rely on deep learning techniques, enabling the automatic extraction of vehicle image features through convolutional neural networks (CNNs). However, in real traffic scenarios, relying only on a single feature extraction unit makes it difficult to fully understand the vehicle information in the traffic scenario, thus affecting the vehicle detection effect. To address this issue, we propose a lightweight vehicle detection algorithm based on Mamba_ViT. First, we introduce a new feature extraction architecture (Mamba_ViT) that separates shallow and deep features and processes them independently to obtain a more complete contextual representation, ensuring comprehensive and accurate feature extraction. Additionally, a multi-scale feature fusion mechanism is employed to enhance the integration of shallow and deep features, leading to the development of a vehicle detection algorithm named Mamba_ViT_YOLO. The experimental results on the UA-DETRAC dataset show that our proposed algorithm improves mAP@50 by 3.2% compared to the latest YOLOv8 algorithm, while using only 60% of the model parameters.

Research on Classification and Identification of Crack Faults in Steam Turbine Blades Based on Supervised Contrastive Learning.

Steam turbine blades may crack, break, or suffer other failures due to high temperatures, high pressures, and high-speed rotation, which seriously threatens the safety and reliability of the equipment. The signal characteristics of different fault types are slightly different, making it difficult to accurately classify the faults of rotating blades directly through vibration signals. This method combines a one-dimensional convolutional neural network (1DCNN) and a channel attention mechanism (CAM). 1DCNN can effectively extract local features of time series data, while CAM assigns different weights to each channel to highlight key features. To further enhance the efficacy of feature extraction and classification accuracy, a projection head is introduced in this paper to systematically map all sample features into a normalized space, thereby improving the model's capacity to distinguish between distinct fault types. Finally, through the optimization of a supervised contrastive learning (SCL) strategy, the model can better capture the subtle differences between different fault types. Experimental results show that the proposed method has an accuracy of 99.61%, 97.48%, and 96.22% in the classification task of multiple crack fault types at three speeds, which is significantly better than Multilayer Perceptron (MLP), Residual Network (ResNet), Momentum Contrast (MoCo), and Transformer methods.

MMS-EF: A Multi-Scale Modular Extraction Framework for Enhancing Deep Learning Models in Remote Sensing

The analysis of land cover using deep learning techniques plays a pivotal role in understanding land use dynamics, which is crucial for land management, urban planning, and cartography. However, due to the complexity of remote sensing images, deep learning models face practical challenges in the preprocessing stage, such as incomplete extraction of large-scale geographic features, loss of fine details, and misalignment issues in image stitching. To address these issues, this paper introduces the Multi-Scale Modular Extraction Framework (MMS-EF) specifically designed to enhance deep learning models in remote sensing applications. The framework incorporates three key components: (1) a multiscale overlapping segmentation module that captures comprehensive geographical information through multi-channel and multiscale processing, ensuring the integrity of large-scale features; (2) a multiscale feature fusion module that integrates local and global features, facilitating seamless image stitching and improving classification accuracy; and (3) a detail enhancement module that refines the extraction of small-scale features, enriching the semantic information of the imagery. Extensive experiments were conducted across various deep learning models, and the framework was validated on two public datasets. The results demonstrate that the proposed approach effectively mitigates the limitations of traditional preprocessing methods, significantly improving feature extraction accuracy and exhibiting strong adaptability across different datasets.

High-Precision Defect Detection in Solar Cells Using YOLOv10 Deep Learning Model

This study presents an advanced defect detection approach for solar cells using the YOLOv10 deep learning model. Leveraging a comprehensive dataset of 10,500 solar cell images annotated with 12 distinct defect types, our model integrates Compact Inverted Blocks (CIBs) and Partial Self-Attention (PSA) modules to enhance feature extraction and classification accuracy. Training on the Viking cluster with state-of-the-art GPUs, our model achieved remarkable results, including a mean Average Precision (mAP@0.5) of 98.5%. Detailed analysis of the model’s performance revealed exceptional precision and recall rates for most defect classes, notably achieving 100% accuracy in detecting black core, corner, fragment, scratch, and short circuit defects. Even for challenging defect types such as a thick line and star crack, the model maintained high performance, with accuracies of 94% and 96%, respectively. The Recall–Confidence and Precision–Recall curves further demonstrate the model’s robustness and reliability across varying confidence thresholds. This research not only advances the state of automated defect detection in photovoltaic manufacturing but also underscores the potential of YOLOv10 for real-time applications. Our findings suggest significant implications for improving the quality control process in solar cell production. Although the model demonstrates high accuracy across most defect types, certain subtle defects, such as thick lines and star cracks, remain challenging, indicating potential areas for further optimization in future work.

Deep Plug-and-Play Non-Iterative Cluster for 3D Global Feature Extraction

Efficient and accurate point cloud feature extraction is crucial for critical tasks such as 3D recognition and semantic segmentation. However, existing global feature extraction methods for 3D data often require designing different models for different input types (point clouds, voxels, and maps). This article proposes an efficient plug-and-play non-iterative clustering method (NICM) to establish a unified point cloud global feature extraction paradigm suitable for any input type to solve the above problems. The core idea of the NICM is to construct the connection between a single point and other global points based only on the cosine similarity between center points to achieve global feature extraction, which has linear complexity characteristics and can be combined with any existing feature extraction model. Additionally, to better integrate the features extracted by NICM and the original model, this article designs an adaptive feature fusion module is designed based on the gate unit, which retains similar features and effectively fuses dissimilar features based on their importance to downstream tasks. We have applied our method to downstream tasks such as point cloud recognition, part segmentation and scene segmentation. Sufficient experiments have proven that our method can provide comprehensive and robust features for the original model, and effectively improve the performance of downstream tasks.