36 results about "Mean-shift" patented technology

The invention provides a digestive tract tumor detection method and system, relates to the field of medical image detection, and solves the technical problem of low accuracy of tumor boundary detection and automatic positioning in a digestive tract tumor image acquisition process in the prior art. The method comprises the following steps: taking a pan-tumor region image as a training sample of a binary decision tree algorithm, constructing an objective function based on the training sample to obtain a first objective function value, and obtaining a depth threshold value and an optimal classification point of the training sample; optimizing the optimal classification point based on the candidate feature sound to obtain a second objective function value; an unknown digestive tract tumor image is accessed, image points of the digestive tract tumor image are marked through a trained binary decision tree algorithm and a regression forest algorithm, the position distribution probability of the image points is generated, a position distribution thermodynamic diagram is generated based on the position distribution probability, and a tumor core area is obtained through mean shift. And completing the separation of the tumor region and the non-tumor region. The application is used in the digestive tract tumor detection process.

The application provides a kind of spatio-temporal intelligent marine satellite internet of things sensing information screening method and device, belong to data processing technical field, this method is based on convolution bidirectional long short-term memory neural network realizes marine environment time series data repair, based on block time series transformer carries out marine environment data space-time prediction and based on mean shift density clustering algorithm carries out marine hotspot area mining, the marine environment data includes sea surface temperature, salinity, wave height and other sensing data, based on BP neural network carries out marine environment risk assessment and multi-agent reinforcement learning carries out information screening feedback control, meet the complex marine environment detection scene efficient information processing demand, realize low redundancy, high-precision data acquisition transmission adaptive feedback control effect.The present application can be directly applied to marine internet of things in marine observation detection facility networking system, can provide strong support in military application and civil application, has wide and important application prospect and value.

The application provides a remote sensing target detection sample sampling method and device based on active learning. The method comprises the following steps: performing difference judgment on a plurality of first multi-level features and a plurality of second multi-level features based on a feature difference mechanism, and obtaining difference distance information; obtaining stability results based on a sample consistency mechanism and a second data set; performing weighted summation on the difference distance information and the stability results to obtain uncertainty results of the second data set; obtaining a target category and a target category quantity of a first data set, and obtaining a category sampling weight according to the target category and the target category quantity; generating a sampling score of each second sample in the second data set according to the category sampling weight and the uncertainty results; performing clustering processing on the second samples based on a Mean Shift algorithm to obtain clustering clusters; and taking the second samples with sampling scores meeting requirements in each clustering cluster as target remote sensing sampling samples. The time cost of target detection sample acquisition is reduced.

The present application relates to a kind of cluster collaborative control method based on continuous mean shift theory, belong to involve multi-agent distributed collaborative control field, including three parts: desired configuration figure input and contour fitting, target shape physical parameter determination, distributed configuration control algorithm design based on continuous mean shift theory.The present application adopts Bezier curve to express desired shape, has strong representation ability to complex non-convex shape, target shape data storage is less, reduces memory requirement, and continuous control law can avoid the oscillation possibly caused by grid-based discrete control law.The present application uses distributed computing mode, and multi-agent system can enter desired shape area according to local perception information and actively explore unknown shape area, effectively reduce the requirement to agent communication ability, increase engineering feasibility, have strong robustness and scale expandability.The present application has strong adaptive capacity, and does not need global label and initial position configuration.

The invention relates to the technical field of physiological data processing, in particular to a patient physiological data monitoring method and system based on the Internet of Things. The method comprises the steps that multiple pieces of historical physiological data and physiological data of a target patient are acquired, all the historical physiological data are clustered through mean shift clustering, and multiple clusters are obtained; and calculating the classification possibility of the physiological data belonging to each cluster, and selecting the cluster corresponding to the maximum value in the normalized values of the classification possibility as the category of the physiological data so as to complete the classification of the physiological data. According to the scheme, classified monitoring of the physiological data, collected in real time, of the target patient can be timely and accurately carried out.

The invention discloses a model dynamic compression method and system based on a Mean Shift clustering and SVD decomposition algorithm, electronic equipment and a storage medium. The method comprises the steps that S1, a convolutional neural network model subjected to sparse training is acquired; s2, performing clustering and pruning on the convolutional neural network model after the sparse training by using a Mean Shift clustering algorithm; s3, carrying out Finning training on the convolutional neural network model after clustering and pruning; s4, pruning operation is carried out on the convolutional neural network model after finishing the Finishing training based on an SVD decomposition algorithm; s5, after pruning operation is completed, whether the parameters meet the threshold value requirement or not is judged till a final model is output. After the dynamic clipping scheme is added, the three problems which are originally left can be effectively solved: 1, the size of the model is effectively reduced, so that the model can be deployed in equipment with a relatively small memory; 2, the model parameters can be effectively cut according to the requirements of a hardware deployment platform; and 3, for the pruned model, the performance similar to that of the original model can be effectively ensured.

The invention discloses a multi-extended target joint tracking and classification method based on star convex RHM and LMB filters, and belongs to the field of radar target tracking. According to the method, an LMB parameter set is initialized by using target prior information, and then a sensor measurement set is divided through a mean shift algorithm; then, state prediction is achieved by combining survival target parameter updating and new target parameter sampling, and then measurement updating is completed through LMB-to-GLMB, GLMB updating and GLMB-to-LMB; and then trimming and fusing the LMB parameter set, estimating the number of targets, extracting state information, and circularly executing until observation is finished. According to the method, a star convex RHM modeling expansion state is adopted to reduce dimensions, the low detection probability / high clutter scene performance is improved based on an LMB framework, the tracking classification effect and the real-time performance are both considered, and the engineering application value is high.

The invention discloses a doubly-fed wind power plant equivalence method based on a mean shift algorithm, and the method is specifically implemented according to the following steps: building a doubly-fed wind power generation grid-connected system model, obtaining the power data during the normal operation and fault period of a unit, and determining a grouping index; the grouping index is preprocessed, then the grouping bandwidth is calculated, then data of each wind turbine generator set serves as an initial grouping center, and initial grouping of the wind turbine generator sets is completed through a mean shift algorithm; introducing a Devensenberg index (DBI) as an evaluation index, and optimizing the preliminary grouping result to obtain the number of equivalent units; and calculating equivalent unit parameters by adopting a capacity weighting method according to an optimized grouping result to obtain a wind power plant equivalent model. According to the method, grouping fits the real characteristics of the unit, and the data deviation of the grouped unit is reduced.

The invention discloses a multipath clustering and tracking method for a dynamic wireless channel, which belongs to the technical field of wireless communication, and comprises the following steps: acquiring time delay, angle and power parameters of multipath components in each time snapshot; running an adaptive neighborhood robust mean shift clustering algorithm, and combining a neighborhood structure quality index to realize static division of a non-uniform density cluster; constructing a directional bounding box of the cluster in a three-dimensional feature space by utilizing principal component analysis, and accurately representing a geometric boundary of the cluster; executing cluster-level tracking based on a directional bounding box and Kalman filtering, and correcting a prediction state; a cluster-level joint similarity is defined based on the shape and density of the bounding box to determine the birth and death of the cluster, and the remaining multipath is re-clustered to identify a new cluster. According to the method, the problems of over-segmentation or under-segmentation and association fuzziness in the prior art are solved. According to the method, the variable density structure of the multipath cluster and the spatio-temporal evolution trajectory of the variable density structure can be adaptively represented, and the precision and robustness of channel modeling in a high-dynamic non-stationary scene are remarkably improved.

This invention relates to the field of image processing technology and discloses an adaptive smoothing and enhancement method for animation capture trajectories based on temporal feature fusion. The method includes the following steps: S1, acquisition and preprocessing of raw trajectory data; S2, extraction of multi-dimensional temporal features; S3, noise and signal identification based on gated feature fusion; S4, generation of adaptive smoothing kernel and trajectory repair; S5, trajectory enhancement and output: post-processing based on physical constraints is performed on the smoothed trajectory to enhance the dynamic rationality of the action, and a smoothed and feature-enhanced animation capture trajectory is output. This invention intelligently distinguishes between real high-frequency motion and noise by fusing kinematic features, spatial correlation features, temporal semantic features, and attention weights, avoiding the over-smoothing distortion phenomenon of traditional methods. Through a multi-scale feature fusion method optimized by mean drift, key motion regions are automatically identified and assigned higher detail preservation weights, achieving differentiated smoothing enhancement.

The present application relates to the technical field of data processing, more particularly, the present application relates to a network risk assessment method for industrial internet, comprising: acquiring network traffic value and network bandwidth value of industrial internet in real time; determining potential abnormal factors at the current time; determining algorithm bandwidth demand degree for self-adaptive correction of fixed bandwidth in mean shift clustering algorithm at the current time; determining adaptive bandwidth at the current time; clustering by using mean shift clustering algorithm to realize network risk assessment of industrial internet. The present application avoids two extreme problems caused by fixed bandwidth by dynamically adjusting bandwidth, so as to more accurately identify abnormal traffic; according to real-time network traffic value, historical data and potential abnormal factors, the optimal bandwidth suitable for the current situation is calculated, the clustering parameters are adjusted according to the network state, and more flexible and agile risk detection is realized.

The invention discloses a carpet printing multi-level accurate positioning method based on image processing. The method comprises the following steps: S1, collecting and preprocessing a carpet image; s2, extracting pattern features by using a convolutional neural network, and performing classification through a random forest algorithm to obtain pattern feature tags; s3, performing mean shift clustering on the pattern feature tag, identifying a central point and a region boundary, and constructing a CNN-RF model to complete a pre-training process; s4, calculating a deviation according to a clustering result, correcting the error through a CNN-RF model, and adjusting positioning to realize pattern alignment; s5, designing and applying a positioning avoidance scheme, and optimizing a pattern alignment relationship; s6, the alignment deviation in the printing process is monitored, and the positioning strategy is adjusted in real time; and S7, adjusting clustering parameters and feature classification rules, and realizing accurate positioning of dynamically optimized pattern levels. The carpet printing multi-layer pattern alignment precision is improved, manual intervention is reduced, and the production efficiency and the product quality are improved.

This invention discloses a method and system for sample augmentation and lightweight deep learning model construction in power distribution line defect identification. The method includes: performing adaptive contrast stretching preprocessing on the input image and stitching normalized coordinate grids to enhance input information; constructing a lightweight detection model with two branches, one based on segmentation and the other on row classification, where the two branches share a partial feature extraction network and are fused through a feature interaction module to simultaneously output pixel-level segmentation results and row classification prediction results; employing a feature-response distillation method based on channel attention to transfer knowledge to a more streamlined student model; using a binary segmentation map to filter instance embedded features and performing mean-shift clustering to obtain independent instance segmentation results, which are then fused with row classification predictions for verification and supplementation; finally, deploying the compressed model to an embedded device to achieve real-time defect identification and localization. This invention effectively improves the detection accuracy and efficiency of the model in embedded environments.

The application discloses a kind of based on spatial data mining and geological constraint seismic activity fault model construction method, system, equipment and medium.The method is first arranged and analyzed to seismic data, extracts minimum complete subdirectory;Using improved mean shift algorithm for spatial clustering, identify each active fault corresponding small earthquake cluster;Subsequently, each fault cluster is three-dimensional slice, and least square method fitting is generated fault interpretation line;With interpretation line as foundation, initial seismic activity fault three-dimensional model is constructed, and is optimized by multi-source geological constraint, and active fault three-dimensional fine model is obtained;Finally, the model is integrated with digital earth spatial registration.This application can realize the whole process automation and quantization from seismic data analysis to active fault three-dimensional fine model construction and application, improve the objectivity, efficiency, precision and reliability of model construction, provide important technical support for seismic disaster risk assessment, active fault detection.

The application discloses a radar super-resolution forward-looking imaging method and system, constructs a weight vector and a manifold matrix in the distance and azimuth dimensions according to the waveform parameters and the array structure of the radar, and further constructs a range spread function and an azimuth spread function.On this basis, target features are extracted through mean shift theory, and the constructed range spread function and azimuth spread function are used when calculating a mean vector, so that the diffusion mode of the target in the radar image is more in line with the diffusion mode, an image with a smaller full width at half maximum can be obtained, and super-resolution imaging is realized.The application innovatively considers the super-resolution imaging problem from the perspective of the image domain and in combination with the characteristics of the radar.Compared with the previous super-resolution method based on signal processing, the application is more universal, is not easily limited by factors such as the waveform and the array structure, and has no strict performance upper limit, so that the resolution can be further improved on the basis of the existing super-resolution imaging result.

The application discloses a power distribution room inspection robot positioning method based on image recognition, and comprises the following steps: establishing an inspection point position database and an equipment position database to form an initial association; collecting multi-modal inspection data and preprocessing to obtain a multi-source observation sequence; constructing an improved bidirectional state space model to obtain a pose state sequence; determining an inspection point position and generating a point position mapping key; obtaining a clustering center set based on a mean shift clustering algorithm; performing repositioning correction and outputting an inspection point position control instruction; and performing offline inspection and local caching to generate an inspection record and an abnormal event record. Through the stable pose estimation of the improved bidirectional state space model and the repositioning correction of the mean shift clustering, the application realizes stable positioning of an inspection point in a complex environment of a power distribution room.

The invention discloses a three-dimensional molecular structure aided design method based on a neural network and text description. After a reference data set is constructed, a text-2D structure alignment model used for aligning text description and a two-dimensional molecular graph, a 2D-3D noise structure alignment model used for aligning the two-dimensional molecular graph and a three-dimensional molecular graph and a prior model used for converting the text description into the two-dimensional molecular graph are sequentially trained; through training the conditional diffusion model, the isodenaturation of the three-dimensional molecular structure on translation, rotation, reflection and arrangement is maintained, and through the trained contrast learning model and the trained conditional diffusion model, based on the mean shift of text gradient guide conditional diffusion model sampling, the three-dimensional molecular structure which has expected properties and accords with text description is generated. And after post-processing, the aided design is completed. According to the method, the three-dimensional molecular structure is generated from the beginning of text description, and compared with the prior art, the chemical effectiveness, atomic stability and molecular stability indexes of generated three-dimensional molecules are improved.

The application relates to a parallel fast label multi-Bernoulli filtering method under a phased array radar system and belongs to the field of radar target tracking. The application comprises the following steps: obtaining a measurement set of a radar, dividing the measurement set by adopting a mean shift algorithm, performing plot condensation on the divided measurement set, performing parallel joint prediction and updating based on the plot condensed measurement set, including LMB state transition, target and measurement grouping, parallel processing of the grouping results, and merging of the group posterior density, pruning and truncating the LMB parameter set after the parallel joint prediction and updating, and estimating the target number and state based on the pruned and truncated LMB parameter set. Through grouping of the target and the measurement, parallel processing of the FLMB filter is realized, the calculation amount is effectively reduced under the condition that the tracking performance loss is small, the real-time performance is improved, and the target detection speed and accuracy are improved.

The invention discloses a customer label clustering analysis method based on semantic embedding, comprising the following steps: S1, collecting original text data of customer labels, and preprocessing to generate a standardized label text data set; s2, executing semantic embedding operation, and generating a label semantic embedding vector data set; s3, the improved sparse coding model is used for dictionary learning, and sparse feature representation is obtained; s4, performing mean shift clustering analysis to obtain a tag clustering grouping result; s5, generating customer label groups and customer portraits according to a label clustering and grouping result; and S6, outputting customer label grouping information for customer management, precision marketing or label recommendation. According to the method, through fusion of semantic embedding, sparse coding and improved mean shift clustering algorithm, high-precision automatic grouping and personalized portrait generation of customer tags are realized.

The embodiment of the present application provides a kind of abnormal target object detection method of dike based on infrared and visible light image fusion, for identifying the dike target object required to be detected in dike disaster scene, method includes: obtaining infrared image and visible light image;Infrared image and visible light image are registered using feature matching algorithm;Registered infrared image and visible light image are fused using image fusion algorithm to obtain fusion image;The following segmentation optimization operation is iteratively executed until the calculated value of the objective function of mean shift algorithm is minimum, to obtain the image segmentation result of fusion image: the fusion image is segmented using mean shift algorithm to obtain current segmentation result, the calculated value of objective function is determined according to current segmentation result, the hyperparameter of mean shift algorithm is optimized according to the calculated value of objective function.This scheme can accurately segment small size dike target object and can overcome the problem of over segmentation.

The invention discloses a defect multi-classification detection method based on feature clustering and an improved UNet model, and the method comprises the steps: carrying out the morphological enhancement of a test defect segmentation result predicted by a defect detection model after unsupervised training, obtaining an enhanced segmentation image, extracting defect features through a RoIAlign pooling method, and carrying out the detection of the defect features. Clustering each defect feature by using a Mean Shift algorithm to obtain a clustering result containing a plurality of defect category pseudo labels, wherein the clustering result is used for indicating the defect category pseudo labels corresponding to each defect enhancement region; training the improved UNet model based on each test sample image, each defect enhancement region in the test sample image and a corresponding defect category pseudo tag; and utilizing the trained improved UNet model to predict a to-be-detected defect fine segmentation result containing a plurality of defect category areas based on a to-be-detected image of the end face of the to-be-detected shaft metal part. On the basis, multi-classification fine segmentation of the end face defects of the shaft type metal parts can be achieved without data labeling, and the defect segmentation precision is improved.

This invention relates to the field of image data processing technology, specifically to a visual inspection method for precision hardware fittings used in loading and unloading containers. The method includes: acquiring pixels in the edge regions of each hole area in an image of the surface of the hardware fittings; performing mean-shift clustering on the resulting angle-grayscale value sequence based on the angle and grayscale value of the pixels; adjusting the drift vector of the mean-shift clustering using the characteristic angles formed between pixels and the degree of anomaly; using the adjusted drift vector to complete the clustering of the angle-grayscale value sequence, obtaining several data segments corresponding to the hole areas; obtaining the defect probability of the data segments based on the grayscale value change characteristics; and using the defect probability to complete defect detection. This invention significantly improves the accuracy of mean-shift clustering, further enhancing the accuracy of defect detection in hole areas.

This invention relates to the field of vision and image processing technology, and discloses a text compliance detection method based on image semantic segmentation. The method acquires and grayscales a text image, obtains an initial mask through semantic segmentation, calculates the local pixel mean drift coefficient, and determines the dynamic edge search bandwidth accordingly. Within the bandwidth, it statistically analyzes the magnitude and distribution characteristics of the first-order spatial gradient, calculates the rectified edge probability score of the target pixel, constructs a topological gravitational potential energy using the probability score as the gravity source, calculates the adhesion blocking coefficient by solving the Hessian matrix determinant and combining it with the global drift variance, then extracts the potential energy maximum point as the stroke center, extracts the cross-section along the potential surface normal, performs weighted integration, and calculates the equivalent stroke width. Finally, it calculates the width dispersion and combines it with the global drift mean to calculate the compliance discrimination coefficient, outputting the detection result. This technical solution effectively overcomes semantic drift and false edge traps, and accurately decouples local ink topological adhesion.

This invention relates to the field of power distribution system condition assessment technology, and discloses a comprehensive assessment method and system for the operating status of power distribution systems. The method includes: dividing historical data into time period subsets based on default time period characteristics that characterize the thermodynamic state of equipment; independently constructing load loss models for each time period subset to obtain a healthy baseline model cluster; after acquiring real-time data, calling the model corresponding to its assigned time period to calculate the normalized loss residual, and applying multidimensional statistical signature analysis of mean drift, instantaneous mutation, and increased volatility in parallel to the residual sequence for comprehensive assessment. This invention utilizes time periods as proxy variables for thermal state, thereby avoiding interference from equipment thermal inertia on the assessment, no longer relying on ambient temperature data, and obtaining a purer characteristic signal reflecting the health status of equipment.

The application provides a speed spectrum automatic picking method based on MeanShift clustering analysis, comprising the following steps: step 1, pre-processing artificial picking speed control points; step 2, training an initial constraint model of a work area speed; step 3, cleaning a speed spectrum according to the speed constraint model; step 4, picking a time-speed pair based on the MeanShift method; step 5, updating and optimizing the work area speed constraint model according to the speed picking result in step 4; step 6, repeating steps 3 to 5 until an iteration condition is reached; and step 7, outputting a speed spectrum automatic picking result. The speed spectrum automatic picking method based on MeanShift clustering analysis can effectively reduce the labor and time cost consumed in manual picking, and further improve the precision of speed automatic picking by introducing a three-dimensional speed field nonlinear multivariate regression model based on DNN as a constraint model for pre-processing the speed spectrum.

The application discloses a target tracking method based on an improved CamShift algorithm, and relates to the field of visual target tracking. The method comprises the following steps: acquiring image data, and pre-processing the image data; using a target detection algorithm to detect the target of the pre-processed image data, so as to obtain a tracking target; tracing the historical track of the tracking target, using a Lagrange interpolation method to generate a track prediction window, and using a Kalman filtering algorithm to correct the track prediction window, so as to obtain a corrected track prediction window; improving a continuous adaptive mean shift algorithm through a search window dynamic adjustment strategy, and using the improved continuous adaptive mean shift algorithm to track the tracking target based on the corrected track prediction window, so as to obtain a tracking result. The one-dimensional interpolation polynomial is constructed through the Lagrange interpolation method, and the tracking performance of the continuous adaptive mean shift algorithm on high-speed or nonlinear targets is effectively improved.

The application discloses a kind of field wheat stem tiller number extraction method, comprising the following steps:First, the sample of wheat stem tiller number and point cloud data are collected, and the data is preprocessed, in data preprocessing, introduce Kalman filtering algorithm, to remove redundant noise, improve initial point cloud data quality;Second, after data is voxelized, the mathematical relationship between porosity and voxel point cloud density is used to interpolate missing voxels in the canopy, to reduce the influence of occlusion on the algorithm, to obtain relatively complete canopy point cloud information;Finally, the mean shift algorithm is used to cluster the interpolated canopy, and the cluster number is the stem tiller number.The wheat stem tiller number extracted by the method is compared with the field measured stem tiller number, and the feasibility of the algorithm is verified.The application solves the noise and occlusion problem in the application of ground-based laser radar to a certain extent, and provides technical support for the estimation of other growth parameters by ground-based laser radar in the future.

The invention relates to the technical field of meteorological data processing and artificial intelligence, in particular to a local air temperature hundred-year homogenization day sequence processing method based on multi-source data fusion, which comprises the following steps: a multi-source data preprocessing stage: integrating core air temperature data, non-natural factor metadata and natural factor data to construct a data set; preliminarily screening abnormal windows based on an LSTM time sequence prediction model, and obtaining potential sections of breakpoints through multi-window consistency verification; the method comprises the following steps: generating an expected natural air temperature sequence by using a GBRT natural factor simulator, calculating a residual error to eliminate a natural fluctuation false breakpoint, identifying a single or coupled unnatural factor through multi-dimensional statistical characteristics such as mean shift and variability, and forming an enhanced breakpoint list containing influence factor types, weights and confidence coefficients; and performing correction by adopting a differentiation model or a weight combination according to factor types, and outputting a final homogenized sequence. According to the method, the breakpoint judgment accuracy and correction precision are improved, and high-reliability hundred-year air temperature data support is provided for climate change research.

The embodiment of the present application provides a kind of collaborative robot target tracking method, device and computer readable storage medium, the method comprises the following steps: from the real-time image frame of the camera device of collaborative robot transmission extraction predetermined starting image frame as current image frame;Actual hue and actual saturation of image in target tracking frame are established two-dimensional joint histogram according to;Two-dimensional joint probability map is obtained by the inverse projection of two-dimensional joint histogram;Optimized probability map is obtained by using adaptive filter to two-dimensional joint probability map spatial filtering;With the next image frame of current image frame update current image frame;Based on the mean shift algorithm model, the target tracking frame is re-determined in the updated current image frame according to optimized probability map and jumps back to the previous step.This embodiment can effectively provide the accuracy of target tracking.

A method, apparatus, non-transitory computer readable medium, and system for image generation includes obtaining an input image and an input prompt. In some cases, the input image depicts a scene and the input prompt indicates a target element to be added to the scene. The image generation model generates a normalized output based on the input image and the input prompt by performing a channel shift on a preliminary output of the image generation model. A synthetic image is generated including the scene of the input image and the target element of the input prompt that is harmonized with the scene.