33 results about "Filter noise" patented technology

This invention discloses a weakly supervised semantic segmentation method and system for remote sensing images based on adversarial background interference, belonging to the fields of pattern recognition and machine learning. The method includes: First, prototyping a background prototype mining approach. By constructing a pure background prototype library and calculating pixel-level similarity, a background mask is generated to suppress false activations of interfering backgrounds from a priori level. Second, foreground-background decoupling is designed. A contrastive loss based on prediction entropy weighting is introduced, driving the model to aggregate background features and move away from foreground features in the feature space, enhancing semantic discriminative power. Finally, a pseudo-label optimization strategy is employed. Through hole filling and dual-threshold screening, the structural integrity of pseudo-labels is repaired and noise is filtered, providing high-quality supervisory signals. This method aims to solve the technical problems in existing weakly supervised semantic segmentation of remote sensing images, where severe background interference leads to high pseudo-label noise and incomplete foreground target segmentation.

This invention discloses a multimodal collaborative denoising product recommendation method based on modality balance. The method first constructs behavior-aligned multimodal semantic encoding and projects it onto the behavior semantic space to obtain a multimodal feature product sequence. Next, for the multimodal feature product sequence, a multimodal-aware collaborative denoising module is constructed to collaboratively filter noise from each modality, resulting in a denoised intermediate product sequence. By introducing positional encoding in conjunction with the collaborative denoising module, an enhanced product sequence is obtained. Finally, based on the enhanced product sequence, cross-modal fusion weights are generated, prediction scores are calculated, and the product with the highest score is recommended. A joint loss function is constructed, and the global parameters are iteratively updated using a backpropagation algorithm. This invention effectively solves the problems of noise interference and modality learning imbalance in multimodal recommendation, suppresses the excessive dominance of strong modalities in the early stages of training, and improves the accuracy and robustness of product recommendations.

This application belongs to the field of intelligent manufacturing technology, specifically relating to a method and system for predicting and generating maintenance suggestions for metal tool wear. It aims to solve the problems of low accuracy in tool wear prediction and delayed maintenance decisions. The method includes: constructing a multi-source physical feature sensing system to collect and align multi-dimensional signals during the machining process in real time; performing heterogeneous data preprocessing and feature enhancement, and filtering noise using adaptive wavelet packet decomposition; extracting multi-dimensional quantitative indicators and selecting a subset of core features; constructing a prediction model that integrates a deep residual network, a bidirectional long short-term memory network, and an attention mechanism to output the real-time wear level and remaining useful life; this application achieves accurate capture of wear status and optimized scheduling of maintenance resources through multi-source data fusion and deep learning evolutionary analysis, thereby improving machining quality and reducing unplanned downtime.

The application discloses a multi-modal semantic recognition method and system based on cross-modal contrast learning, belongs to the technical field of artificial intelligence and multi-modal semantic recognition, and solves the problems of mode quality difference and difficult sample recognition in multi-modal semantic recognition.The method comprises the following steps: S1, data preprocessing: standardizing and preprocessing visual, speech and text multi-modal data; S2, multi-modal feature extraction: extracting feature representations of visual, speech and text modes respectively; S3, cross-modal contrast learning: mapping the multi-modal features to a unified space, and enhancing the semantic consistency between modes through contrast learning; S4, adaptive mode fusion: dynamically adjusting the fusion weight according to the mode quality, and filtering noise features through a gating mechanism; S5, joint optimization: constructing a joint loss function of classification loss and contrast loss; S6, semantic recognition: outputting a target category recognition result; and S7, difficult sample mining: identifying difficult samples based on prediction uncertainty and strengthening training.

This invention discloses an adaptive wavelet denoising method for noise characteristic verification of PEPT trajectory data. The method first performs statistical analysis on the PEPT trajectory data under static conditions to obtain its noise characteristics. Then, it uses an adaptive wavelet denoising method to obtain the denoised PEPT trajectory and noise data. It then verifies whether the statistical characteristics of the filtered noise data conform to the noise characteristics under static conditions. Based on the verification results, the denoising parameters are dynamically adjusted for iterative feedback until the verification conditions for noise characteristics are met, at which point the method stops. This method uses "filtered components conforming to noise characteristics" as the convergence criterion, providing an objective statistical verification basis for denoising quality. It can effectively filter out as much noise data as possible from the PEPT trajectory without excessively smoothing out the real high-frequency motion. Moreover, it does not rely on the prior motion model of the tracer particles, thus having stronger universality.

This invention relates to the field of industrial defect detection technology, and proposes a cross-scale industrial surface defect detection method and system, including the following steps: acquiring template image I a And the industrial product image to be inspected I b And preprocess it; based on the Blob algorithm, the preprocessed image I a I b Feature extraction is performed to obtain image features I'. a 、I' b The image I is estimated using a Homography estimation neural network. a and I b and image features I' a and I' b Perform alignment calibration separately; for the aligned and calibrated image I a I b The difference image is processed by subtraction, and a sliding window is used to filter noise from the difference image to obtain the industrial surface defect detection result. This invention introduces a Blob analysis algorithm to locate small-scale defects in multi-scale defects, and combines it with a Homography estimation neural network to align and calibrate the standard sample image used for the defect sample to be detected, so as to achieve better difference results.

The application discloses a kind of comprehensive evaluation and early warning system and method for mine power supply safety, it is related to mine power supply safety management and intelligent early warning technical field, deployment is used in mine intrinsic safety multi-parameter acquisition terminal, integration is multiple sensor, according to parameter characteristic set acquisition frequency;Filter noise after extracting equipment parameter time domain and frequency domain characteristics;With CNN-LSTM hybrid model combined with transfer learning generates health index;Dynamic threshold value is divided into three levels and early warning, information is pushed in multiple channels;With gas, ventilation, emergency power supply system;With MySQL and MongoDB storage, block chain stores key records and builds digital twin model, three color standard equipment health state.The application can synchronously collect multidimensional parameters, quantify the influence of environment on equipment health, dynamically adjust the early warning threshold to reduce false and missed reports, link multiple systems to improve emergency efficiency, distributed storage and block chain ensure data reliability, digital twin assists intuitive management, and improve the intelligent level of mine power supply safety and management.

This invention discloses a method, terminal device, and storage medium for noise identification and filtering in lidar, comprising: clustering line segments according to distance similarity in the scanning order for each scan line, marking suspected noise line segments in each scan line to obtain a set of suspected noise line segments; performing rapid clustering on each suspected noise line segment in the set to obtain multiple targets, and deleting point noise corresponding to targets containing fewer than a threshold line segments from the point cloud; calculating the linear representation of the suspected noise line segments and determining whether it is a centripetal line, and deleting the linear noise corresponding to the centripetal line from the point cloud. This invention is applied to the field of lidar, solving the problem of identifying and filtering point and linear noise in lidar, with low time consumption, compatibility with any multi-line mechanical rotating lidar, strong generalization ability, and the ability to retain real target point clouds with similar noise while filtering noise, resulting in extremely low false detection and false detection rates.

The application discloses a method and system for finding potential associated addresses based on communication data, which comprises the following steps: obtaining full-amount communication data of a target object; constructing an explicit associated address set of known addresses; setting a night period, extracting candidate residence points based on the rule that the last call of the day and the first call of the next day access the same base station and do not fall into the explicit associated address set; calculating the spherical distance between the candidate residence points and the explicit associated address set for secondary filtering; performing spatial matching between the remaining residence points and a real estate address entity library containing asset valuation information to generate a potential associated address list; and counting the occurrence frequency, calculating the confidence score by using an exponential decay model and generating an early warning level. The application solves the problem of sparse data by using the rule that the first and last communication data access the same base station, effectively filters noise and known address interference by combining the exclusion mechanism and multi-dimensional value evaluation, and realizes high-precision automatic mining of potential associated addresses and quantitative risk early warning.

The application discloses a digital event triggering method and system based on baseband data sequence edge detection, relates to the field of baseband data detection, and comprises the following steps: removing direct current bias, suppressing low-frequency drift and filtering noise from an original baseband data sequence to obtain a preprocessed baseband data sequence; determining a morphological structure element according to a signal-to-noise ratio and an expected event width of a target service scenario, and extracting a local mutation profile through an open operation and a top-hat transformation; generating a double threshold based on a morphological gradient, statistical characteristics and noise estimation to extract a candidate event interval; performing multi-dimensional feature matching on the candidate event segment in combination with a reference event feature template to determine a target effective event; and determining an event time center point and start and end boundaries according to a local event energy sequence and event positioning constraint information and outputting triggering information containing an event timestamp and event segment data. Thus, the stability, accuracy and time positioning reliability of digital event triggering in a complex interference environment are improved.

The application provides a Zhuang language audio extraction method, including obtaining a first text set and Zhuang language audio through video processing, the first text set including a plurality of first texts A and corresponding text time axes a, step two, inputting a video, extracting an audio file, performing Fourier transform on the audio file to obtain an audio spectrum, setting a waveform amplitude threshold, filtering noise, the noise being audio with a waveform amplitude lower than the threshold, obtaining target audio B and a target audio time axis b, step three, aligning the text time axes a and the target audio time axes b, and adjusting the length of the text time axes a to be consistent with the length of the target audio time axes b; the scheme first aligns the audio and the text in the time domain, and then aligns them in the frequency domain based on the alignment in the time domain, so that the result is more accurate, and the audio and the text are aligned in the frequency domain and the time domain, so that the text and the audio information completely correspond.

This invention provides a multi-channel receiving method, apparatus, storage medium, and electronic device for ultrasound imaging. The method includes: S1, controlling the multi-channel transmission signals on the channel device; S2, controlling each element on the probe to receive signals transmitted by at least two channels on the channel device; S3, filtering noise from the signals transmitted by at least two channels on the channel device received by each element to obtain effective signals; and S4, performing ultrasound imaging based on the effective signals. The multi-channel receiving method, apparatus, storage medium, and electronic device for ultrasound imaging of this invention utilize multi-channel transmission and single-channel reception, and perform noise filtering to effectively eliminate the influence of noise and improve the signal-to-noise ratio.

Embodiments of the present disclosure disclose a voice control-based video playing method and device, electronic equipment and a medium. A specific implementation of the method comprises: collecting a video playing voice signal in a noisy environment to obtain a video playing audio signal; filtering noise from the video playing audio signal to obtain a filtered audio signal; performing noise separation processing on the filtered audio signal to obtain a target audio signal; performing amplitude normalization processing on the target audio signal to obtain a normalized audio signal; performing text conversion processing on the normalized audio signal to obtain audio text information; generating an audio text vector according to the audio text information; performing video retrieval on a target video corresponding to the audio text vector to obtain a video retrieval result; and generating a video playing control instruction according to the audio text information and the video retrieval result to play the video. The implementation can reduce the time for retrieving and playing the video and the error operation for playing the target video.

This invention belongs to the technical field of information technology, specifically disclosing a fault-tolerant method and system for streaming media services that supports both pre-broadcast and post-continuous push. The method includes: real-time monitoring of network fluctuations and high packet loss rates; obtaining the current buffered data reception speed and end-to-end latency accumulation value from the streaming media transmission data; processing these values ​​using a Kalman filter algorithm to filter noise and obtain a smoothed network feature vector; determining whether a chain reaction has occurred based on the obtained smoothed network feature vector; if the latency accumulation exceeds a preset threshold and the buffered data reception speed is below average, a potential interruption risk is identified, and a risk assessment report is generated. This invention solves the problem of potential interruption risks caused by network fluctuations and high packet loss rates in existing streaming media transmission technologies.

The embodiment of the application provides a sensor circuit, which comprises a power supply circuit, a signal processing circuit and a filtering circuit. The power supply circuit is used to provide voltage to the signal processing circuit so that the signal processing circuit can normally operate. The filtering circuit is connected to the output end of the signal processing circuit and the ground end of the power supply circuit respectively, and filters noise signals from the ground end. In this way, for a high-precision requirement scene, the voltage at both ends of the signal processing circuit can be stabilized, noise can be suppressed, and finally the sensor circuit can achieve better performance and higher accuracy.

This invention discloses a method and system for identifying the country of origin of radio signals based on multi-dimensional features, belonging to the field of signal identification technology. The method includes receiving raw data of a target signal transmitted by a radio receiving device; after null filtering and noise suppression preprocessing, extracting signal feature data, audio data, and direction of arrival data; comprehensively comparing the signal feature data and the receiving device's geographical location data with a pre-set database of known radio stations to obtain station matching results; demodulating the audio data and identifying the language using a pre-trained model to obtain language identification results; calculating the signal's azimuth data based on the direction of arrival data, the receiving device's geographical location, and border line geographical data; and performing dynamic weighted analysis on the three types of results to determine the country of origin of the signal and its corresponding confidence value. Through multi-dimensional information complementarity and cross-validation, the method solves the problem of unclear signal sources in border scenarios and is suitable for border radio management.

The application discloses a power plant equipment real-time fault early warning method based on an AI diagnosis algorithm, relates to the technical field of power diagnosis, and comprises the following steps: collecting multi-source signals, interpolating and aligning the signals and filtering noise to form a unified operation data frame; extracting a change rate, a fluctuation amplitude and statistical characteristics based on the unified operation data frame, identifying a working condition and normalizing characteristics by using a finite state machine, and generating a structured characteristic frame; inputting the structured characteristic frame into a double-layer LSTM time sequence AI diagnosis model, performing online inference and order-preserving calibration, and outputting a predicted state set; positioning a potential fault mode through the predicted state set, combining a comprehensive verification score calculated based on resonance characteristics, and generating a disposal record set; and performing adaptive learning and parameter updating through the disposal record set. Through the design of the double-layer LSTM time sequence AI diagnosis model and the combination of the online order-preserving calibration algorithm, trend prediction and abnormal confidence adaptive quantization of key operation parameters of the power plant are realized.

The application discloses a kind of based on geometry feature mapping and enhanced EEGNet on-load tap-changer fault diagnosis method, it is related to on-load tap-changer fault diagnosis technical field, for solving industrial field vibration signal non-stationary, weak fault is easy to cover and depth model decision is not transparent operation and maintenance field problem.The application first carries out Hilbert-Savgol filter noise reduction and phase space plotting signal preprocessing operation, then the data after preprocessing is carried out nesting convex hull mask and distance field calculation geometry feature coding operation, finally, data processing is carried out using enhanced attention EEGNet network.The application eliminates the black box attribute of deep learning model by constructing the strong coupling explanation logic of thermal distribution feature-geometric topological motion-physical operation mechanism, provides intuitive, traceable diagnostic evidence for operation and maintenance personnel.

The present application belongs to the field of geological disaster monitoring and early warning, and discloses a landslide deformation monitoring method based on SAR image offset tracking and multi-stage filtering. The registered SAR image data set is divided into image pairs, the original offset field of each image pair is calculated by using the block phase cross-correlation algorithm, and the original offset field is comprehensively processed by multi-stage filtering based on correlation threshold and signal-to-noise ratio threshold filtering, median filtering, statistical aggregation, amplitude threshold filtering, vector consistency filtering and density clustering analysis, so that noise is effectively removed, systematic deviation is corrected, and real and coherent landslide deformation signals are refined, thereby providing a more accurate and reliable landslide deformation monitoring solution. The present application focuses on image pixel feature tracking and post-processing, integrates various technologies to filter noise step by step, can realize the identification and long-term dynamic monitoring of landslides, improves the accuracy and reliability of landslide monitoring, and provides technical support for preventing and controlling landslide disasters.

The application belongs to the field of magnetic resonance sounding signal noise filtering, and is a kind of magnetic resonance sounding signal extraction method based on intelligent optimization manifold learning, the parameter group of the local linear embedding manifold learning method is initialized, the genetic algorithm in the intelligent optimization algorithm is used, the signal-to-noise ratio is taken as the fitness function, and the parameter group in the local linear embedding manifold learning method is optimized, the local linear embedding manifold learning method uses the optimized parameter group to sequentially perform first processing and second processing on the magnetic resonance sounding signal, removes random noise, and obtains the final denoised magnetic resonance sounding signal. The application effectively retains the signal characteristics through the nonlinear dimension reduction of manifold learning, avoids the information loss caused by frequency band selection, maintains the local relationship between data points in the dimension reduction process, ensures that adjacent points in m-dimensional space remain adjacent in d-dimensional space, only compresses and filters noise, and realizes signal extraction.

The application provides a kind of unmanned underwater vehicle target generation type active sonar received signal authenticity evaluation method, system and related device, including hybrid position coding module, denoising module, feature enhancement module, joint feature extraction module, respectively for embedding signal position information, filtering noise, signal feature enhancement, signal feature extraction, finally to the active sonar received signal of unmanned underwater vehicle target generated by deep learning is classified, the corresponding classification probability of unmanned underwater vehicle target echo signal and noise signal is used as authenticity evaluation result.The application provides the feature extraction capability of unmanned underwater vehicle target echo signal and noise signal, is suitable for unmanned underwater vehicle target echo signal and noise signal classification, provides more accurate method for the authenticity evaluation of unmanned underwater vehicle target generation type active sonar received signal.

The application relates to a parameter determination method and device of a micro-perforated panel for spectrum noise reduction of intelligent electrical appliances, applied to the field of sound absorption and noise reduction, wherein the method comprises the following steps: determining a frequency attenuation curve of the micro-perforated panel based on initial parameters of the micro-perforated panel; obtaining original spectrum data of an intelligent electrical appliance to be reduced in noise; determining a filtering noise reduction amount according to the original spectrum data and the frequency attenuation curve; and predicting target parameters corresponding to a predicted noise reduction amount meeting a preset noise reduction threshold range in the micro-perforated panel based on a preset Bayesian optimization strategy and by using a target noise reduction prediction model. The target noise reduction prediction model is obtained by training a preset regression model based on the filtering noise reduction amount and the initial parameters of the micro-perforated panel, and is used for predicting filtering noise reduction amounts corresponding to different micro-perforated panel parameters. Through the application, the accuracy and efficiency of micro-perforated panel parameter design are improved.

The application discloses a gas identification method suitable for low-resolution high-noise spectral data, and is suitable for low-resolution high-noise spectral data obtained by a medium-long wave infrared spectrometer or a hyperspectral imager in an open space in a passive and passive measurement mode, and performs the following steps: a) extracting spectral radiation values of a measured gas at corresponding characteristic spectral position wave number coordinates by taking preset typical gas spectral characteristic points as templates, and filtering noise interference outside the characteristic spectral position wave number coordinates; b) judging a positive-negative matching number coefficient of multiple characteristic spectrums, judging a positive-negative matching direction coefficient of the multiple characteristic spectrums, and jointly judging a characteristic vector cosine similarity, so that the dimension of the judgment is increased, and the positive-negative temperature difference change of the gas and the background is adapted; and c) comprehensively judging by multiple gas similarity sorting methods. The application effectively matches the spectral gas identification accuracy, and is suitable for application scenes limited by conditions such as an open space environment, long-distance passive detection, and low sensor sensitivity.

The present application relates to the technical field of network attack tracing, and particularly relates to a network attack tracing method and system based on malware samples; the present application maps high-dimensional heterogeneous static semantic vectors and dynamic semantic vectors to a unified dimension space through a fully connected projection layer, eliminates feature scale differences, and dynamically calculates semantic correlation weights of static API control flow features and runtime behavior sequences using a cross-modal attention mechanism, generates interaction vectors to strengthen potential attack chain coupling relationships, and then filters noise signals through vector splicing, aggregation and dimension reduction operations, finally outputs discriminative and significant fusion feature vectors, solving the feature flooding and semantic fragmentation problems caused by simple splicing of dynamic and static features in traditional methods; solving the technical problem that feature information in two dimensions of dynamic analysis and static analysis has not been effectively fused in the prior art to realize all-around and in-depth tracing of attack behaviors.

This invention provides a noise reduction method for machine tool spindle rotation signals based on VMD-PE-MWSTD, belonging to the field of precision measurement technology. The method includes: first, acquiring and preprocessing the rotation signal of the machine tool spindle; then, using Variational Mode Decomposition (VMD) to decompose the signal into several Intrinsic Mode Function (IMF) components; next, calculating the permutation entropy (PE) of each IMF component, and automatically identifying and filtering noise-dominated IMF components and useful signal-dominated IMF components based on a preset PE threshold; processing the filtered noise-dominated IMF components using an improved wavelet soft thresholding (MWSTD) method to remove noise while retaining the effective components in the signal; finally, reconstructing the denoised IMF components and the useful signal-dominated IMF components to obtain the final denoised rotation signal. This invention effectively separates signal and noise, significantly improves the signal-to-noise ratio, and retains the true characteristics of the signal, providing a high-quality data foundation for machine tool spindle accuracy monitoring and health diagnosis.

The application provides a crack detection method and device based on a bounding box guide constraint and electronic equipment, and relates to the technical field of image analysis and processing. The method comprises the following steps: obtaining crack labeling information, the crack labeling information comprising a target bounding box defined on a to-be-detected image and surrounding a target crack, and an artificial anchor point labeled for the target crack; performing dynamic expansion based on the target bounding box to obtain a corresponding background region, and performing adaptive image segmentation based on local contrast analysis of an image region in the target bounding box and the background region to generate a crack segmentation image; extracting at least one high-confidence connected domain from the crack segmentation image as a crack sample according to the artificial anchor point, learning a crack feature model to filter noise of the crack segmentation image, and performing structure optimization and vectorization reconstruction on crack pixels after noise filtering to output a final crack path. The application improves the crack detection accuracy in a complex environment.

This invention discloses a method for denoising 3D point cloud data in complex scenes, belonging to the field of data processing technology. The method includes: screening candidate ground points; performing ground fitting algorithm to fit the ground and segment the ground point cloud; clustering the non-ground point cloud after removing the ground point cloud to obtain point cloud clusters belonging to each obstacle; constructing a 3D bounding box estimation model based on graphical envelope; estimating the 3D bounding box of the clustering results to obtain the minimum 3D bounding box of each obstacle point cloud; identifying and filtering dust, water, and high-inverse-dilation crosstalk noise in the scene to obtain high-precision point cloud data. This invention can effectively detect and filter noise in complex scenes and has good real-time performance and robustness.

This invention discloses a non-contact sleep staging assessment method based on millimeter-wave radar and TinyML, relating to the fields of sleep monitoring, embedded artificial intelligence, and IoT health management. The method sequentially completes millimeter-wave micro-motion signal acquisition and joint filtering noise reduction, multi-dimensional feature extraction of respiratory rhythm / respiratory amplitude / body movement events, PCA feature compression and 8-bit linear quantization, TinyML edge inference, three-level stabilization post-processing, task-scheduled local storage, and edge-cloud collaborative data interaction, ultimately outputting standardized sleep staging results and a sleep health assessment report. This invention abandons traditional wearable monitoring methods, achieving interference-free and highly private sleep data acquisition. Through lightweight algorithms adapted to resource-constrained microcontrollers, it addresses the technical pain points of edge inference timing jumps, unreliable data storage, and lack of edge-cloud collaboration.