25results about How to "Improve distinguishability" patented technology

The invention provides a self-adaptive short-circuit fault current-limiting control method and system for a primary and secondary fusion circuit breaker, and relates to the technical field of circuit protection and manufacturing, and the method comprises the steps: collecting a current value, a voltage value and a state quantity of a circuit breaker, and forming a first operation parameter set; performing feature extraction based on the amplitude abrupt change, the phase offset and the harmonic component, generating a first feature vector, and matching the first feature vector with a preset short-circuit fault feature template to obtain a fault recognition result, a fault type and a position identifier; a dynamic current-limiting control curve is calculated according to the identifier and the current value, a current-limiting control parameter set is generated, and a current-limiting execution mechanism is controlled to obtain current-limiting response data; after the action is completed, a second operation parameter set is collected, current limiting effect analysis is executed to generate state feedback data, and a current limiting control parameter set is updated based on the feedback data; and factory calibration can be carried out on the power electronic current limiting module serving as a current limiting execution mechanism. According to the method, the adaptive capacity and accuracy of current limiting control are improved.

This application relates to the field of industrial data processing and intelligent control technology, and discloses a method for dynamic modeling and prediction of industrial big data combined with digital twins. The method includes: acquiring operating status data, process execution data, control command records and prediction results, and constructing a sequence of control action events; segmenting the operating status data according to the sequence of control action events to form natural operating data segments and intervention operating data segments, extracting distribution characteristics, fluctuation characteristics and response characteristics, and generating a segmented behavioral feature set; calculating the control reflexive influence quantity and generating a reflexive drift marker, dividing the natural evolution dataset and the control feedback dataset, and predicting them separately; generating control risk constraint quantity, screening, delaying or limiting the control commands to be executed, and iteratively correcting the dataset according to the execution results; this method can distinguish between control feedback changes and natural evolution changes, reduce drift misjudgments, and suppress multi-round oscillations and parameter fluctuations.

The invention discloses a key control detection method and device for a gamepad and a medium, and relates to the technical field of man-machine interaction, and the method comprises the steps: building a mapping table of a key acoustic reference file and a key region, and carrying out the continuous sampling of the vibration acoustic response of a key surface structure during the operation of a user, generating a key acoustic sampling frame set attached with a unified time identifier; comparing the key acoustic sampling frame set with a key acoustic reference file, identifying a key drop point area, and aggregating to generate a key drop point event set; and analyzing an acoustic response signal corresponding to the key drop point event set, judging a target shearing direction through an energy migration track and a phase change mode of the acoustic response signal, and binding the target shearing direction with the key drop point event set to generate a key control event set. According to the method, the over-complete dictionary associated with the key drop point area identifier is constructed based on the key acoustic reference file, and the consistency and real-time performance of complex control instruction analysis are enhanced.

The embodiment of the present disclosure relates to a kind of parking lot point cloud feature extraction method, device, equipment and medium, wherein, method includes: obtaining the point cloud data and image of parking lot;From the point cloud data, the point cloud data of parking line is extracted, and the parking number is extracted from the image;According to the relative position relationship between the three-dimensional points contained in the point cloud data of parking line, the point cloud data of single parking line is separated from the point cloud data of parking line;According to the distance between parking number and single parking line in three-dimensional space, the parking number with the distance less than preset distance from single parking line is assigned to the point cloud data of single parking line.The scheme provided in the embodiment of the present disclosure can improve the distinguishability of parking line point cloud data, and provide accurate and reliable data basis for parking lot high-precision map mapping.

The invention discloses a titanium steel composite plate defect type identification method based on image identification, and relates to the technical field of image processing, and the method comprises the steps: inputting a standardized image into a pre-trained self-supervised feature extraction network, carrying out comparative learning, rotation prediction and shielding prediction tasks, extracting plate surface texture and microdefect features, and carrying out the fusion, generating a high-dimensional feature vector; performing spatial feature analysis on the high-dimensional feature vector to generate a candidate box set; inputting the candidate box set into a pre-trained multi-stage cascade identification network, and sequentially carrying out rough classification, common defect fine classification and rare defect enhancement identification to generate defect types; superposing the defect type to the standardized input image to generate a defect annotation image; and correcting the defect annotation image to obtain correction information, and performing fine adjustment on the pre-trained self-supervised feature extraction network and the pre-trained multi-stage cascade recognition network to form closed-loop optimization. According to the invention, the distinguishability of micro-defects and rare defects is enhanced.

The invention relates to a distributed multi-station radar three-dimensional imaging method assisted by Beidou high-precision positioning information. The method comprises the following steps: acquiring and returning ship high-precision motion information; transmitting an orthogonal radar signal, receiving mixed echoes and then separating the mixed echoes; motion phase compensation is carried out on the echoes by using Beidou information; generating a high-resolution three-dimensional image by adopting a time domain imaging algorithm; and automatically estimating the ship size and spacing based on the three-dimensional image. According to the invention, the high-precision position and speed information provided by the Beidou terminal of the target itself is obtained and is used as an external reference to carry out accurate phase compensation on the echoes received by the distributed multi-station radar so as to correct the phase error introduced by the motion of the target. According to the method, the problems of image defocusing and insufficient resolution caused by the movement of the target during radar imaging of the moving target in the prior art are solved, high-quality three-dimensional imaging and parameter extraction of ships moving in dense channels are realized, and the channel monitoring capability under severe weather conditions is remarkably improved.

This invention relates to a multi-field-of-view coding method and system based on intermediate image plane polarization mask modulation, belonging to the field of space optics technology. It addresses the technical problems in existing technologies where feature aliasing and mutual interference easily occur when multi-field-of-view signals are superimposed, and where encoded features are easily distorted in the presence of optical errors or system deviations. The coding method includes the following steps: multi-field-of-view division; polarization coding parameter allocation; polarization modulation implementation; optical folding and signal superposition; and coding result output. The multi-field-of-view coding method and system based on intermediate image plane polarization mask modulation of this invention, by introducing polarization modulation in the imaging stage, maps multi-field-of-view spatial information to the polarization dimension, realizing an extension from traditional "scalar intensity coding" to "vector polarization coding."

The application discloses a kind of accurate trigger synchronous multichannel vibration data acquisition system, comprising: data acquisition and pre-processing module, for reading data and generating vibration state sample and action state pair;Mutual inhibition reserve pool echo state network module, for executing dynamic feature extraction to vibration state sample;Event sensitive trigger gate module, for generating gated activation vector and timing screening is carried out to reserve pool activation state, output event response feature vector;Reverse memory drive DRQN model module, for receiving event response feature vector and executing forward recursion and reverse recursion, output Q value estimation and policy action;Error training module, for calculating the Q value error of policy action and generating training signal;Experience replay module, for constructing priority experience replay queue based on Q value error and selecting training segment to update DRQN model parameter.The application realizes the high-precision trigger and synchronous acquisition of multichannel vibration signal, improves the stability of feature and policy decision performance.

The invention provides an all-vanadium redox flow battery pump fault detection method based on a self-adaptive GCN, and the method comprises the following steps: carrying out the operation experiment of an all-vanadium redox flow battery in a plurality of operation states, collecting a battery voltage signal and an electrolyte flow signal in the experiment process, and marking the corresponding battery state as a training set; performing feature extraction and feature fusion on the battery voltage signal and the electrolyte flow signal to obtain a multi-modal node feature matrix of the all-vanadium redox flow battery; the Euclidean distance between the nodes is calculated to serve as the similarity between the nodes, a similarity matrix is constructed, neighbors with the highest similarity are screened out for each node from the similarity matrix, and an adjacent matrix is constructed; constructing a graph convolutional neural network model based on the adjacent matrix, and introducing an adaptive smoothing factor used for dynamically adjusting the smoothing strength of neighborhood features during volume accumulation; and inputting to-be-detected battery operation data into the trained graph convolutional neural network model to obtain a battery pump fault diagnosis result.

The application provides an enhanced generation method based on topology-aware graph coding and adaptive subgraph retrieval, including multi-granularity semantic-aware segmentation and knowledge graph construction, structure feature matrix construction and explicit topology position coding generation, constrained subgraph diffusion and dynamic pruning based on semantic-topology joint scoring, and graph-text consistency loss constraint generation based on attention matrix structured alignment; in the coding layer, explicit topology position coding is used to avoid the overhead and excessive smoothing caused by online GNN aggregation; in the retrieval layer, connected subgraphs instead of fragmented nodes are used as enhanced context; in the alignment layer, the internal attention matrix of the large language model is directly constrained to be consistent with the subgraph adjacency matrix, the structured guidance is realized from the reasoning mechanism level, so that the logical illusion is inhibited and the multi-hop reasoning accuracy is improved.

This invention discloses a target trajectory background modeling method and system based on bidirectional difference and grayscale correction, belonging to the field of image data processing technology. The method includes: S1, periodically acquiring basic image trajectory data and performing preprocessing on the acquired data; S2, constructing a frame index sequence, generating a frame pairing table and a grayscale coefficient table, performing grayscale correction and bidirectional difference calculation on the paired frame sequence to obtain the candidate motion mask for the current frame; S3, obtaining the initial background region of the current frame, evaluating the background confidence level of each pixel, and generating an effective background region; S4, comprehensively evaluating the stability level and fluctuation constraint capability of the background, dividing the stable and weak background regions, and outputting the frame-level motion mask to complete the background closed-loop output. This solves the problem that traditional background modeling methods directly use arithmetic mean modeling in a few-frame sequences, leading to the background grayscale being lowered by high-contrast moving pixels, thus generating motion artifacts and causing weak target detection failure.

The application relates to a small target chicken positioning method based on double-domain enhancement and dynamic priori, which comprises the following steps: performing multi-scale feature extraction on a breeding image, a chicken image and a background image respectively, performing double-domain enhancement on shallow features and generating breeding coding features; prototype construction is performed on chicken image features and background image features to generate chicken feature prototypes and background feature prototypes, and similarity enhancement and background suppression are performed; based on foreground scores and foreground k A content embedding and refined position priori of an object query are generated by selecting a high response feature point, and are input into a DETR decoder to output the boundary box coordinates and the category score of the chicken target. The application can alleviate the small target missing detection problem caused by the smoothing of shallow features, reduce the false detection risk caused by background interference, improve the small target positioning ability of the traditional DETR model which is limited by the lack of target area position priori in object query initialization, and effectively improve the positioning detection precision.

This invention discloses a network traffic anomaly detection method based on the fusion of SSM and Transformer, belonging to the field of network traffic monitoring technology. The method includes the following steps: S1, construction of raw traffic observations; S2, construction of joint feature representations; S3, dual-view sample generation: constructing raw view samples and generating corresponding perturbation view samples; S4, multi-scale variable behavior fragment decomposition: performing fragment decomposition on the raw view samples and perturbation view samples at different time scales; S5, hybrid coding feature extraction: extracting intra-session dynamic features using the SSM modeling branch and extracting inter-session correlation features using the Transformer modeling branch; S6, normal behavior representation learning; S7, traffic anomaly determination. This invention captures features at different time granularities through a multi-scale patch hierarchical structure, achieving high-precision and high-stability time-series anomaly detection.

The invention relates to the field of optical fiber sensing, and discloses an optical fiber sensing signal diagnosis method based on TimesNet two-dimensional time-frequency transformation, which comprises the following steps: S1, inputting an original signal, and carrying out adaptive time-frequency denoising through learnable wavelet packet transformation; s2, extracting a dominant frequency component based on the enhanced signal, and constructing a two-dimensional time-frequency tensor through zero padding and remodeling; s3, using depth separable convolution to extract features, using grouping convolution to capture local space-time coupling, using point-by-point convolution to fuse abnormal features, and combining a SimAM attention mechanism to enhance the weight of a key area; s4, modeling time sequence evolution through minLSTM, and realizing parallel long-range dependence modeling by adopting a logarithm domain correlation scanning algorithm; and S5, performing joint training through a gradient conflict coordination mechanism based on the shared features by using a multi-task output head, and outputting fault detection, type classification and positioning results at the same time. According to the method, two-dimensional orthogonal expression is constructed through physical prior dimension raising and noise reduction, and the signal-to-noise ratio, the feature interpretability and the diagnosis precision are remarkably improved.

The present application relates to the technical field of intelligent analysis of flavonoids, and particularly relates to an intelligent analysis method for electrochemical recognition of flavonoids. The method comprises the following steps: obtaining a response signal matrix in time-frequency dimension, and performing frequency domain compression and nonlinear interaction mapping to obtain a one-dimensional nonlinear compressed current response sequence; introducing a variable window integral singular response recognition mechanism based on the one-dimensional nonlinear compressed current response sequence to calculate a singular response strength; determining a singular response occurrence point based on the singular response strength, and performing weighted fusion on a local window tensor; performing asynchronous reconstruction processing on the obtained singular weighted fusion tensor to obtain a characteristic representation of the flavonoid target after asynchronous reconstruction processing, and performing intelligent analysis on the flavonoids. The method solves the problems of low resolution, poor stability and repeatability of traditional electrochemical recognition analysis methods for flavonoids, and the inability to obtain characteristics of various target objects in flavonoids, thereby improving the analysis accuracy.

The application discloses a kind of target-oriented photovoltaic receiver light communication and energy integration system limited dynamic range constraint emission waveform design method, first generate MLS sequence, simultaneously to the original signal of input PAM modulation obtains PAM sequence, PAM sequence is as slow envelope to MLS sequence mean envelope modulation, and the alternating current signal after envelope modulation is obtained, direct current bias B is superimposed in the alternating current signal to constitute the final emission waveform.The application B determines reference light intensity and effective AC budget.In available budget, communication level interval and perception MLS amplitude are determined by amplitude allocation factor ρ together.Under fixed AC budget, ρ determines the trade-off between communication and perception, and B determines the upper limit of feasible dynamic range.The application makes full use of limited amplitude resources, realizes low-bandwidth photovoltaic receiver-oriented efficient information extraction, accurate echo perception and stable energy transmission, and realizes flexible coordination of communication and energy functions.

This disclosure relates to the field of wearable electronic device technology, and discloses a smart ring and its reminder control method. The smart ring includes a ring shell, a health monitoring sensor disposed on the inner surface of the ring shell, at least two electrodes, an electrical stimulation driving circuit electrically connected to the electrodes, and a control circuit. The control circuit determines the user's physiological state based on physiological monitoring data collected by the health monitoring sensor, and when a preset target physiological event is determined to occur, controls the electrical stimulation driving circuit to output a micro-electrical stimulation sequence corresponding to the target physiological event through the electrodes to provide a reminder to the user. Therefore, health monitoring results can be promptly fed back to the user on the ring itself in a low-power manner, improving the timeliness and distinguishability of the reminder.

The application discloses a transformer production quality online detection method and system, and the method comprises the following steps: clamping a transformer to be detected and reading a unique identifier, measuring the primary terminal contact resistance and the secondary terminal contact resistance, and re-measuring if the contact resistance is out of limit; applying a bipolar high-slope pulse between the primary and the secondary, synchronously sampling the voltage, the response current and the reference noise, and aligning the sampling time; setting two levels of gate windows with the pulse edges, coherently superimposing the response current in the inner window, and calculating the offset amount according to the reference noise to obtain the net response in the outer window; calculating the cross-correlation peak value, the gated energy and the frequency band energy ratio from the net response, and generating the first index in combination with the batch reference. The application realizes the synchronous online identification of insulation hidden troubles and connection hidden troubles under the production line condition, and forms a stable judgment and a full-flow tracing closed loop.

The invention discloses a method for searching and identifying road diseases by using persistent coherence, and relates to the technical field of road defect detection. According to a topological data analysis method based on persistent coherence, path finding is carried out by using persistent coherence. Comprising the following steps: step 1, zooming an image again; 2, finding out a topological structure by using persistent coherence; step 3, bonding the area near the position; and 4, analyzing the shapes and sizes of different segmented regions to detect and identify the road disease. According to the topological fingerprint generation method based on continuous coherence, a novel and powerful technical path is provided for solving the essential feature extraction problem of a two-dimensional object, and path diseases can be found and recognized more accurately.

Embodiments of the present application provide a robot positioning method and device, computer equipment, storage medium and computer program product, and relate to the technical field of computer vision. The method comprises: obtaining a current temperature and point cloud fusion feature matrix; encoding the temperature and point cloud fusion feature matrix to obtain a current encoding feature vector; obtaining a relative position change matrix of a target robot between a previous time step and a current time step and a temperature feature difference matrix of the target robot between the previous time step and the current time step according to a historical encoding feature vector of the previous time step adjacent to the current time step and the current encoding feature vector; and obtaining target current pose information of the target robot at the current time step according to the relative position change matrix and the temperature feature difference matrix. The method improves positioning accuracy, path planning efficiency and accuracy, and map updating accuracy of the corresponding space.

The application discloses a kind of network construction methods for extracting space-time features of dynamic gesture RGB video frame, belong to computer vision and deep learning field.Aiming at the problems that existing network weakly resists blur, hand small target extraction effect is poor, gesture frame offset and time sequence redundancy influence training precision, the application constructs single mode space-time feature extraction network based on ShuffleNetV2-GRU.Spatial branch adds learnable gaussian blur branch, adopts SPDConv optimization down sampling, Stage4 layer is replaced by lightweight DCNv2 to extract hand irregular features, and multi-scale feature fusion is completed by WAFF-ASFF.The time sequence branch accesses TAM module between spatial network and GRU to correct frame offset, and relies on adjacent frame weighting module to strengthen key time sequence information.Preprocessing distinguishes trajectory, gesture posture and makes differential enhancement, and combines hand ROI to optimize input data.The application has excellent extraction precision, model lightweight, inference is fast, and is suitable for embedded device deployment.

The invention discloses a nematode edge auxiliary semantic segmentation method based on adaptive contrast. The method aims at achieving accurate nematode segmentation. The method comprises the following steps: S1, preprocessing a nematode microscopic image, dividing the nematode microscopic image into a training set, a test set and a verification set, and inputting the preprocessed image into a dynamic contrast regulation and control module DCRM to generate a contrast enhanced image; s2, inputting the enhanced image into an adaptive boundary saliency extraction module AEPD, extracting an edge saliency map and an edge mask, and calculating edge saliency loss; s3, inputting the shallow features and the edge information into an edge-assisted semantic enhancement module ESEM, fusing the edge-guided shallow features and the multi-scale semantic features, and generating an enhanced semantic graph; and S4, executing query sampling based on the enhanced semantic graph and predicting a final segmentation mask to obtain a segmentation result of the nematodes or eggs. According to the method, the nematode segmentation precision is effectively improved through edge-assisted semantic segmentation.

The application relates to the technical field of image processing, in particular to a clinical ultrasonic image auxiliary screening system based on deep learning. The system comprises an ultrasonic image preprocessing module, a lesion region segmentation module, an image lesion labeling module and a lesion auxiliary screening module, can acquire clinical ultrasonic images and carries out gray processing and neighborhood gray balance processing, so as to generate clinical ultrasonic balanced images; carries out pixel normalization and lesion region segmentation based on tissue boundary blur on the clinical ultrasonic balanced images, so as to generate clinical ultrasonic image lesion region blocks; carries out image lesion labeling through the clinical ultrasonic image lesion region blocks, so as to generate a clinical ultrasonic lesion labeling image set; constructs an image lesion auxiliary screening model and carries out lesion auxiliary screening, so as to output the lesion position, lesion type and lesion severity corresponding to the lesion existing on the clinical ultrasonic image. The application can realize accurate analysis of clinical ultrasonic images, so as to assist in efficiently completing lesion screening.

The invention discloses a multilevel radar signal sorting method and equipment based on pulse parameter cascading. The method comprises the steps of receiving radar pulse signals, constructing corresponding three-dimensional pulse description words, performing normalization processing, mapping the radar pulse signals into data points in a three-dimensional feature space, and calculating Euclidean distances to establish a distance matrix; calculating the local density of each pulse data point based on K neighbor and natural neighbor, introducing a relative local abnormal factor, and identifying and eliminating false alarm pulses; k-nearest neighbor search is executed again, the DBSCAN algorithm is adopted for primary pulse sorting, the neighborhood radius is constructed based on the third quartile and the quartile distance of the K-nearest neighbor distance, and the minimum point number threshold value is K value; based on inter-cluster K neighbor correlation and boundary natural neighbor connectivity, verification merging is carried out; and according to a preset PRI search range and step length, carrying out two-dimensional plane mapping on the arrival time sequence of each cluster pulse after combination, constructing a comprehensive judgment criterion to identify a real PRI and extract a pulse sequence, and obtaining a sorting result.

This invention discloses an acoustic imaging device and method for detecting pavement voids, belonging to the field of non-destructive testing technology for roads. The device includes an excitation module, an accelerometer, a multi-aperture reconfigurable microphone array module, a synchronous acquisition module, a sound field feature construction module, a multi-source fusion module, a geometric inversion module, and a coordinate registration and imaging output module. The excitation module applies a transient impact to the pavement under test. The accelerometer acquires the acceleration signal at the excitation point, and the multi-aperture reconfigurable microphone array module acquires multi-channel near-field sound pressure signals around the excitation point. The sound field feature construction module determines the impact initiation time and impact energy based on the acceleration signal at the excitation point, and performs time alignment, energy normalization, and feature extraction on the multi-channel near-field sound pressure signals. This invention improves the stability and quantitative characterization capability of pavement under-slab void detection, and is suitable for detecting under-slab voids in cement concrete pavements, composite pavements, and bridge deck paving.