Abstract ：

The expansion of tunnel scale has led to a massive demand for inspections. Light Detection And Ranging (LiDAR) has been widely applied in tunnel structure inspections due to its fast data acquisition speed and strong environmental adaptability. However, raw tunnel point-cloud data contain noise point clouds, such as non-structural facilities, which affect the detection of tunnel lining structures. Methods such as point-cloud filtering and machine learning have been applied to tunnel point-cloud denoising, but these methods usually require a lot of manual data preprocessing. In order to directly denoise the tunnel point cloud without preprocessing, this study proposes a comprehensive processing method for cross-section fitting and point-cloud denoising in subway shield tunnels based on the Huber loss function. The proposed method is compared with classical fitting denoising methods such as the least-squares method and random sample consensus (RANSAC). This study is experimentally verified with 40 m long shield-tunnel point-cloud data. Experimental results show that the method proposed in this study can more accurately fit the geometric parameters of the tunnel lining structure and denoise the point-cloud data, achieving a better denoising effect. Meanwhile, since coordinate system transformations are required during the point-cloud denoising process to handle the data, manual rotations of the coordinate system can introduce errors. This study simultaneously combines the Huber loss function with principal component analysis (PCA) and proposes a three-dimensional spatial coordinate system transformation method for tunnel point-cloud data based on the characteristics of data distribution.

Keyword ：

principal component analysis (PCA) point-cloud denoising cross-section fitting 3D laser scanning technology Huber loss function

Cite：

Copy from the list or Export to your reference management。

Abstract ：

BackgroundUnsupervised traumatic brain injury (TBI) lesion detection aims to identify and segment abnormal regions, such as cerebral edema and hemorrhages, using only healthy training data. Recent advancements in generative models have achieved success in unsupervised anomaly detection by transforming abnormal patterns into normal counterparts. However, current mask-free image generators often fail to maintain semantic consistency of anatomical structures during the restoration process. This limitation negatively impacts residual-based anomaly detection, particularly in cases where structural deformations occur due to the mass effect of TBI lesions.PurposeThis study aims to develop a semantic-consistent, unsupervised TBI lesion detection and segmentation method that minimizes false positives by preserving normal tissue consistency during the image generation process while addressing mass effect-related tissue deformations.MethodsWe propose the semantic-consistent diffusion model (SCDM) for unsupervised TBI lesion detection, focusing on the localization and segmentation of various lesion types from noncontrast CT scans of TBI patients. Leveraging the high-quality image generation capabilities of unconditioned diffusion models (DM), we introduce a normal tissue retainment (NTR) regularization to ensure that normal tissues remain unaltered throughout the iterative denoising process. Furthermore, we address normal tissue compression and deformation caused by the mass effect of TBI lesions through diffeomorphic registration, reducing erroneous activations in residual images and final lesion maps.ResultsExtensive experiments were conducted on three publicly available brain lesion datasets and one internal dataset. These datasets comprised 75, 51, 92, and 56 CT scans, respectively. Thirty seven CT scans without TBI lesions were used for training and validation, while the remaining scans were used for testing. The proposed method achieved average DSC of 0.56, 0.51, 0.47, and 0.52 and AUPRC of 0.57, 0.48, 0.53, and 0.50 on the BCIHM, BHSD, Seg-CQ500, and internal datasets, respectively, surpassing state-of-the-art unsupervised methods for TBI lesion detection and segmentation. An ablation study validated the effectiveness of the proposed NTR regularization and diffeomorphic registration-based mass effect simulation.ConclusionsThe results suggest that the proposed SCDM enables effective TBI lesion detection and segmentation across diverse TBI CT scans. It significantly reduces false positives by addressing inconsistencies in normal tissue during the iterative image restoration process and mitigating mass effect-induced tissue deformations.

Keyword ：

semantic-consistent diffusion model deformable registration unsupervised traumatic brain injury localization

Cite：

Copy from the list or Export to your reference management。

Abstract ：

While the utilization of transportation systems is on the rise, significant data quality concerns persist, including data loss and noise arising from network transmission delays and detector malfunctions. Various methods for data imputation exist, among which diffusion-based approaches have demonstrated competitive outcomes. Nonetheless, diffusion models, primarily employed in matrix-structured data like images, fail to fully exploit the inherent graph structure of traffic data. To enhance the quality of data filling, we propose a novel method that combines spatio-temporal transformer and a conditional diffusion model (STCDM). The introduction of the conditional diffusion model involves using observable traffic data as conditional information in the reverse process, allowing it to learn the underlying probability distribution and guide the generation of high-quality data samples. The spatio-temporal transformer module is selected as the basic denoising function, capturing comprehensive spatio-temporal context information of traffic data. Our experimental results, conducted on public transportation datasets with various missing patterns and rates, indicate that STCDM exhibits superior performance by achieving up to a 1.11% improvement over the second-ranked conditional score-based diffusion model across popular performance metrics.

Keyword ：

data mining management and control neural nets traffic demand forecasting traffic modelling

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Ptychography offers several advantages, including a lensless design, a simple optical setup, high spatial resolution, and the elimination of the need for high-quality optical components. In applications like surface topography measurement, ptychography setups are typically configured in reflective mode. However, existing studies indicate that normal-incidence configurations are often complex and constrained by the arrangement of beam splitters. Furthermore, tilted illumination with vertical detection typically involves tilted plane correction methods for pre-correcting diffraction patterns or by developing more refined models, which can degrade reconstruction accuracy and increase computational overhead. This study derives a forward propagation model for reflective off-axis diffraction, where the beam maintains the same angle relative to both the sample plane and the recording plane, thereby eliminating the need for a pre-correction process. Based on our propagation model, we developed a regularized ptychography iterative engine combined with a total variation denoising algorithm, effectively suppressing sensor noise and potential experimental inaccuracies. Reflective samples were successfully reconstructed, and comparisons with pre-correction methods on USAF 1951 targets demonstrated a significant improvement in reconstruction speed as well as enhanced accuracy. Finally, the proposed method accurately retrieved the surface morphology of atomic force microscope test samples. © 2025 Elsevier Ltd

Keyword ：

Optical design Diffraction patterns Optical beam splitters Lenses Image resolution

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Fault detection and diagnosis in HVAC systems are essential for maintaining energy efficiency and indoor comfort. However, the scarcity of fault samples, particularly for rare faults, leads to severe data imbalance, degrading model performance and increasing false alarms. While deep learning methods have improved diagnostic accuracy, they often struggle to capture the complex spatiotemporal interactions of HVAC faults using one-dimensional data. To address this challenge, we propose a novel data augmentation framework based on Denoising Diffusion Probabilistic Models, integrating diffusion models with Gramian Angular Field transformation. This approach effectively captures intricate dynamic patterns and generates high-quality synthetic fault samples, helping to mitigate data imbalance. Experimental results on the ASHRAE dataset demonstrate that our method outperforms existing approaches in sample quality, data distribution alignment, and diagnostic accuracy, achieving a 3.78 % improvement over CVAE-GAN while significantly reducing false positives and missed detections for rare faults. Additionally, we introduce a comprehensive evaluation framework to ensure that generated samples meet high application standards. By providing a more robust and generalizable solution for HVAC fault detection, this study contributes to the advancement of intelligent building management and energy-efficient operation. © 2025 Elsevier Ltd

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Speech recognition refers to the process of receiving and understanding human speech input through a computer, converting it into readable text or instructions. In order to improve the denoising effect and speech recognition effect of f uzzy speech, a fuzzy speech recognition algorithm based on continuous density hidden Markov model and self-organizing feature map is proposed. Firstly, the conventional Wiener filtering algorithm is improved by using the dynamic estimation algorithm of noise power spectrum, and the endpoint detection of noisy speech signal is performed by using spectral entropy, and the noise power spectrum of the silent segment is dynamically updated according to the detection results to obtain a more ideal priori signal to noise ratio; Secondly, the fuzzy speech is input into the Wiener filter to eliminate the noise in the speech signal; then, Mel-Frequency Cepstrum Coefficient (MFCC) of speech signal is extracted as speech feature; Finally, combined with the continuous hidden Markov model and the self-organizing feature neural network in the artificial intelligence algorithm, through the process of adjusting parameters, Viterbi decoding, and the time adjustment of the voice signal in the same state, the speech classification and recognition are realized according to the speech characteristics. In the experiment, comparative experiments were conducted on the LibriSpeech dataset using speech recognition algorithms based on convolutional neural networks and recurrent neural networks, speech recognition algorithms based on residual networks and gated convolutional networks, speech recognition algorithms based on multi-scale Mel domain feature map extraction. The experimental results show that the algorithm has good denoising performance. With the increase of added environmental noise intensity, the algorithm can maintain the Signal-to-Noise Ratio (SNR) of speech signals between 88dB-98dB; This algorithm can accurately detect the sound areas in the signal, and the endpoint detection accuracy is high; The accuracy and recall of the Continuous Density Hidden Markov Model-Self-Organizing Feature Neural Network (CDHMM-SOFM) designed in the algorithm increase with the number of iterations, and the highest levels of accuracy and recall can reach 0.89, respectively; The minimum recognition time of this algorithm is only 8.2 seconds, and the highest recognition rate can reach 98.7%; after applying this algorithm, the user’s error rate ranges from 0.0031 to 0.0084. The above results indicate that the algorithm has good application performance. © 2025, Zarka Private University. All rights reserved.

Keyword ：

continuous hidden Markov model Speech recognition wiener filter Mel-frequency cepstrum coefficient self-organizing feature neural network

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Among hyperspectral imaging technologies,interferometric spectral imaging is widely used in remote sening due to advantages of large luminous flux and high resolution.However,with complicated mechanism,interferometric imaging faces the impact of multi-stage degradation.Most exsiting interferometric spectrum reconstruction methods are based on tradition model-based frame-work with multiple steps,showing poor efficiency and restricted performance.Thus,we propose an interferometric spectrum reconstruction method based on degradation synthesis and deep learning.Firstly,based on imaging mechanism,we proposed an mathematical model of interferometric imag-ing to analyse the degradation components as noises and trends during imaging.The model con-sists of three stages,namely instrument degradation,sensing degradation,and signal-independent degradation process.Then,we designed calibration-based method to estimate parameters in the model,of which the results are used for synthesizing realistic dataset for learning-based algorithms.In addition,we proposed a dual-stage interferogram spectrum reconstruction framework,which sup-ports pre-training and integration of denoising DNNs.Experiments exhibits the reliability of our degradation model and synthesized data,and the effectiveness of the proposed reconstruction method.

Cite：

Copy from the list or Export to your reference management。

Abstract ：

3D Reconstruction Using Images has made strides in small-scale, uncomplicated scenes but struggles with complex, large-scale architectural forms. Targeting early-stage architectural design, we introduce ArchiDiff, a platform for 3D architectural form generation and editing from images to point clouds. First, we curated a dataset specifically tailored for architectural design, ArchiCloudNet. Second, we proposed a 3D generation method using a conditional denoising diffusion model, with an arbitrary object segmentation model to enhance recognition capabilities in complex input. Finally, we incorporate an interactive feature enabling instantaneous 2D image editing through simple drag-and-drops with simultaneous updates to 3D forms, giving designers improved control. We evaluated ArchiDiff's generation accuracy against cutting-edge baselines on ArchiCloudNet and two other datasets, RealCity3D and BuildingNet. We also validated it with real sketches from early-stage architectural design. The experiments indicated that our model could generate accurate architectural point clouds, providing rapid-response modification and effective processing of complex backgrounds. Demostration: http://39.101.72 .109:3000/archidiff.

Keyword ：

3D diffusion model 3D interaction Architectural design Single-view reconstruction

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Few-shot object detection involves detecting novel objects with only a few training samples. But very few samples are difficult to cover the bias of the new class in the deep model. To address the issue, we use self-supervision to expand the coverage of samples to provide more observation angles for new classes. In this paper, we propose a multi-task approach that combines self-supervision with few-shot learning to exploit the complementarity of these two domains. Specifically, our self-supervision as an auxiliary task to improve the detection performance of the main task of few-shot learning. Moreover, in order to make self-supervision more suitable for few-shot object detection, we introduce the denoising module to expand the positive and negative samples and the team module for precise positioning. The denoising module expands the positive and negative samples and accelerate model convergence using contrastive denoising training methods. The team module utilizes location constraints for precise localization to improve the accuracy of object detection. Our experimental results demonstrate the effectiveness of our method on the Few-shot object detection task on the PASCAL VOC and COCO datasets, achieving promising results. Our results highlight the potential of combining self-supervision with few-shot learning to improve the performance of object detection models in scenarios where annotated data is limited.

Keyword ：

Few-shot object detection End-to-End Detector Self-supervised learning

Cite：

Copy from the list or Export to your reference management。

Abstract ：

This paper designs a weighted minimum mean square error (WMMSE) based distortionless convolution beamformer (DCBF) for joint dereverberation and denoising. By effectively using WMMSE with the constraint of distortionless, a DCBF is deduced, where the outputs of the weighted prediction error (WPE) filter and the WPE-based minimum variance distortionless response (MVDR) beamformer are combined to initialize target signal for balancing signal distortion, residual reverberation and residual noise. In addition, two optimization factors are introduced to reduce the reverberation and noise when the initialized target signal is used for the solution of beamformer. As a result, the designed beamformer is presented as a linear combination of the WMMSE-based convolution beamformer (CBF) and weighted power minimization distortionless response (WPD) filter. The experimental results demonstrate the superior performance of the designed beamformer for joint dereverberation and denoising compared to the reference methods. © 2024 Elsevier B.V.

Keyword ：

Errors Convolution Reverberation Mean square error Beamforming

Cite：

Copy from the list or Export to your reference management。