Synthetic Data Research Articles

Ground-truth-free Deep Learning for 3-D Seismic Denoising and Reconstruction with Channel Attention Mechanism

Seismic denoising methods using supervised methods rely on a large number of high-quality paired training datasets to reach satisfactory performances. There are two ways to generate labels for network training: one is to simulate the synthetic data using the wave equation, and the other is to utilize denoised data obtained via conventional methods. However, using these labels will limit the networks' noise attenuation performance compared with using large volumes of noise-free data as labels. Here, we propose a ground-truth-free way for three-dimensional (3-D) seismic data processing. First, we use the 3-D patch scheme to divide the noisy seismic data into many fixed-size blocks and then flatten the obtained 3-D patches to expand the training set and capture more higher-order waveform characteristics from the input noisy data. Next, the obtained training dataset is sent into the proposed deep learning (DL) network, where the encoder blocks compress the feature map to extract the waveform features, and the decoder blocks reconstruct the denoised feature map. Notably, the convolutional bottleneck attention module (CBAM) and efficient channel attention (ECA) module are applied to guide the network to focus on signal fluctuation features with fewer network parameters. In addition, the concatenation mechanism is used to enable deep networks to reuse shallow-layer waveform features and mitigate overfitting during training. Finally, the unpatching scheme is used to reconstruct the denoised 3-D seismic data. Numerical experiments demonstrate that the proposed method outperforms benchmark approaches in terms of signal-to-noise ratio (SNR) improvement and useful signal preservation.

Summary Augmenter: A Text Augmentation Framework to Improve Summarization Quality

Data augmentation in Natural Language Processing (NLP) faces various challenges that hinder its widespread adoption, unlike its ever-present usage in the field of vision. It is even more the case for the text summarization task where one should focus on both article and summary. In this paper, we review the effect of back translation augmentation, present the diverse beam search decoding strategy, and masking as a method to generate synthetic data for text summarization. The approaches will be evaluated by ROUGE score, novelty, summary length, and GPT-4 to analyze their effectiveness. Our proposed framework presents multiple combinations of back translation and masking for articles, along with diverse augmentation for summaries. Although applicable to networks of any size, we decided to use BART-large, a relatively smaller model, in order to conduct a larger number of experiments. The experiments demonstrated superior performance across all specified metrics when compared to fine-tuning BART-large on the CNN/Dailymail dataset. Specifically, we showed a significant improvement in novelty; 158% and 56% increase rate for bigrams and unigrams, respectively. It could eliminate some copyright concerns around generating content similar to human writing. Additionally, the GPT-4 assessment indicates that models trained using the augmentation technique tend to capture important information more effectively than the baseline model.

Enhancing spatial modeling and risk mapping of six air pollutants using synthetic data integration with convolutional neural networks

Air pollution poses significant risks to human health and the environment, necessitating effective air quality management strategies. This study presents a novel approach to air quality management by integrating an autoencoder (AE) with a convolutional neural network (CNN) algorithm in Tehran city of Iran. One of the primary and vital problems in deep learning is model complexity, and the complexity of a model is affected by data distribution, data complexity, and information volume. AE provide a helpful way to denoise input data and make building deep learning models much more efficient. The proposed methodology enables spatial modeling and risk mapping of six air pollutants, namely, particulate matter 2.5 (PM2.5), particulate matter 10 (PM10), sulfur dioxide (SO2), nitrogen dioxide (NO2), ozone (O3), and carbon monoxide (CO). For air pollution modelling, data from a spatial database containing the annual average of six pollutants from 2012 to 2022 was utilized. The model considered various parameters influencing air pollution: altitude, humidity, distance to industrial areas, NDVI (normalized difference vegetation index), population density, rainfall, distance to the street, temperature, traffic volume, wind direction, and wind speed. The risk map accuracy was assessed using the area under the receiver operating characteristic (ROC) curve for six pollutants. Among them, NO2, PM10, CO, PM2.5, O3, and SO2 exhibited the highest accuracy with values of 0.964, 0.95, 0.896, 0.878, 0.877, and 0.811, respectively, in the risk map generated by the CNN-AE model. The findings demonstrated the CNN-AE model’s impressive precision when generating the pollution risk map.

Ground-truth-free Deep Learning for 3-D Seismic Denoising and Reconstruction with Channel Attention Mechanism

Seismic denoising methods using supervised methods rely on a large number of high-quality paired training datasets to reach satisfactory performances. There are two ways to generate labels for network training: one is to simulate the synthetic data using the wave equation, and the other is to utilize denoised data obtained via conventional methods. However, using these labels will limit the networks' noise attenuation performance compared with using large volumes of noise-free data as labels. Here, we propose a ground-truth-free way for three-dimensional (3-D) seismic data processing. First, we use the 3-D patch scheme to divide the noisy seismic data into many fixed-size blocks and then flatten the obtained 3-D patches to expand the training set and capture more higher-order waveform characteristics from the input noisy data. Next, the obtained training dataset is sent into the proposed deep learning (DL) network, where the encoder blocks compress the feature map to extract the waveform features, and the decoder blocks reconstruct the denoised feature map. Notably, the convolutional bottleneck attention module (CBAM) and efficient channel attention (ECA) module are applied to guide the network to focus on signal fluctuation features with fewer network parameters. In addition, the concatenation mechanism is used to enable deep networks to reuse shallow-layer waveform features and mitigate overfitting during training. Finally, the unpatching scheme is used to reconstruct the denoised 3-D seismic data. Numerical experiments demonstrate that the proposed method outperforms benchmark approaches in terms of signal-to-noise ratio (SNR) improvement and useful signal preservation.

Enhancing spatial modeling and risk mapping of six air pollutants using synthetic data integration with convolutional neural networks

Air pollution poses significant risks to human health and the environment, necessitating effective air quality management strategies. This study presents a novel approach to air quality management by integrating an autoencoder (AE) with a convolutional neural network (CNN) algorithm in Tehran city of Iran. One of the primary and vital problems in deep learning is model complexity, and the complexity of a model is affected by data distribution, data complexity, and information volume. AE provide a helpful way to denoise input data and make building deep learning models much more efficient. The proposed methodology enables spatial modeling and risk mapping of six air pollutants, namely, particulate matter 2.5 (PM2.5), particulate matter 10 (PM10), sulfur dioxide (SO2), nitrogen dioxide (NO2), ozone (O3), and carbon monoxide (CO). For air pollution modelling, data from a spatial database containing the annual average of six pollutants from 2012 to 2022 was utilized. The model considered various parameters influencing air pollution: altitude, humidity, distance to industrial areas, NDVI (normalized difference vegetation index), population density, rainfall, distance to the street, temperature, traffic volume, wind direction, and wind speed. The risk map accuracy was assessed using the area under the receiver operating characteristic (ROC) curve for six pollutants. Among them, NO2, PM10, CO, PM2.5, O3, and SO2 exhibited the highest accuracy with values of 0.964, 0.95, 0.896, 0.878, 0.877, and 0.811, respectively, in the risk map generated by the CNN-AE model. The findings demonstrated the CNN-AE model’s impressive precision when generating the pollution risk map.

Electricity GANs: Generative Adversarial Networks for Electricity Price Scenario Generation

The dynamic structure of electricity markets, where uncertainties abound due to, e.g., demand variations and renewable energy intermittency, poses challenges for market participants. We propose generative adversarial networks (GANs) to generate synthetic electricity price data. This approach aims to provide comprehensive data that accurately reflect the complexities of the actual electricity market by capturing its distribution. Consequently, we would like to equip market participants with a versatile tool for successfully dealing with strategy testing, risk model validation, and decision-making enhancement. Access to high-quality synthetic electricity price data is instrumental in cultivating a resilient and adaptive marketplace, ultimately contributing to a more knowledgeable and prepared electricity market community. In order to assess the performance of various types of GANs, we performed a numerical study on Turkey’s intraday electricity market weighted average price (IDM-WAP). As a key finding, we show that GANs can effectively generate realistic synthetic electricity prices. Furthermore, we reveal that the use of complex variants of GAN algorithms does not lead to a significant improvement in synthetic data quality. However, it requires a notable increase in computational costs.

ScBoolSeq: Linking scRNA-seq statistics and Boolean dynamics.

Boolean networks are largely employed to model the qualitative dynamics of cell fate processes by describing the change of binary activation states of genes and transcription factors with time. Being able to bridge such qualitative states with quantitative measurements of gene expressions in cells, as scRNA-seq, is a cornerstone for data-driven model construction and validation. On one hand, scRNA-seq binarisation is a key step for inferring and validating Boolean models. On the other hand, the generation of synthetic scRNA-seq data from baseline Boolean models provides an important asset to benchmark inference methods. However, linking characteristics of scRNA-seq datasets, including dropout events, with Boolean states is a challenging task. We present scBoolSeq, a method for the bidirectional linking of scRNA-seq data and Boolean activation state of genes. Given a reference scRNA-seq dataset, scBoolSeq computes statistical criteria to classify the empirical gene pseudocount distributions as either unimodal, bimodal, or zero-inflated, and fit a probabilistic model of dropouts, with gene-dependent parameters. From these learnt distributions, scBoolSeq can perform both binarisation of scRNA-seq datasets, and generate synthetic scRNA-seq datasets from Boolean traces, as issued from Boolean networks, using biased sampling and dropout simulation. We present a case study demonstrating the application of scBoolSeq's binarisation scheme in data-driven model inference. Furthermore, we compare synthetic scRNA-seq data generated by scBoolSeq with BoolODE's, data for the same Boolean Network model. The comparison shows that our method better reproduces the statistics of real scRNA-seq datasets, such as the mean-variance and mean-dropout relationships while exhibiting clearly defined trajectories in two-dimensional projections of the data.

Development Of A Spatio-Temporal Acoustic Pressure Measurement Technique Utilizing Background Oriented Schlieren Technique

We developed an image and optical-based technique to measure spatio-temporal acoustic pressure fields using BOS instead of a hydrophone. Fast checker demodulation (FCD) and vector tomography (VT) were employed for displacement detection and axisymmetric tomography, respectively. In experiments, BOS and hydrophones of different diameters measured focused ultrasound, revealing BOS's higher spatial resolution and ability to capture temporal evolution at 4.554 MHz. However, BOS showed systematic errors under amplified pressure. Using synthetic data, we identified that FCD's underestimation issues could be defined by the integrated power spectra of distorted images. Thus, BOS with FCD and VT provides high-resolution measurements but requires criteria to avoid displacement underestimation.

An uncertainty-aware domain adaptive semantic segmentation framework

Semantic segmentation is significant to realize the scene understanding of autonomous driving. Due to the lack of annotated real-world data, the technology of domain adaptation is applied so that the model is trained on the synthetic data and inferred on the real data. However, this domain gap leads to aleatoric and epistemic uncertainty. These uncertainties link to the potential safety issue of autonomous driving in normal weather and adverse weather. In this study, we explore the scientific problem that has received sparse attention previously. We postulate that the Dual Attention module can mitigate the uncertainty in the task of semantic segmentation and provide some empirical study to validate it. Furthermore, the utilization of Kullback-Leibler divergence (KL divergence) helps the estimation of aleatoric uncertainty and boosts the robustness of the segmentation model. Our empirical study on the diverse datasets of semantic segmentation demonstrates the effectiveness of our method in normal and adverse weather. Our code is available at: https://github.com/liubo629/Seg-Uncertainty-dual-attention.

FlowFit3: Efficient Data Assimilation Of LPT Measurements

Lagrangian particle tracking (LPT) techniques such as Shake-The-Box provide accurate flow measurements by following the movement of tracer particles within a flow. The technique gives measured values of position, velocity and acceleration at the locations of the scattered particles as opposed to a Eulerian grid like with window based techniques such as PIV. This is advantageous for e.g spatially highly resolved flow statistics. Often it is also desirable to have access to a grid-structured representation of the data, especially for the calculation of spatial gradients of flow field measures. While a simple interpolation onto a grid is possible, data assimilation techniques using physics based regularization such as FlowFit2 or VIC+ and VIC#are frequently used nowadays. Through knowledge of the Navier-Stokes equations a higher spatial resolution can be obtained than from interpolation alone. We present the new and highly efficient FlowFit3 approach and its evaluation on synthetic data.

High-Resolution Two-Phase Velocimetry Of Aspherical Particles Using Wavelet-Based Optical Flow Velocimetry (WOFV) Benchmarked With Fully Resolved Direct Numerical Simulations

Several techniques exist to measure the carrier and dispersed phases in multi-phase flows. Particle image velocimetry (PIV) analyzes the carrier phase by cross-correlating particle images, but its resolution is limited by the size of the interrogation window, making it difficult to resolve fine-scale turbulent structures. Particle tracking algorithms capture translational motion of dispersed particles well, but struggle with rotational motion, especially for irregular and aspherical particles. Currently, there is no method that universally measures both the carrier phase velocity field and the translational and rotational motions of dispersed particles. This study evaluates wavelet-based optical flow velocimetry (wOFV) for motion estimation in multi-phase flows, focusing on dispersed ellipsoidal particles and their surrounding turbulent carrier flow using synthetic image data. The research proceeds in two phases: first, the rigid motion of ellipses, including translational and rotational components, is analyzed using wOFV-generated dense motion fields. The results highlight the critical role of the regularization weighting parameter λ. Higher λ values improve translational motion estimation, while an optimal λ avoids under-regularization and non-physical structures in rotational motion. wOFV maintains accuracy in combined motion scenarios with optimal λ values. In the second phase, wOFV captures the turbulent carrier flow around an aspherical particle, benchmarked against DNS data and compared to PIV. wOFV outperforms PIV in resolving finer structures near the particle surface and in accurately representing the wake region. Error analyses confirm wOFV’s superior performance, with optimal results within a specific λ range. In conclusion, wOFV is a highly effective tool for the analysis of multi-phase flow dynamics, providing greater resolution and accuracy than PIV, especially in complex scenarios involving aspherical particles.

Benchmarking Data Assimilation Algorithms For 3D Lagrangian Particle Tracking

This paper reports a comparison of three state-of-the-art data assimilation (DA) algorithms for 3D Lagrangian particle tracking (LPT): Vortex-In-Cell sharp (VIC#), FlowFit3, and neural-implicit particle advection (NIPA). Particle trajectories, termed “tracks,” are spatially sparse, and a reconstruction algorithm is commonly employed to estimate dense Eulerian fields using position, velocity, and/or acceleration data from the tracks. DA algorithms for LPT combine the tracks with a set of governing equations (or constraints) to enhance the accuracy of the velocity field, relative to interpolation methods, and infer additional quantities like pressure. We compare the performance of VIC#, FlowFit3, and NIPA with respect to their accuracy and computational cost. Accuracy is assessed in terms of the mean inter-particle distance, frame rate, and magnitude of tracking errors, and costs are reported in wall time. Synthetic and experimental data sets for homogeneous isotropic turbulence and turbulent boundary layer flows are evaluated; direct numerical simulations are used for the synthetic tests, and realistic localization errors are added to the simulated tracks. All three DA methods perform well in cases with the highest spatio-temporal sampling of the flow, and all three methods are relatively robust to the frame rate and magnitude of localization errors. NIPA is more resilient to sparse seeding than FlowFit or VIC#, which exhibit similar performance, but NIPA is also the most expensive technique by some margin. DA reconstructions are superior to interpolation across the board, with a reduction of error up to 80% in the experimental demonstration.

Depth from Defocus Technique For High Number Density Particle Images

The Depth from Defocus (DFD) imaging technique is used to measure the size and number concentration of particles in dispersed two-phase flows, but until now it has primarily been applied to low concentration particle images. This study explores how the technique can be extended to handle overlapping images caused by neighboring particles, significantly broadening the application scope of the DFD technique and enabling measurements at higher particle number/volume concentrations. The processing algorithms are experimentally validated using a dedicated apparatus that can systematically vary particle size, shape, and degree of image overlap. Additionally, this study explores the use of Convolutional Neural Networks (CNN) for this task, comparing these results with those obtained using conventional analyses in terms of accuracy, tolerable concentration limits, and computational speed. This approach requires a large teaching dataset of images, which is only practical and feasible if the dataset can be synthetically generated. An image generation procedure for out-of-focus neighboring spherical particles, resulting in a known blurred image overlap, is therefore first developed. This procedure is validated using laboratory images with known particle size distribution, position, and image overlap before creating the teaching dataset. The trained processing scheme is then applied to both synthetic datasets and experimental data, allowing the evaluation of the technique’s limits in terms of image overlap and tolerable volume concentration, as a function of particle size distribution.

High-Resolution Two-Phase Velocimetry Of Aspherical Particles Using Wavelet-Based Optical Flow Velocimetry (WOFV) Benchmarked With Fully Resolved Direct Numerical Simulations

Several techniques exist to measure the carrier and dispersed phases in multi-phase flows. Particle image velocimetry (PIV) analyzes the carrier phase by cross-correlating particle images, but its resolution is limited by the size of the interrogation window, making it difficult to resolve fine-scale turbulent structures. Particle tracking algorithms capture translational motion of dispersed particles well, but struggle with rotational motion, especially for irregular and aspherical particles. Currently, there is no method that universally measures both the carrier phase velocity field and the translational and rotational motions of dispersed particles. This study evaluates wavelet-based optical flow velocimetry (wOFV) for motion estimation in multi-phase flows, focusing on dispersed ellipsoidal particles and their surrounding turbulent carrier flow using synthetic image data. The research proceeds in two phases: first, the rigid motion of ellipses, including translational and rotational components, is analyzed using wOFV-generated dense motion fields. The results highlight the critical role of the regularization weighting parameter λ. Higher λ values improve translational motion estimation, while an optimal λ avoids under-regularization and non-physical structures in rotational motion. wOFV maintains accuracy in combined motion scenarios with optimal λ values. In the second phase, wOFV captures the turbulent carrier flow around an aspherical particle, benchmarked against DNS data and compared to PIV. wOFV outperforms PIV in resolving finer structures near the particle surface and in accurately representing the wake region. Error analyses confirm wOFV’s superior performance, with optimal results within a specific λ range. In conclusion, wOFV is a highly effective tool for the analysis of multi-phase flow dynamics, providing greater resolution and accuracy than PIV, especially in complex scenarios involving aspherical particles.

A Single Camera Calibration-Free Approach For Dispersion Size Measurement Using Depth From Defocus

The Depth from Defocus (DFD) imaging technique is used to measure the size and number concentration of particles in dispersed two-phase flows, but until now it has primarily been applied to low concentration particle images. This study explores how the technique can be extended to handle overlapping images caused by neighboring particles, significantly broadening the application scope of the DFD technique and enabling measurements at higher particle number/volume concentrations. The processing algorithms are experimentally validated using a dedicated apparatus that can systematically vary particle size, shape, and degree of image overlap. Additionally, this study explores the use of Convolutional Neural Networks (CNN) for this task, comparing these results with those obtained using conventional analyses in terms of accuracy, tolerable concentration limits, and computational speed. This approach requires a large teaching dataset of images, which is only practical and feasible if the dataset can be synthetically generated. An image generation procedure for out-of-focus neighboring spherical particles, resulting in a known blurred image overlap, is therefore first developed. This procedure is validated using laboratory images with known particle size distribution, position, and image overlap before creating the teaching dataset. The trained processing scheme is then applied to both synthetic datasets and experimental data, allowing the evaluation of the technique’s limits in terms of image overlap and tolerable volume concentration, as a function of particle size distribution.

Improving prediction of blood cancer using leukemia microarray gene data and Chi2 features with weighted convolutional neural network

Blood cancer has emerged as a growing concern over the past decade, necessitating early diagnosis for timely and effective treatment. The present diagnostic method, which involves a battery of tests and medical experts, is costly and time-consuming. For this reason, it is crucial to establish an automated diagnostic system for accurate predictions. A particular field of focus in medical research is the use of machine learning and leukemia microarray gene data for blood cancer diagnosis. Even with a great deal of research, more improvements are needed to reach the appropriate levels of accuracy and efficacy. This work presents a supervised machine-learning algorithm for blood cancer prediction. This work makes use of the 22,283-gene leukemia microarray gene data. Chi-squared (Chi2) feature selection methods and the synthetic minority oversampling technique (SMOTE)-Tomek resampling is used to overcome issues with imbalanced and high-dimensional datasets. To balance the dataset for each target class, SMOTE-Tomek creates synthetic data, and Chi2 chooses the most important features to train the learning models from 22,283 genes. A novel weighted convolutional neural network (CNN) model is proposed for classification, utilizing the support of three separate CNN models. To determine the importance of the proposed approach, extensive experiments are carried out on the datasets, including a performance comparison with the most advanced techniques. Weighted CNN demonstrates superior performance over other models when coupled with SMOTE-Tomek and Chi2 techniques, achieving a remarkable 99.9% accuracy. Results from k-fold cross-validation further affirm the supremacy of the proposed model.

Improving prediction of blood cancer using leukemia microarray gene data and Chi2 features with weighted convolutional neural network

Blood cancer has emerged as a growing concern over the past decade, necessitating early diagnosis for timely and effective treatment. The present diagnostic method, which involves a battery of tests and medical experts, is costly and time-consuming. For this reason, it is crucial to establish an automated diagnostic system for accurate predictions. A particular field of focus in medical research is the use of machine learning and leukemia microarray gene data for blood cancer diagnosis. Even with a great deal of research, more improvements are needed to reach the appropriate levels of accuracy and efficacy. This work presents a supervised machine-learning algorithm for blood cancer prediction. This work makes use of the 22,283-gene leukemia microarray gene data. Chi-squared (Chi2) feature selection methods and the synthetic minority oversampling technique (SMOTE)-Tomek resampling is used to overcome issues with imbalanced and high-dimensional datasets. To balance the dataset for each target class, SMOTE-Tomek creates synthetic data, and Chi2 chooses the most important features to train the learning models from 22,283 genes. A novel weighted convolutional neural network (CNN) model is proposed for classification, utilizing the support of three separate CNN models. To determine the importance of the proposed approach, extensive experiments are carried out on the datasets, including a performance comparison with the most advanced techniques. Weighted CNN demonstrates superior performance over other models when coupled with SMOTE-Tomek and Chi2 techniques, achieving a remarkable 99.9% accuracy. Results from k-fold cross-validation further affirm the supremacy of the proposed model.

Graph-based semi-supervised learning with non-convex graph total variation regularization

Graph total variation (GTV) is a widely employed regularization in graph-based semi-supervised learning (GSSL), which enforce the piece-wise smoothness of the label values concerning the underlying graph structure. However, due to its inherently biased estimation, GTV regularization tends to cause an underestimation of label values on boundary nodes . In this paper, we propose a novel GSSL model with non-convex GTV regularization. Specifically, we construct the non-convex GTV regularization by subtracting the generalized Moreau envelope from the original GTV regularization term, which can reduce the estimation bias and thus effectively estimate the label values of the boundary nodes. We demonstrate that under certain conditions of the non-convex control parameter, our proposed GSSL model maintains global convexity, thereby providing a theoretical guarantee for algorithm convergence. Additionally, we present an efficient alternating direction multiplier method (ADMM) to solve the proposed model. Finally, we validate the effectiveness on both synthetic data and background subtraction. The proposed GSSL model outperforms state-of-the-art model with a 17.97% improvement in SNR value on synthetic data, and increases average recall, average precision, and average F-measure by 4.92%, 1.19%, and 1.57% respectively on the pan–tilt-zoom(PTZ) challenge of background subtraction.

Ethics-aware face recognition aided by synthetic face images

Current face recognition models trained on large-scale face datasets have achieved promising performance. However, using face images to train a face recognition model without consent would lead to severe privacy and ethical issues. Moreover, existing face recognition models also exhibit uneven performance on different races, thus perplexing vulnerable populations. To address the aforementioned two issues, this work investigates an ethics-aware face recognition method and examines whether we can leverage synthesized faces to achieve a high-accuracy racial balanced recognition model. In a nutshell, we introduce a race-controllable and identity-innumerable face synthesis approach to generate synthetic face images, and then employ the synthesized images to improve face recognition accuracy and mitigate recognition imbalance among different races despite the scarcity of consenting images (less than 100 individuals). More importantly, the synthetic data enable us to analyze the potential impacts of races on face recognition models quantitatively and facilitate the eradication of racial imbalance in face recognition. Extensive experiments demonstrate that employing our synthetic face data improves face recognition accuracy by a large margin while mitigating the recognition imbalance across different race groups.

Multi-Scale and Irregularly Distributed Circular Hole Feature Extraction from Engine Cylinder Point Clouds

The circular hole structures on automotive engines possess stringent mechanical processing requirements, so it is of vital importance to perform quality inspections on all manufactured circular hole structures. The detection of circular holes on automotive engines presents a significant challenge due to their numerous, multi-scale, and irregular distribution. Additionally, the data pertaining to circular holes is often incomplete, further complicating the detection process. In this paper, we proposed a multi-scale and irregularly distributed circular hole detection method for engine cylinder blocks, which enables the efficient extraction of all hole feature points within the engine, thereby facilitating quality inspection. First, the utilization of compartmentalization analysis techniques enhances the perceptual capacity for internal hole features from various angles. Second, by employing curvature center contractility method, hole-wall points are contracted towards their circular center positions, further enhancing the identification accuracy of small holes and holes with missing data. The proposed method is tested on both synthetic data and raw data, and compared with existing extraction and circular hole fitting methods. The experiment results demonstrate that compared to other methods, our method achieves the best feature point detection accuracy and hole primitive parameter calculation accuracy. Notably, even in special situations such as those with insufficient hole points and rounded structures, our method maintains exceptional discriminative capability and stability.