90 results about "Multi feature fusion" patented technology

This invention relates to the field of remote sensing image processing technology, specifically disclosing a building change detection method based on a dual-branch encoder and multi-feature fusion. By modeling the building body and edges separately, this invention elevates edge information to an equal level of importance with body information, providing a new design approach to solve the long-standing problems of boundary adhesion and ambiguity in building change detection, and facilitating the handling of irregular boundaries. A feature cross-fusion module effectively promotes the fusion of building integrity, ensuring the semantic consistency and accuracy of the results. In the decoder section, a hierarchical feature fusion noise removal module maximizes the identification and removal of abnormal image patches. This invention achieves efficient and high-precision building change detection by inputting the acquired dual-temporal remote sensing image (including the preceding and following temporal images) into the constructed building change detection model and outputting the building change detection results.

This invention discloses a fuel tank anti-theft detection method and system based on differential and multi-feature fusion recognition. The method includes: executing a fuel tank anti-theft detection algorithm within a time recognition window to determine abnormal fuel tank opening behavior; if multiple consecutive time recognition windows determine abnormal fuel tank opening, an alarm message is generated and sent out. The advantages of this invention are as follows: 1. The dual-sensor differential structure can distinguish between overall vibration and local structural displacement; 2. Data collected by dual sensors can effectively distinguish between overall vehicle vibration and local abnormal fuel tank cap opening behavior; wherein, direction-sensitive differential can enhance the recognition ability of the opening direction movement, and multi-feature fusion recognition can significantly reduce the false alarm rate and improve recognition accuracy; 3. It provides self-learning suppression alarm, which can improve environmental adaptability; 4. It achieves long-term low-power operation through hierarchical power consumption management, suitable for scenarios without external power supply.

This invention belongs to the field of QR code printing control technology, specifically disclosing a variable QR code printing control method. This method includes: importing order parameters containing QR code printing position, content type, and application scenario characteristics; real-time acquisition of the flatness and colorimetric features of the substrate surface; quantifying and generating integrity assurance coefficient and clarity maintenance coefficient based on flatness features and printing position; quantifying and generating color stability coefficient and readability maintenance coefficient based on colorimetric features; comparing and verifying each quality indicator with quality indicator thresholds dynamically generated based on order parameters; if verification passes, printing is executed; otherwise, printing parameters are dynamically adjusted before printing is executed. This invention, through multi-feature fusion and dynamic verification adjustment, significantly improves the printing quality, physical durability, and visual readability of QR codes in complex outdoor environments, and overall enhances the adaptability of the QR code printing process.

This invention discloses a multi-feature fusion time series prediction method, system, device, and medium for power load peak prediction, belonging to the field of power system load prediction technology. It includes: acquiring and preprocessing raw power load time series data; extracting and constructing a load peak time series for feature construction to obtain a prediction feature set; constructing supervised learning samples to transform the time series prediction problem into a regression problem; and training the supervised learning samples using a machine learning model to obtain a load peak prediction model for predicting and evaluating future load peaks. The beneficial effects of this invention are as follows: Through a feature engineering method of multi-feature fusion, this invention can effectively capture the historical dependence, short-term volatility, and long-term periodicity in power load time series, thereby significantly improving the accuracy of load peak prediction. It has a predictive advantage, especially for power systems with large load fluctuations, in environments with large-scale integration of new energy sources.

This application provides an intelligent image comparison method and related equipment for the hydrophobicity of synthetic insulators using multiple features, belonging to the field of power equipment testing technology. The application includes: acquiring multiple reference standard images and multiple images of insulators to be tested; performing image preprocessing on the reference standard images and the images of insulators to be tested respectively to obtain multiple preprocessed reference images and multiple preprocessed images to be tested; constructing multiple image pairs; for each image pair, extracting multiple image features and calculating a set of single-feature similarities corresponding to the multiple image features; performing weighted fusion on the set of single-feature similarities based on a dynamic weighted fusion algorithm to obtain a comprehensive similarity score for the image pair; determining the hydrophobicity level of the insulator image to be tested relative to the reference standard image in the image pair based on the comprehensive similarity score of each image pair, and outputting the comparison result. This application can improve the accuracy of judgment through multi-feature fusion and comprehensive comparison of multiple reference samples.

This invention discloses an automatic identification method for PCB debug interfaces based on multi-feature fusion, comprising: establishing electrical connections between the debug interface pins of the PCB under test and a parameter measurement unit and a programmable protocol generator respectively through a multiplexed switch matrix; performing static electrical feature measurements on the debug interface pins using the parameter measurement unit to obtain electrical feature data; comparing and matching the electrical feature data with a pre-stored interface feature database to generate at least one candidate interface type hypothesis; automatically configuring a corresponding protocol excitation signal for each generated candidate interface type hypothesis, and applying the protocol excitation signal to the corresponding candidate pin through the multiplexed switch matrix; monitoring the response signal of the candidate pin under excitation, and confirming or excluding the candidate interface type hypothesis based on whether the response signal conforms to the expected protocol specification of the current candidate interface type; and outputting the finally confirmed interface type and its corresponding pin definition.

The application discloses a kind of instance segmentation and target detection hybrid identification method based on multi-feature fusion, belong to computer vision technical field.The method constructs multi-task fusion network;Through the feature extraction of image to be measured of main network, shared fusion feature map M1 is obtained by multi-scale feature pyramid fusion;It is respectively input into target detection and instance segmentation module, and detection result and segmentation result are obtained;Then feature map M1, target detection result and instance segmentation result are jointly input into posture estimation module, and posture estimation result is obtained;Three kinds of results are uniformly mapped and aligned in space, are spliced and weightedly fused in channel dimension, are input into fully connected layer learning task weight combination, and the identification result containing target class, spatial position, pixel-level segmentation mask and key point coordinate is output.The application is collaboratively designed by feature sharing, multi-task fusion and cross-frame matching mechanism, to improve system operation efficiency while ensuring identification accuracy.

The application discloses a zero-intervention blade tip timing signal extraction method based on multi-feature fusion, and belongs to the technical field of non-contact vibration testing of rotary machines. The application firstly constructs a candidate time window to intercept a pulse section based on a key phase signal; secondly, four types of arrival time features, i.e. a peak value moment, an energy barycenter moment, a cross-correlation peak value moment and a fitted zero-crossing point moment, are extracted in parallel in the candidate window; then, local signal-to-noise ratios and other signal quality indexes are calculated, feature fusion weights are adaptively determined, and a fused arrival time is obtained; finally, a median absolute deviation is used for outlier rejection, and a stable and reliable arrival time sequence is output in combination with adjacent blade time interval consistency verification. The application can significantly improve the arrival time extraction precision and robustness under the conditions of low signal-to-noise ratio, waveform distortion and working condition fluctuation, and is suitable for blade health monitoring and online vibration measurement in a closed casing environment of an aero-engine, a gas turbine and the like.

The application discloses a kind of based on multi-feature fusion and model learning's electromagnetic detection defect qualitative and quantitative method, builds multi-modal electromagnetic detection system, acquires the multi-frequency eddy current of measured component / multi-channel magnetic flux leakage signal and lift-off data, after adaptive denoising, interference suppression and normalization preprocessing, extract four categories 44-dimensional multidimensional features, obtain optimal feature subset by three-level screening strategy;Adopt channel attention and cross attention mechanism to realize multi-feature adaptive fusion, build "shared backbone+double branch" dual-task joint learning model, complete transfer pre-training, fine-tuning and lightweight compression;Finally output defect qualitative and quantitative results, support dual-technology result fusion and risk rating.The application realizes the full-depth coverage detection of special equipment metal components from surface to middle-deep defects, qualitative classification accuracy is greater than or equal to 98%, depth quantitative average relative error is less than or equal to 5%, single-frame inference time is less than or equal to 10ms, which can be directly deployed on handheld devices to realize real-time detection on site.

This application discloses a method for processing and recognizing image data from a scope, relating to the field of image data processing technology, including the following steps: S1, preprocessing the raw image data acquired by the scope, extracting the RGB values, grayscale values, position information, and gradient features of pixels to obtain preprocessed image data; S2, based on the preprocessed image data, establishing a dynamic background environment model by fusing pixel gradient features through multi-Gaussian distribution, comparing pixel features of newly acquired image data, and decomposing the image data into a background environment part and a suspected target part. This application, through dynamic background modeling, grayscale and contour multi-feature fusion, dual-criteria similarity matching, and trajectory verification mechanisms, can accurately distinguish targets from environmental interference under harsh conditions such as low light, dust, and complex mountain forest backgrounds, effectively filtering out isolated noise and background fluctuations, and significantly reducing the probability of misidentification and missed identification.

The application discloses a kind of based on multi-feature fusion's gas turbine rub-impact fault diagnosis method, step one, the running data of gas turbine is collected in real time by multiple sensors;Step two, feature extraction is carried out to vibration data, and the key time-frequency features related to rub-impact are calculated;Step three, for speed data, the rub-impact energy index is calculated according to slight collision energy model;Step four, feature extraction is carried out to temperature data, and whether temperature anomaly appears is judged according to temperature feature condition;Step five, the weight of each feature is adjusted, and the weighted algorithm is used to fuse multiple features;Step six, the fault decision threshold is determined using ROC curve, and whether the fault occurs is judged;The application can improve the precision of gas turbine rub-impact fault diagnosis, monitor the running state of potential machine in real time, give early warning to abnormal condition, and adjust the operating condition of gas turbine according to the diagnosis result, so as to reduce the probability of rub-impact fault.

This invention discloses a clutter false alarm suppression method based on multi-feature fusion, comprising: acquiring radar echo signals and performing pulse compression processing to obtain time-domain echo data; performing clutter suppression and coherent accumulation on the time-domain echo data to obtain a range-Doppler matrix; performing constant false alarm detection on the range-Doppler matrix to obtain several potential target points; extracting the multi-pulse echo signals and unsuppressed frequency domain signals corresponding to the potential target points from the time-domain echo data, extracting the corresponding clutter-suppressed frequency domain signals from the range-Doppler matrix, and constructing multi-dimensional features for each potential target point based on the extracted signals; inputting the multi-dimensional features into a pre-trained classifier, determining whether the potential target point is a target or false alarm clutter based on the output results, and outputting the final target point trace. This invention effectively suppresses clutter false alarms in complex environments and improves radar detection reliability through multi-dimensional feature fusion and secondary decision-making by the classifier.

The application relates to the technical field of logistics, and particularly discloses a comprehensive sorting and identifying method and system for express items based on multi-feature fusion, initial feature entry is completed in a unified distribution warehouse, visual features of the items are collected by a feature collection terminal at multiple angles, a unique feature fingerprint is generated after binding a user ID, and the unique feature fingerprint is stored in a database; end identification is completed in a regional unmanned relay station, and includes three links of warehouse entry identification, shelf mounting, user notification and user pick-up; and a return and replacement processing flow and an unmanned pick-up flow are further arranged. The application discloses the cumbersome operation of abandoning the traditional bar code multiple scanning, realizes 'one-time collection and whole-process identification', and the item warehouse entry and pick-up links do not need manual intervention for code scanning, so that the efficiency of logistics sorting and end distribution is greatly improved; the unified distribution warehouse side of the application is fine and configured to guarantee feature collection accuracy, the relay station side is simplified and configured to control cost, and the application can be flexibly adapted to various logistics scenes such as the unified distribution warehouse, the regional unmanned relay station and the intelligent express cabinet, and has high practicability.

This invention relates to the field of nondestructive testing technology, solving the problem of large thickness assessment errors caused by the combined effects of the cladding layer lift-off effect and the shielding effect of the metal protective layer in existing technologies. Specifically, it relates to a pulsed eddy current thickness assessment method based on multi-feature fusion. This method extracts multi-dimensional feature parameters such as centroid time, information diffusion energy, and information diffusion width as inputs, and performs normalization and weighted fusion on each feature to construct a fused feature. This fused feature comprehensively characterizes the time distribution, energy distribution, and diffusion characteristics of the signal, achieving high-precision quantitative assessment of the wall thickness of clad equipment. This invention uses multi-dimensional features such as centroid time, information diffusion energy, and information diffusion width as inputs, and comprehensively characterizes the signal through feature normalization and weighted fusion methods. It fully utilizes the overall information of the pulsed eddy current signal, improving the accuracy and stability of wall thickness assessment, thereby achieving accurate measurement of the wall thickness of clad pipes.

This invention relates to the field of roadbed engineering construction scheduling technology, specifically to an intelligent scheduling method and system for roadbed filling and excavation operations based on multi-source data. The method includes the following steps: A multi-feature fusion network based on an attention mechanism and implicitly constrained by the scheduling decision space obtains global situational awareness features based on the semantic segmentation layer, road network state matrix, and dynamic safety risk index distribution map; a spatiotemporal graph convolutional network is used to model the dynamically changing vehicle-road network system, and combined with mechanical efficiency profiles and vehicle predicted trajectories, a transportation topology efficiency map representing the overall capacity and bottlenecks of the current transportation network is obtained; engineering rules are introduced as a reward function, and scheduling decisions are based on multi-agent collaborative reinforcement learning, combining engineering scheduling units, global situational awareness features, and the transportation topology efficiency map to generate optimal scheduling instructions. This invention achieves precise, dynamic, and collaborative scheduling of roadbed filling and excavation operations, effectively improving scheduling efficiency.

The application is based on a scientific research scheme evaluation method and system based on multi-feature fusion, comprising: constructing a data set; the data set comprises personal information, ability and research idea of the evaluated person; the data set is encoded to obtain a personal information feature vector, an ability feature vector and a research idea feature vector; the extracted personal information feature vector, ability feature vector and research idea feature vector are fused, and after fusion, evaluation prediction is carried out to obtain the ability representation of the evaluated person; the data set covering personal information, ability and research idea is systematically constructed, high-dimensional feature vectors are extracted through deep feature coding technology, and an advanced fusion mechanism is adopted to realize effective interaction of multi-source features, thereby improving the accuracy, objectivity and automation level of scientific research idea evaluation, and the quality of the final work or the evaluation result in the preliminary idea stage is predicted, thereby reducing unnecessary resource investment.

The application discloses an IGBT residual life prediction method based on multi-feature fusion and KPCA optimization. The method first collects IGBT collector current and voltage and other multi-source signals, extracts time domain, frequency domain, time-frequency domain and derived statistical domain features after pretreatment; then adopts a two-stage strategy of comprehensive evaluation index preliminary screening and mutual information regression fine screening to eliminate redundant features and retain high correlation features. On this basis, nonlinear dimension reduction is carried out by using kernel principal component analysis to construct a high-robustness health index, and the exponential weighted moving average and adaptive gradient detection are combined to accurately divide the degradation into three stages. Finally, the CNN-BiLSTM model is used to deeply mine the time sequence degradation features, and the MC Dropout algorithm is introduced to realize the accurate prediction and uncertainty quantification of the residual life. The application effectively solves the one-sidedness of single-source signal representation and the feature redundancy interference problem, and significantly improves the prediction accuracy and reliability.

The application relates to a soil organic matter prediction method based on multi-feature fusion, and belongs to the technical field of soil organic matter prediction. The method comprises the following steps: pre-processing acquired soil data, obtaining spectral features, performing time domain reconstruction on frequency domain signals, and obtaining time domain data; obtaining an optimal delay time and an optimal embedding dimension based on the time domain data, and performing phase space reconstruction to obtain a phase space trajectory; extracting chaotic features based on the phase space trajectory, and taking the extracted chaotic features, the optimal delay time and the optimal embedding dimension as final chaotic features; obtaining a vegetation index based on the spectral features and taking the vegetation index as an index feature; and inputting the spectral features, the final chaotic features and the index feature into a constructed double-flow low-rank interaction network model to obtain a soil organic matter prediction result. The application aims to solve the technical problem that the spectral features extracted by the prior art cannot comprehensively represent the complex nonlinear characteristics of soil, thereby leading to low prediction accuracy.

This invention discloses a method and system for inverting the nutrient content of eucalyptus canopy leaves based on multi-feature fusion, belonging to the field of UAV remote sensing and forestry information technology. The method includes: acquiring and preprocessing multispectral image data from a UAV; extracting vegetation index features and texture features of the eucalyptus canopy; filtering the fused features based on mutual information to construct a multi-feature fusion dataset; constructing an inversion model for the nitrogen, phosphorus, and potassium content of eucalyptus leaves using a random forest regression algorithm; and verifying and optimizing the accuracy of the inversion model. This invention effectively compensates for the information gaps of single features by fusing spectral and spatial structure information, significantly improving the inversion accuracy of eucalyptus canopy leaf nutrient content, and providing reliable technical support for precise nutrient management of eucalyptus plantations.

The application discloses a kind of based on RBM-SegNet's watershed boundary extraction method, device, medium and equipment, comprising: obtaining target area digital elevation model data and carrying out spatial resampling;With SegNet as base model, construct RBM-SegNet semantic segmentation model, its encoder uses residual network as main stem and introduces residual connection, encoder and decoder multilevel embedding bottleneck attention module, decoder specific upsampling stage is equipped with multiple feature fusion module;Four-channel topographic feature input data is handled using the model trained, and boundary probability graph is output and optimized, finally, the watershed boundary that conforms to topographic feature and hydrology rationality is obtained.The method reduces parameter dependence, improves boundary extraction accuracy and integrity under complex terrain.

The present application relates to the technical field of optical fiber sensor demodulation, and particularly relates to a multi-feature fusion sapphire fiber Fabry-Perot sensor demodulation method, system and device. The method comprises the following steps: collecting an interference spectrum signal, constructing a two-dimensional feature representation form and a frequency domain feature representing interference characteristics; inputting the two-dimensional feature representation form and the frequency domain feature into a feature mapping model; extracting an interference fringe of a specific interference cavity according to the feature mapping model, calculating a rough interference order of the interference fringe of the specific interference cavity according to the spectrum signal; calculating an accurate interference order according to the rough interference order, and inputting an accurate optical path difference into the mapping model to establish a mapping relationship between the environment pressure value and the accurate optical path difference. The present application introduces different dimensional spectral feature information, realizes stable extraction of target cavity interference information, and overcomes the problems of insufficient demodulation accuracy and robustness of existing demodulation methods under the conditions of multi-cavity interference, non-ideal spectrum and environmental disturbance.

This invention provides a method and system for earthquake magnitude detection based on deep learning and multi-feature fusion, belonging to the field of deep learning technology. The system includes a data preprocessing module and an earthquake magnitude detection module. The method includes: acquiring raw three-component earthquake waveform data; preprocessing the raw three-component earthquake waveform data to obtain preprocessed three-component earthquake waveform data; obtaining a training set from the preprocessed three-component earthquake waveform data; building a magnitude detection model; training the model until a set number of training rounds is reached, and saving the best-performing model from the trained rounds as the final magnitude detection model; and inputting the preprocessed three-component earthquake waveform data to be detected into the final magnitude detection model to obtain the magnitude detection result. This invention can calculate the magnitude in one step, without requiring multiple revisions or manual intervention, and improves calculation accuracy.

The application provides a lithium battery lithium precipitation detection method and system based on laser ultrasonic multi-feature fusion, and relates to the field of battery detection. The detection method comprises the following steps: controlling a laser generating device to irradiate a lithium battery sample with laser and obtaining time domain data of ultrasonic signals from the lithium battery sample; converting the time domain data into frequency domain data, fusing the time domain data and the frequency domain data through a cross attention mechanism to obtain ultrasonic fusion features; and inputting the ultrasonic fusion features into a trained lithium precipitation prediction model to obtain the lithium precipitation probability of the lithium battery sample. This detection method can realize non-contact nondestructive detection while improving the detection accuracy.

The application provides a multi-feature fusion network traffic classification method, system, device and storage medium, the method comprising: obtaining target traffic, the target traffic comprising multiple different types of traffic; inputting the target traffic into a network traffic classification model to obtain the traffic type in the target traffic; wherein the network traffic classification model comprises a multi-core convolutional neural network, a bidirectional long short-term memory network and a Transformer model, the multi-core convolutional neural network is used for extracting local features of the target traffic, the bidirectional long short-term memory network is used for extracting time sequence dynamic features according to the local features, and the Transformer model is used for obtaining attention features according to the time sequence dynamic features. The network traffic classification method in the embodiment of the application can effectively distinguish similar types of traffic, is effective in encrypted and obfuscated traffic classification, and has application potential in enhancing a network security system.

This invention discloses an intelligent classification method and system for fundus images based on multi-feature fusion, relating to the field of medical artificial intelligence technology. The method includes: acquiring fundus image data; constructing a fundus image classification model based on a Swing Transformer network; sequentially extracting local fundus features from the fundus image data in the EfficientNetV2-L network branch; segmenting the fundus image data into 8×8 fundus image blocks, segmenting them into 4×4 fine-grained local windows in the improved Swing Transformer network branch, and calculating the local attention of the fundus image blocks; mapping the continuous relative coordinates of each fundus image block to bias values ​​using a two-layer multilayer perceptron to determine the global fundus features; and performing cross-attention fusion of the global fundus features output from the improved Swing Transformer network branch and the local fundus features output from the EfficientNetV2-L network branch to obtain the fused fundus features.

This invention relates to the field of text processing technology and discloses a school name normalization method based on multi-feature fusion and NLP semantic understanding. The method includes obtaining the text of the school name to be normalized and the corresponding administrative division information; based on the administrative division information, selecting a set of candidate standard schools from a pre-constructed standard school feature library, where each record in the standard school feature library contains at least a standard school name, its administrative division code, and a semantic vector; and performing semantic matching between the semantic vector of the school name text to be normalized and the semantic vectors of each standard school name in the candidate standard school set. This invention aims to solve the problems of data statistics difficulties caused by non-standard school name writing and high mismatch rates due to simple character matching, achieving efficient, accurate, and secure school entity normalization.

The application provides a multi-feature fusion-based no-reference screen image quality evaluation method and system, which comprises the following steps: calculating the gradient amplitude, relative gradient amplitude and gradient direction mapping of the input screen content image, applying a local ternary pattern operator on the mapping for texture coding, combining the gradient domain values to statistically weight the coding modes, and generating a gradient-weighted local ternary pattern histogram; inputting the screen content image into a pre-trained deep convolutional neural network for feature extraction, obtaining the deep feature map at the end of the network, and converting the deep feature map into a global perception feature vector through a global average pooling operation; fusing the gradient-weighted local ternary pattern histogram and the global perception feature vector, inputting them into a neural network quality prediction model, and outputting the no-reference quality evaluation score of the screen content image after nonlinear mapping of the model. The application realizes accurate mapping from multi-dimensional features to quality scores.