42 results about "Nearest neighbor clustering" patented technology

The invention provides an automatic unstructured road boundary line extraction method based on a vehicle-mounted laser point cloud. The method comprises steps: scanning line segmentation is carried out; a topology relation network is built; a moving window method is adopted to acquire ground points; clustering is carried out through a K nearest neighbor clustering method, a pseudo ground point cluster is eliminated according to the number of laser points in the point cluster, and road surface laser point cloud data are acquired; the outer contour point of the road surface laser point cloud data is set to be at the position of the road boundary line, outer contour boundary point extraction is carried out on each scanning line, two ground points with the farthest distance in each scanning line are acquired, and outer contour points at two sides of the road are finally acquired; clustering analysis is carried out, and a left contour point set and a right contour point set of the road areacquired; and boundary line tracking optimization is carried out to acquire the optimal boundary line. Extraction on an unstructured road and a road boundary can be quickly realized based on the laserpoint cloud data direction, an effective solution is provided for automatic unstructured road boundary line extraction, and the automatic road boundary line extraction degree is improved.

A reconfigurable processor architecture. A reconfigurable processor is an array of a multiplicity of various functional elements, between which the interconnections may be programmably configured. The inventive processor is implemented on a single substrate as a network of clusters of elements. Each cluster includes a crossbar switching node to which a plurality of elements is connected via ports. Additional ports on the crossbar switching node connect to the switching nodes of nearest neighbor clusters. The crossbar switching nodes allow pathways to be programmably set between any of the ports, and any pathway may be set to be either registered or unregistered. The use of clusters of processing elements allows complete freedom of local connectivity for effective configuration of many different processing functions. Wide area interconnection is more restricted, but, since it is less used, does not significantly restrict configurability. The inventive processor thus provides 1) high configurability with a low cost of switching network overhead; 2) constant clock speed, independent of configuration; and 3) very high clock speed since all communication is local or nearest neighbor.

An improved system and method for clustering text or content described by text is provided. Each text in a set of texts may be represented as a dimensional vector of words. Singleton texts that may not be similar to another text may be excluded from the set of texts for clustering. Texts identified as good nearest neighbors may then be grouped in the same cluster. In addition, metadata describing content may be used for clustering items of aggregated content from content feeds. Metadata describing items of content from content feeds may be converted into a set of texts and texts identified as good nearest neighbors may then be clustered. Items of content feeds described by the clustered texts may then be similarly clustered. Any types of items of content that may be described by text may be clustered, including audio, images, video, multimedia content, and so forth.

The invention provides an electric power system payment channel evaluating method based on large data and belongs to the field of electric power systems. The method comprises the steps of: firstly based on investigation questionnaire data, adopting a K-means clustering algorithm to carry out figure analysis on user payment behaviors, and identifying corresponding relations between various kinds of user attributes and the various kinds of user payment behaviors; utilizing a K nearest neighbor classification algorithm to carry out nearest neighbor clustering on electric power system payment channel evaluation factors, using a clustering result as a payment channel evaluation index, and establishing a payment channel evaluation system; utilizing a simulated genetic annealing algorithm to determine the weight of the payment channel evaluation index so as to form a payment channel evaluation module; and for new areas or new periods, utilizing electric power system practical payment data to judge whether the payment channel evaluation module is suitable, and if not, establishing a new evaluating module. The invention fills a technical void in the electric power system field, the quantitative evaluation of the electric power system payment channel is realized, and a scientific decision making basis are provided for the planning, construction and reconstruction of the payment channel.

The invention provides a weld joint contour feature recognition and weld pass real-time planning method, and the method comprises the steps: obtaining a weld joint contour image in real time through a laser visual sensor, and carrying out the gray scale processing to obtain a gray scale image; building a visual attention model, processing the gray scale image through the visual attention model, obtaining a comprehensive saliency map, and extracting laser stripes through combining with the Otsu threshold segmentation method and the nearest neighbor clustering algorithm; segmenting the extracted laser stripes based on the section span of monotonous slope, and obtaining a monotonous slope section and a slope catastrophe point; finally formulating an algorithm according to the slope catastrophe point, and planning a weld pass. The method solves a problem that conventional weld pass planning is low in precision, and the method achieves the real-time planning of the weld pass.

The invention discloses a data mining method of the neural network based on nearest neighbor clustering. An improved nearest neighbor clustering learning algorithm is used to train the neural network, so that on the premise that the precision requirement is met, the number of nodes of a hidden layer is reduced, the network structure is simplified, the learning speed of the neural network is increased, and the learning efficiency and precision of the neural network are further improved; and further data mining is carried out on the neural network, and the mining efficiency for a database of a large practical power system via the neural network is further improved. The data mining method aims at ensuring the safe, stable, high-quality and economical operation of the power system, and satisfying higher requirements for mass data of the digital power system and reliability, consistency and sharing performance of data.

The invention discloses a GC-MS non-target method used for rapid screening of plant differential metabolites, and belongs to the technical field of flavours and fragrances, and metabonomics study technology. According to the GC-MS non-target method, modern GC-MS full scanning technology is adopted to obtain chemical ingredient information of volatile substances in plant; multidimensional Gaussian smoothing is adopted to extract substance information in a TIC picture; dynamic programming algorithm is adopted to calibrate chromatographic peak time shifting among samples, realize aligning of chromatographic peaks belongs to one same substance, achieve comparability of substance information among different samples; nearest neighbor-clustering algorithm is adopted to realize compound registration; and at last variance analysis is adopted to screen compounds with significant difference from different groups. According to the GC-MS non-target method, compound mass spectrum information is output so as to obtain a MSP file; and the MSP file can be input into the NIST database to search compounds. The GC-MS non-target method can be used for locking compounds with dissimilarities in plant samples rapidly, and is suitable for rapid analysis determination of a large amount of samples.

The invention provides a point cloud-based method and system of extracting building feature lines. The method comprises: extracting a plurality of planes from massive point cloud plane data; extracting two planes from the plurality of planes, and judging whether the two planes intersect; calculating and extracting an intersection line of the two planes if the two planes intersect; and judging whether two other intersecting planes exist in the plurality of planes, if yes, returning to the step of calculating and extracting an intersection line of the two planes, and otherwise, ending a process.According to the point cloud-based method of extracting the building feature lines provided by the invention, a shared nearest neighbor clustering algorithm is used for plane dividing of a building point cloud of three-dimensional ground laser scanning, the number of clusters does not need to be input in a clustering process, a method of plane intersection is used at the same time to extract building edge line features, a plane intersection distinguishing method is relatively simple, and edge line point coordinate calculation is accurate.

The invention relates to an intelligent control method of a steering engine electro-hydraulic loading system. The method comprises: a controller uses an ant colony clustering algorithm to obtain clustering information containing the clustering number and a cluster center; an RBF neural network uses the clustering number and the clustering center as the number of nerve cells of a hidden layer and a nerve cell center of the hidden layer, uses error information e and a system output signal y that are outputted by a force sensor and a displacement sensor as a pair input, and carries out on-line identification on a steering engine electro-hydraulic loading system by using a nearest neighbor clustering algorithm, thereby obtaining a neural network output; and a single nerve cell PID uses Jacobian information provided by an RBF neural network to carry out on-line self tuning on control parameters kp, ki, and kd by using a gradient descent method and then a loading force instruction signal is outputted to control movement of a steering machine. According to the invention, an on-line real-time control problem is solved; and thus the loading precision and response precision of the system are improved.

The invention discloses a real-time and stage theory line loss fast estimation method with the clustering technique adopted. The method comprises the steps of extracting a real-time fracture surface feature vector, calculating a center fracture surface weight coefficient vector matched with the real-time fracture surface feature vector, fracture surface clustering, and theory line loss calculation. According to the real-time and stage theory line loss fast estimation method with the clustering technique adopted, when data are collected completely, center fracture surfaces are extracted by means of the nearest-neighbor clustering technique, and then fracture surfaces with similar operating states in a power grid are gathered together; under the situation of data missing, a central moment close to the data missing moment is found out by means of the nearest-neighbor clustering technique, and transmission loss of the power grid at the moment is assigned to the data missing moment; therefore, stage and real-time theory line loss values can be obtained fast, stage theory line loss calculation is achieved, the speed of the stage theory line loss calculation is increased, and precise comparison between the statistics line loss rate and the stage and real-time theory line loss values is achieved.

The invention provides a density self-adapting non-uniform point cloud simplifying treatment method. The method comprises the following steps of: 1, performing K-near neighbor calculation of the whole point cloud; 2, performing self-adaptive and uniform re-sampling according to density information and curvature information of each point; and 3, setting an initial point cloud (D), outputting a simplified point cloud (FD), and working out the average density of the point cloud, which is a weighted average of the density of all points. In the method, the average density threshold value of the point cloud is set; the initial point cloud(D) is sampled by an uniform grid curvature adaptive sampling method so as to obtain a sub-point set(SD) of the initial point cloud(D); a similarity degree matrix S is obtained by working out the degree of similarity between points in the SD, and the u value of the points in the SD by indexing; by adopting a near neighbor clustering algorithm and using the S and u as AP algorithm inputs, a representative degree matrix and an adaptive degree matrix between the points are worked out; and the information between the points are updated by iteration until the target value is reached and the final point set(FD) is obtained. The method has the advantages of simplifying calculation, reducing occupied memory capacity and effectively simplifying non-uniform point cloud.

The invention provides a fast adaptive nearest neighbor clustering method based on a structured anchor graph. First of all, using K-Means algorithm generates representative anchors; Then, the initialsimilarity matrix is constructed for the original data points and anchor points. Then, the similarity matrix is updated iteratively by using the structured anchor graph technique, and the nearest neighbor assignment is performed adaptively. Finally, the clustering results are obtained directly according to the connected components of the graph corresponding to the final similarity matrix. The method of the invention reduces the dependence of the large-scale spectral clustering task on the initial similarity matrix weight, and can quickly obtain high-quality clustering results by iteratively optimizing the anchor graph structure.

The invention discloses a data-driven marine diesel engine fault detection method. The method comprises the following steps of clustering working modes of historical data of a diesel engine by adopting a k nearest neighbor clustering method to obtain a working mode type of the diesel engine; establishing a KPI model corresponding to each working mode by adopting a partial least squares regressionmethod PLS, and calculating an inspection threshold value; inputting real-time data to be detected; matching the working modes of the diesel engine by judging whether a distance between a real-time data point to be detected and a clustering center of the historical data is smaller than a preset threshold d or not; inputting to-be-detected data into the KPI model of the corresponding mode, and performing fault test statistics calculation to obtain KPI fault test statistics; and realizing data-driven marine diesel engine fault detection through comparison. By adoption of the technical scheme, real-time, wide-application-range and high-efficiency online fault diagnosis can be achieved.

The invention relates to a power distribution network multi-fault first-aid repair recovery method and system considering pre-disaster pre-scheduling. The method comprises the steps: determining the outage probability of an element, and generating an anticipated accident set; calculating a system direct-current power flow; sequentially quitting the load nodes in the anticipated accident set to judge whether the system has faults or not, and if so, merging the fault nodes into an accident determination set; judging whether all nodes in the system are judged to be finished or not, and if so, obtaining an accident determination set; if not, obtaining the information of the next node; determining a system fault position and a subarea to which a fault element belongs based on the accident determination set, and constructing a path scheduling scheme; determining a DG output prediction curve by adopting a DG output prediction method based on nearest neighbor clustering; and based on the DG output prediction curve, determining a power distribution network first-aid repair recovery scheme by constructing a multi-source collaborative optimization model. The fault first-aid repair efficiencycan be effectively improved, the first-aid repair sequence is optimized, the first-aid repair plan is adjusted in real time, and the power supply capacity of available resources of the power distribution network is optimized.

The invention relates to a clustering analysis method based on sample density and self-adaptive adjustment of a clustering center. Firstly, a clustering number K and the clustering center are initialized based on the sample density and the sample neighborhood, then the clustering number K is adaptively adjusted by adopting a nearest neighbor cluster merging method, and finally, the optimal clustering result is obtained, so that the effectiveness of the clustering analysis result is greatly improved.

The invention discloses an outdoor facility inspection method. According to the outdoor facility inspection method, a color image is read, and the image is converted into a grey-scale image; line segments in the image are calculated by employing probability Hof transform, and the line segments with length exceeding a set threshold are selected; the K near neighbor clustering algorithm is utilized, and the clustering result is traversed till clustered line segments can form a profile structure of equipment; a detection-needing area is selected; and a localized binary image in the area is calculated; a white connected component of the binary image is simplified through a binary image refinement algorithm to form a curve line segment; the binary image containing the curve line segments is traversed, points at end points of the curve line segments are extracted, and the points are marked at end point queues; the end point queues are classified and processed by utilizing a gradient histogram and a support vector machine, and an inspection result is acquired. The method can adapt to a real illumination situation and further has advantages of high detection accuracy, faster processing speed, high stability and relatively high noise resistance performance and inclination resistance performance.

The invention provides a density peak-core fusion-based adaptive clustering method, and belongs to the field of pattern recognition. The method comprises the following steps: (1) firstly, carrying outdensity neighbor clustering based on a density peak value: calculating the density of each data point in the to-be-clustered data set by using a core density-k-nearest neighbor density estimation method, determining density peak points through an adaptive threshold value, and performing density nearest neighbor clustering by taking the density peak points as a class center to obtain an initial clustering result; And (2) performing core fusion operation based on intra-class divergence: adaptively determining a core point in each initial class, fusing the two initial classes if the core pointsof the two initial classes are adjacent and the intra-class divergence can be reduced after fusion, and fusing all the initial classes to be fused to obtain a final clustering result. The method is simple, accurate and efficient, and can perform self-adaptive clustering on the data set with any shape and density distribution.

The invention, which relates to the analysis field of the user-project data, designs an improved recommendation method based on combination of a nearest neighbor clustering algorithm and a slope one algorithm. According to the invention, the method is based on combination of the k-means clustering method and the slope one algorithm. The k-means clustering method is used for finding similar individual projects for a user to form a neighbor project set; a user-project investigation data table is calculated by using the slope one algorithm; and then data prediction is carried out and the data are recommended to the user. When possibility of project investigation on a user is determined, similarity between the data projects is taken into consideration, so that the algorithm prediction result becomes accurate. A predication investigation possibility problem is converted into a project recommendation problem for solution. The possibility of investigation on the user is predicted based on the neighbor project slope one algorithm, so that the user can take related measures in advance.

The invention discloses a method of quickly screening difference of dry sulphitation radix paeoniae alba and non-dry sulphitation radix paeoniae alba by UPLC-MS-MS. The method comprises the followingsteps: (1) collecting medicinal materials of dry sulphitation radix paeoniae alba and non-dry sulphitation radix paeoniae alba to be differentiated to prepare a standard decoction sample; (2) feedingthe sample into an UPLC-MS-MS system for analysis to obtain mass spectrometric data; (3) directly importing originally collected data into Compound Discoverer 2.1 to analyze data; (4) correcting chromatographic peak drift by adopting a dynamic planning calculating method, aligning the chromatographic peaks, and registering matters, the retaining times of which are the closest, into a compound by means of a nearest neighbor clustering method; and (5) calculating different matters among samples by means of variance analysis and determining the difference between dry sulphitation radix paeoniae alba and non-dry sulphitation radix paeoniae alba. The compound mass spectrum information is output as an original file which does not need to be converted into other formats and can be directly imported into a Compound Discoverer 2.1 library to search for compounds. The method can quickly lock different compounds in traditional Chinese medicinal material samples and is suitable for quick analysisand determination of bulk samples.

The invention discloses a multi-source data visual analysis and display method and system. The multi-source data visual analysis and display method comprises the steps of receiving all data information, cleaning data, deleting abnormal data, and obtaining a data sample set; according to the type of the data sample, selecting the size of a clustering number K value; distributing the data samples inthe data sample set to the nearest neighbor cluster according to the minimum distance principle; recalculating the centers of the K clusters according to the nearest clustering result, and taking thecenters as new clustering centers; judging whether convergence occurs or not, namely judging whether the clustering center does not change any more, and if not, repeating until convergence to obtainan optimal clustering result; according to the optimal clustering result, marking corresponding labels on the data samples and enabling the labels to correspond to the display templates in a one-to-one mode, and conducting display. The multi-source data visual analysis and display system comprises a data sample processing module, a clustering number selection module, an approaching and clusteringmodule, a clustering center operation module, an optimal clustering operation module and a display module.

The invention relates to a method for solving copper conductor temperature by optical fiber temperature for photoelectric composite submarine cables. The method mainly comprises the following steps: (1) listing the thermal load operating current range, ambient temperature range, environment convection heat exchange coefficient range during operation of the photoelectric composite submarine cable, and carrying out uniform discretization respectively; (2) using the optical fiber temperature Tf and parameters I, G and H as the input variable X, and the copper conductor temperature Tc as the output variable; (3) through a test, measuring and acquiring the output optical fiber temperature Tf and the copper conductor temperature Tc under different thermal load operating current I, ambient temperature G and environment convection heat exchange coefficient H of the photoelectric composite submarine cable, wherein X and Tc form mx*n*p input-output sample pairs; (4) selecting a Gaussian function based on a radial basis function of a multilayer feedforward neural network; (5) studying the m*n*p input-output sample pairs by the multilayer feedforward neural network, wherein the study algorithm is a nearest neighbor clustering algorithm, the values of wi, ci and sigmai are obtained after studying, i=1, 2, 3,...N; and (6) solving the copper conductor temperature Tc from I, G, H and Tf by using a formula.

The invention discloses a multi-reconnaissance-platform pulse-to-pulse agile radar radiation source frequency information association method, and the method comprises the steps: firstly carrying out the classification of radar radiation source frequency samples reported by two reconnaissance platforms through employing a nearest neighbor clustering algorithm, and calculating the number of frequency samples corresponding to each reconnaissance platform in each class; respectively constructing frequency sample interception frequency vectors of the two reconnaissance platforms; calculating cosinesimilarity between the vectors of the interception times of the two frequency samples; and finally, comparing the cosine similarity with a judgment threshold, and judging whether the radiation sourcefrequency information reported by the two reconnaissance platforms is matched. According to the method, the accuracy of frequency information association of the pulse-to-pulse agile radar radiation source of the multi-reconnaissance platform is effectively improved.

The invention belongs to the field of image recognition, particularly relates to a wheel positioning system, method and device for automatic car washing, and aims to solve the problems of resource waste and poor washing effect caused by indiscriminate washing of a car body and wheels in an existing automatic car washing method. The method comprises the steps that an image acquisition module acquires wheel point set images through single-line laser radars arranged on the two sides of a parking space, a point set clustering module performs clustering through a nearest neighbor clustering methodof an adaptive threshold to generate wheel clustering point sets, and a point set fitting module performs fitting to generate wheel clustering circle images; the wheel positioning module obtains the wheel circle center coordinates and the wheel radius, wheel position information is obtained according to the wheel circle center and the wheel radius in combination with the position of the single-line laser radar, and then the wheel is accurately cleaned. Wheels are accurately positioned in automatic car washing, and wheel washing and car body washing can be distinguished, so that the purposes ofimproving the washing effect and saving resources are achieved.