196 results about "Filter noise" patented technology

The invention discloses a cross-modal remote sensing image unsupervised adaptive method based on unreliable pseudo tag guidance. The method comprises the following steps: step 1, acquiring data of a source domain and a target domain and preprocessing the data; 2, constructing a teacher-student self-training framework, generating a pseudo tag for a target domain by a teacher network, and filtering noise through confidence evaluation; 3, dividing the image into reliable pixels and unreliable pixels according to the confidence coefficient, and generating a mask; 4, constructing positive samples, negative samples and anchor point features; 5, designing an unreliable sample guide pixel contrast loss function; 6, optimizing the loss function training model until the optimal performance is achieved; and 7, predicting test set data by adopting the trained model to obtain a semantic segmentation result. According to the method, the potential of pseudo labels of which modals are difficult to label is fully mined, the problem that unreliable pixels are insufficient in utilization during training is solved, and finally more effective cross-modal domain alignment and better cross-modal unsupervised domain adaptive semantic segmentation precision are realized.

The invention provides an underground ore body positioning identification method and system based on a ground penetrating radar, and relates to the technical field of ore body detection, and the method comprises the steps: inputting a corrected multi-dimensional point cloud feature set into a pre-trained closed-loop convolutional neural network, optimizing the network parameters through transfer learning, and obtaining a multi-dimensional point cloud feature set; outputting the category confidence and the spatial position probability distribution point cloud of the ore body target; on the basis of the spatial position probability distribution point cloud, generating ore body depth and horizontal positioning point cloud through a reverse projection algorithm in combination with an electromagnetic wave two-way travel time calculation rule and a Kanay resistivity measurement result; and performing multi-scale two-dimensional empirical mode decomposition on the reverse projection positioning point cloud, and filtering noise components to extract a high-confidence target point cloud. According to the invention, the accuracy and stability of ore body identification are improved.

The invention relates to the technical field of radar echo denoising, in particular to a radar echo signal filtering noise self-adaptive suppression method and a radar echo signal filtering noise self-adaptive suppression system. Acquiring radar station position and beam pointing data, and synchronously acquiring forest vegetation data and environment interference data of a beam pointing target area to form radar environment data; the radar environment data are identified through the multi-band decision model, and multi-modal waveform parameters are obtained; based on the multi-modal waveform parameters, radar equipment is controlled to perform periodic detection on the target area, and radar echo data are obtained; the method comprises the steps of identifying radar echo data and radar environment data, extracting dynamic spectrum characteristics in the radar echo data, performing clutter suppression processing on the dynamic spectrum characteristics by using the radar environment data, calculating the confidence probability that an early fire exists in a target area, and executing space-time consistency joint verification; according to the invention, fire early warning is carried out through the confidence probability of the fire.

The invention provides an image video retrieval method based on a domain fine-tuning large language model, which comprises the following steps: performing fine-tuning on a pre-training model to obtain a fine-tuning pre-training model for intention classification and keyword extraction; performing dynamic iteration screening on an optimal prompt template through Monte Carlo tree search in combination with a hidden Markov model (HMM); performing noise filtering on the keyword list, and predicting category labels of the filtered keywords through a conditional random field model to obtain a keyword enhancement set; combining with the user intention to generate a query condition, and obtaining a candidate resource set; and according to the similarity between the user query text and the candidate resource set, and in combination with the optimal prompt template, obtaining the resource path with the highest matching score between the user query and the candidate resource, and obtaining the retrieved image or video, so that the identification deviation possibly occurring when a general model processes proper nouns and terminologies can be effectively solved, and the user experience is improved. And the retrieval accuracy and response speed are improved, so that the retrieval accuracy and professional adaptability are improved.

The invention provides a motor stator fault detection system and method, and relates to the technical field of data processing, and the method comprises the steps: synchronously collecting a three-phase current signal and a stator winding temperature signal during the operation of a motor in real time through a current sensor and a temperature sensor; performing time-frequency domain analysis and spectrum feature extraction on the three-phase current signal to obtain a current unbalance degree and a current harmonic distortion rate, and forming a current feature parameter set; performing filtering noise reduction and trend elimination processing on the stator winding temperature signal to obtain a processed signal; on the basis of the processed signals, temperature abnormal points are recognized through a k-nearest neighbor clustering algorithm of the three-dimensional Euclidean distance, the temperature instantaneous change rate is calculated according to the temperature abnormal points, and a temperature characteristic parameter set containing the abnormal point density and the temperature rise rate is obtained. According to the invention, stator fault detection and differential early warning are realized.

The invention discloses an ultrasonic phased array weld defect intelligent identification method based on adaptive filtering, and belongs to industrial nondestructive detection image identification methods, the technical scheme is that the method comprises the following steps: sampling and sorting pipeline weld PAUT detection images; dividing the collected image samples into a training set and a test set; pixel classification and filtering noise reduction are carried out on the training set images through a hybrid adaptive filter; performing abnormal signal labeling on the preprocessed training set image and the unprocessed test set image respectively; training a target recognition network through the preprocessed training set; and collecting a new fan-shaped scanning image of the pipeline weld defect as network input, and identifying the weld defect. The ultrasonic phased array weld defect intelligent identification method based on self-adaptive filtering has the beneficial effect that the ultrasonic phased array weld defect intelligent identification method based on self-adaptive filtering is provided.

The invention discloses a root cause alarm analysis method and system based on time sequence association and multi-modal fusion, and the method comprises the steps: taking digital twin dynamic threshold cleaning as a core, fusing multi-source data time sequence alignment verification, and forming a multi-source data set of time sequence alignment and preliminary restoration through exception handling and cross-system time consistency verification; a cleaning threshold is optimized in combination with a working condition to filter noise, a label confidence weighting method is fused, confidence is given to a label through twin simulation, and a multi-modal data set with a multi-dimensional label is constructed. A learning model is constructed and trained, exclusive feature extraction branches are designed for multi-source data, a twin working condition adaptation enhancement module is introduced to strengthen core modal extraction, and the proportion of abnormal modals in fusion is reduced by means of a dynamic attention filtering mechanism. Based on the model and a twinning system, root cause penetration type positioning is completed through dynamic sequencing of root cause confidence in combination with a twinning reduction time sequence. And the root cause analysis efficiency and accuracy are improved.

The invention relates to the technical field of traffic information collection, in particular to a tunnel traffic flow detection method based on a video cloud networking technology, and the method comprises the following steps: obtaining video flow data collected by each monitoring camera in a tunnel in real time; performing compression coding processing on the video stream data; performing noise filtering and frame synchronization preprocessing on the coded video data; in a preset edge computing node, inputting the structured video data into a vehicle flow detection model to extract vehicle feature parameters in the video frame; performing tracking association on the target vehicle in the video frame through an asymmetric Kalman filtering algorithm, and comparing the vehicle flow detection data with a preset vehicle flow threshold; and in the cloud analysis platform, analyzing the structured detection data through the tunnel traffic flow prediction model. According to the invention, noise filtering and time-space alignment can be carried out on the collected vehicle operation video data, and the accuracy of tunnel traffic flow detection is improved.

The invention discloses a monitoring system and method for determining an abnormal geologic body based on distributed optical fiber sound wave vibration, and relates to the technical field of distributed optical fiber sound wave detection.The monitoring system comprises a distributed optical fiber sensing unit, a signal excitation unit, a photoelectric demodulation unit, a signal processing and analyzing unit, a result recording and outputting unit and the like; the distributed optical fiber is used for continuously collecting stratum vibration response; the signal excitation unit applies controllable disturbance to a known reference position to induce a potential abnormal geologic body to generate an inactive excitation response signal; the photoelectric demodulation unit performs phase demodulation on backscattered light transmitted back by the optical fiber to form distributed vibration measurement data; and the signal processing and analyzing unit is used for filtering and denoising the data, establishing a quantitative mathematical model of a signal incident angle and signal intensity, solving an epicentral distance and an incident angle and calculating the spatial position of an abnormal geologic body by combining propagation attenuation correction and least square inversion, and is suitable for dangerous section identification and avoidance decision before mining and construction.

The invention discloses a self-adaptive air pressure regulation and control system of quartz sand unpowered powder selecting equipment, and relates to the technical field of automatic control. According to the method, wind pressure and airflow disturbance signals are collected, a high-resolution map is generated through Hampel filtering and noise reduction, features are extracted through the convolutional neural network, key influence areas are identified, the problem of unstable wind pressure caused by airflow disturbance in a mountainous area is solved, and the particle separation precision is improved; a natural potential energy distribution matrix is constructed based on building floor height difference, velocity field reconstruction and airflow channel optimization are driven, technological parameters are adjusted in combination with a gradient descent algorithm, precise utilization of natural potential energy is achieved, energy consumption is reduced, uneven efficiency is improved, airflow distribution is simulated through a fluid dynamic model, and sedimentation behaviors are analyzed and predicted through particle trajectory. A support vector machine is used for classifying tracks and generating regulation and control instructions, PID control is used for stabilizing air pressure to form a closed loop, the problems that fine powder is mixed with coarse powder, coarse particles are left and the like are avoided, the equipment blockage risk is reduced, and the product purity stability is improved.

The invention discloses an ammonia gas leakage multi-sensor fusion automatic monitoring method and system, and the method comprises the steps: deploying a plurality of gas detection sensors in a monitoring region, synchronously collecting the ammonia gas concentration and environmental parameters, and forming a multi-source data set; carrying out noise filtering and drift correction by utilizing the edge computing node, and extracting spatial-temporal characteristics to identify concentration anomaly; determining a leakage source coordinate through a data fusion algorithm, simulating a diffusion path and a trend, and constructing a risk diffusion model; and finally, an emergency trigger signal is extracted, ventilation equipment and valve control are activated, and real-time linkage response is achieved. According to the method, data layer fusion and a risk diffusion model are combined, so that the high efficiency of accurate positioning and emergency response of a leakage source is ensured, the safety and reliability of harmful gas monitoring and control are remarkably improved, and a technical guarantee is provided for gas leakage risk management in a complex environment.

The invention discloses an intelligent electric power measurement and decision-making method based on an Internet of Things architecture, and aims to solve the problems of large temperature measurement deviation, poor dynamic adaptation, unquantified aging influence and extensive meter change decision-making in the prior art. The method comprises the following steps: collecting multi-source data, denoising and fusing, and generating a fused temperature observation value; inputting historical time sequence data formed by the data and the fusion temperature into the TCN, and outputting a state matrix increment to correct a Kalman filtering state equation; optimizing a filtering noise covariance matrix through PSO (Particle Swarm Optimization); based on the correction equation and the optimal noise matrix, the initial temperature of the contact is obtained through improved Kalman filtering; in combination with an SVR model, the aging compensation amount is calculated by taking the multi-source denoising data and the initial temperature as input, and the final temperature is obtained through superposition; and calculating an aging influence coefficient according to the aging compensation amount and the final temperature, and executing early warning or meter replacement. According to the invention, temperature prediction precision and long-term stability are improved, intelligent meter changing decision is realized, and equipment safety and operation and maintenance efficiency are guaranteed.

The invention discloses a city safety monitoring and early warning method and system, and the technical key points are that the method comprises the following steps: S1, multi-source data collection: collecting heterogeneous city safety monitoring data through monitoring terminals distributed in all scenes of a city, the heterogeneous data comprising real-time video stream data, structured numerical data and semi-structured log data; s2, data preprocessing and fusion: inputting the heterogeneous data acquired in the step S1 into a multi-source data fusion model based on an attention mechanism, filtering noise data through a data cleaning module, and extracting spatial-temporal features of each data source through a feature extraction module; s3, performing dynamic risk assessment; s4, grading early warning and collaborative linkage are carried out; and S5, performing closed-loop optimization. According to the method, the multi-source data fusion precision can be improved by more than 35%, the weights of different data sources are automatically identified by constructing a heterogeneous data fusion model based on an attention mechanism, the problem of data islands is solved, and a reliable basis is provided for risk assessment.

The invention discloses a dynamic sleep memory intervention method and system based on an ultrathin eyeshade, the system comprises an ultrathin eyeshade body and a cloud platform, and the eyeshade integrates electroencephalogram acquisition, memory content playing, master control and communication modules. The electroencephalogram acquisition module adopts an eight-channel super-soft spandex conductive electrode and a noise reduction unit combining adaptive notch filtering with wavelet threshold filtering; the memory playing module is a hemispherical dome miniature bone conduction loudspeaker, and silent leakage outside 30cm is realized; a dynamic sleep memory dry prediction algorithm is built in the main control module, and playing parameters can be adjusted according to delta wave intensity; the cloud stores personalized memory content and intervention records, and push content can be intelligently updated. The method comprises the steps of equipment starting initialization, electroencephalogram collection and noise reduction, deep sleep recognition, dynamic playing and data encryption synchronous closing, the problems that existing equipment is heavy and intervention is rigid are solved, non-inductive wearing, accurate intervention and privacy protection are achieved, and the method is suitable for sleep memory consolidation.

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.

The invention discloses an RGB-T counting method and device based on semantic perception complementary feature mining. The method comprises the steps that firstly, images in a training data set are preprocessed; then constructing an RGB-T counting network based on semantic perception complementary feature mining, wherein the RGB-T counting network comprises a feature extractor based on Transform, a mode feature adaptive method from coarse to fine and a semantic perception agent module; the feature extractor based on the Transform is used for extracting high-level semantic representations of two different modal images; the mode feature adaptive method from coarse to fine is used for filtering noise features and mining fine complementary features; and the semantic perception agent module is used for enhancing semantic consistency perception in network output and the feature extractor, and predicting a density map and a counting result through a density regression layer. According to the method, the RGB-T crowd counting task can be efficiently completed, and the counting result is better than that in the prior art.

The invention relates to a touch control processing method and device, electronic equipment and a storage medium. The touch control processing method comprises the following steps: in response to a detected touch control operation, obtaining a first induction signal corresponding to the touch control operation; acquiring a correction parameter associated with the touch screen to which the first induction signal belongs; wherein the correction parameter is used for filtering noise introduced when the sensing signal of the touch screen is filtered, the correction parameter is obtained by carrying out noise reduction processing on a sample sensing signal, and the sample sensing signal is a signal obtained based on data associated with the touch screen to which the first sensing signal belongs; correcting the first induction signal by using the correction parameter to obtain a second induction signal; and determining a touch position according to the second sensing signal. According to the embodiment of the invention, the noise or redundant information in the first induction signal can be effectively reduced, and the principal component in the first induction signal is reserved, so that the quality of the second induction signal is improved, and the user experience is improved.

An example medical device includes a memory; and processing circuitry coupled to the memory, the processing circuitry is configured to: receive, from one or more electrodes coupled to the medical device, a cardiac signal of a patient; determine an initial indication of a cardiac event based on the cardiac signal; determine whether a particular feature of the cardiac signal satisfies a filter threshold; and in response to the particular feature of the cardiac signal satisfying the filter threshold, activate a filter to filter noise from the cardiac signal.

The invention discloses a streaming video subtitle generation method and system based on a localized large model, and a storage medium, and belongs to the technical field of artificial intelligence and natural language processing. The method comprises the following steps: constructing an asynchronous streaming audio extraction queue, and realizing parallel processing of millisecond-level response and background fragmentation of a first audio segment of a video; a decoupled voice activity detection model is adopted to carry out refined slicing on the audio stream, and noise filtering and muting are carried out; text decoding is carried out through a speech recognition model subjected to fine tuning of synthetic data, and term enhancement in professional fields is supported; carrying out multi-modal fusion and time sequence calibration on an identification result to realize sound and picture synchronization and text standardization; and finally, rendering subtitles through a real-time callback mechanism and a cross-platform interface. The method solves the problems of high delay of long video subtitle generation, time axis drift and low terminology recognition rate, is especially suitable for Linux / Windows cross-platform localization deployment, and has high privacy security and low-cost field adaptive capability.

The invention provides a butterfly-shaped cross main tail inductor for a voltage-controlled oscillator, relates to the technical field of phase noise optimization of the voltage-controlled oscillator, and aims to realize phase noise optimization of the voltage-controlled oscillator on the premise of not occupying more area. The butterfly-shaped crossed main and tail inductor comprises a main inductor formed by winding a splayed metal wire, a tail inductor formed by winding a splayed metal wire, two metal interlayer through holes, two main inductor ports and two tail inductor ports; the main inductor is used for providing an inductance value required by resonance of the LC voltage-controlled oscillator; the tail inductor is used for filtering noise generated by the tail current source and improving the impedance of a common mode point, so that the overall phase noise performance of the voltage-controlled oscillator is improved; and the metal interlayer through hole, the main inductor port and the tail inductor port are used for providing inductor wiring and circuit interfaces. The voltage-controlled oscillator is simple in structure, phase noise can be reduced without improving the design of the voltage-controlled oscillator, meanwhile, occupied area is reduced, and the voltage-controlled oscillator has very high cost performance.

The application discloses an OFDM transmission method and system of a marine channel. The method can include: a transmitting end transmits a signal, converts the signal into a time domain through an IFFT, eliminates interference, converts the signal back into a frequency domain through an FFT, and obtains a frequency domain signal; divides the frequency domain signal into useful data and pilot data, and performs channel estimation on the pilot data; performs equalization, parallel-serial conversion and demodulation on the channel estimated signal and the useful data, and a receiving end obtains a final receiving signal. The application can effectively filter noise interference through an improved channel estimation method.

The invention relates to the field of confocal microendoscopes, in particular to a fiber core positioning method and device of a confocal microendoscope and a terminal. The method comprises the steps of obtaining multiple frames of optical fiber end face images; fusing and filtering the plurality of frames of optical fiber end face images to generate a fiber core response image; positioning a fiber core bundle estimation center of the fiber core response image; diffusing from inside to outside based on the fiber core bundle estimation center, and positioning the position of a first target fiber core; and removing the pseudo fiber core points which are close to each other. In a calibration stage or a fiber core tracking stage, firstly, fusion processing is carried out on collected multi-frame optical fiber end face images, noise in the images is filtered, filtering is carried out on the fused images, the position of a fiber core is highlighted, interference is filtered, then an estimation center of a fiber core beam is positioned through a distance clustering method, and the fiber core is tracked. Therefore, the position of each fiber core is diffused outwards from the estimation center, the anti-interference capability is remarkably improved, the positioning precision is high, the response speed is high, and the imaging performance and reliability are improved.

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.

The application provides a ground penetrating radar-based underground ore body positioning and identification method and system, and relates to the technical field of ore body detection.The method comprises the following steps: inputting a corrected multi-dimensional point cloud feature set into a pre-trained closed-loop convolutional neural network, optimizing network parameters through transfer learning, and outputting a category confidence of an ore body target and a spatial position probability distribution point cloud; based on the spatial position probability distribution point cloud, combining electromagnetic wave double-path travel time calculation rules and Cagniard resistivity measurement results, and generating an ore body depth and horizontal positioning point cloud through a reverse projection algorithm; and performing multi-scale two-dimensional empirical mode decomposition on the reverse projection positioning point cloud, filtering noise components to extract a high-confidence target point cloud.The application improves the accuracy and stability of ore body identification.

The invention discloses an aerodynamic characteristic correction method for a wind turbine airfoil wind tunnel test in the field of wind turbine airfoil aerodynamic characteristic optimization. The method comprises the following steps: 1) constructing a wind turbine airfoil aerodynamic test platform for the wind tunnel test; 2) carrying out safety check on equipment and an experimental environment; (3) measuring the surface pressure of the wind turbine airfoil under the expected working condition by using a high-frequency pressure sensor to obtain a measured pressure signal; 4) performing frequency domain analysis on the pressure signal to obtain the power spectrum density of the pressure signal of the wind turbine airfoil under the expected working condition; 5) estimating and identifying vortex shedding frequency; (6) pressure signal pulsation and measurement errors caused by the vortex are distinguished, a Kalman filter is designed to filter the pressure signals, and corrected pressure signals are obtained; 7) estimating a pressure signal of a specific pressure measuring point position; and 8) calculating to obtain the lift coefficient and the resistance coefficient of the corrected wind turbine airfoil. According to the method, the vortex shedding frequency can be accurately identified, the noise can be effectively filtered, and the correction method of the real aerodynamic coefficient can be obtained.

A method (100) for filtering noise in a signal, the method performed by a transmitter, a first device, and a second device in a telecommunications network. The method comprises the transmitted splitting the signal into a first and a second part and transmitting the first and second part to the first and second device. The first device passes the first part through a polarization rotator configured using a key shared with the second device. The key comprises indications for which polarization to select for each segment of the first part of the signal. The first device passes the first part through a first polarization splitter, the first polarization splitter configured using the key. The second device passes the second part through a second polarization splitter, wherein the second polarization splitter is configured using the key. A delay element is used to synchronize the first and second device.

The invention discloses a sleep monitoring and intelligent early warning system and method based on a multi-modal time sequence signal, and the system comprises a multi-modal sensor module which is used for collecting an original heterogeneous time sequence signal of a subject in a sleep period; the time sequence data preprocessing module is used for performing timestamp alignment, band-pass filtering noise reduction, baseline drift correction and sliding window segmentation on the original heterogeneous time sequence signals and extracting multi-modal feature vectors; the sleep feature recognition module is used for receiving the multi-modal feature vector, performing deep feature learning and outputting sleep staging probability distribution and a sleep apnea event detection mark; the large language model module is used for performing joint coding on an output result, a historical sleep file of the subject and a clinical diagnosis text to generate a structured sleep quality assessment report and health risk association analysis; and the multi-level intelligent early warning and decision support module is used for triggering early warning signals of different levels and generating personalized sleep intervention suggestions.

The invention discloses an unmanned AGV power transmission system vibration diagnosis monitoring method, device and equipment, and relates to the technical field of equipment state monitoring. Collecting state monitoring data of a plurality of driving transmission parts of the unmanned AGV by using a vibration monitoring device; performing working condition identification and noise filtering on the state monitoring data by using a density-based clustering algorithm so as to filter noise and other working condition data except the normal working condition to obtain normal working condition data; and carrying out feature extraction on the normal working condition data, inputting the extracted multi-dimensional feature data into a Gaussian mixture model to obtain a fault level early warning result, inputting the multi-dimensional feature data into a fault diagnosis model based on a convolutional neural network to obtain an unmanned AGV fault type diagnosis result, and carrying out visual display on the results. According to the invention, weak and transient fault features can be accurately captured, accurate fault grading early warning and fault diagnosis are realized, and the operation reliability and the operation efficiency of the unmanned AGV are improved.

This application discloses a method, system, device, and equipment for processing point cloud data of a lidar. The lidar includes multiple lasers, including a first laser and at least one second laser. A polarizer is provided on the laser emission path of the first laser, and an analyzer is provided on the receiving optical path of the corresponding channel of the first laser inside the lidar. The method includes: acquiring first point cloud data corresponding to the first laser and second point cloud data corresponding to the second laser; calculating the distance between each second target point in the second point cloud data and its multiple neighboring points to filter out at least one second target point whose distance is greater than a preset value, thereby obtaining updated second point cloud data; and determining the first point cloud data and the updated second point cloud data as the final point cloud data acquired by the lidar. This method can improve the accuracy of detecting and filtering noise in the point cloud data while reducing the computational load.

The application relates to the technical field of computer vision, and proposes a dual-domain pseudo-label generation cross-domain target detection method based on DETR. A multi-scale decoding query clustering module can accurately capture the feature information of targets of different scales by performing multi-scale clustering on the query output by the decoder, thereby effectively evaluating the confidence of the pseudo-label. The method has strong cross-domain migration capability, can generate higher-quality pseudo-labels between target domain and source domain style data, optimizes the confidence of the pseudo-labels, and further improves the accuracy and robustness of target detection, significantly reducing the missed detection and false detection problems of pseudo-labels of different scales. The dual-domain cooperation and dual-domain verification module combines the pseudo-labels under different confidence levels of the source domain and target domain style images, enhances the reliability of the pseudo-label generation, effectively filters noise, improves the adaptability of cross-domain data, ensures that the model can smoothly migrate between different fields, and further improves the detection precision and generalization capability.