1421 results about "Extreme learning machine" patented technology

Disclosed is a copper sheet and strip surface defect detection method based an on-line sequential extreme learning machine. The method includes the following steps that a copper sheet and strip surface image is captured through an image capturing module; the captured copper sheet and strip surface image is enhanced according to the median filtering method with the masking size of 7*7 to reduce noise in the copper sheet and strip surface image and the effect of the noise on the quality of the surface image; the copper sheet and strip surface image is subject to tophat transform treatment to reduce the effect of uneven illumination; a copper sheet and strip surface image pre-detection method based on eight-neighborhood difference values is adopted; defects in the surface image are segmented according to an image segmentation method, wherein it is judged that the copper sheet and strip surface image has the surface defects after pre-detection; geometrical characteristics, gray characteristics, shape characteristics, texture characteristics and other characteristics of each defect are extracted, and copper sheet and strip surface defect characteristic dimensions are subject to optimization and dimensionality reduction according to the principal component analysis method; a copper sheet and strip surface defect classifier based on the on-line sequential extreme learning machine is designed, and samples are used for training; characteristics of the copper sheet and strip surface image to be detected are extracted to identify types of the surface defects.

The invention discloses an on-line sequential extreme learning machine-based incremental human behavior recognition method. According to the method, a human body can be captured by a video camera on the basis of an activity range of everyone. The method comprises the following steps of: (1) extracting a spatio-temporal interest point in a video by adopting a third-dimensional (3D) Harris corner point detector; (2) calculating a descriptor of the detected spatio-temporal interest point by utilizing a 3D SIFT descriptor; (3) generating a video dictionary by adopting a K-means clustering algorithm, and establishing a bag-of-words model of a video image; (4) training an on-line sequential extreme learning machine classifier by using the obtained bag-of-words model of the video image; and (5) performing human behavior recognition by utilizing the on-line sequential extreme learning machine classifier, and performing on-line learning. According to the method, an accurate human behavior recognition result can be obtained within a short training time under the condition of a few training samples, and the method is insensitive to environmental scenario changes, environmental lighting changes, detection object changes and human form changes to a certain extent.

The invention provides a human falling detection method based on machine vision. Under the conditions that all detected objects are people in indoor environment and the activities of all people can be captured by a depth camera, the method comprises the following steps of: (1) acquiring an image by using the depth camera; (2) subtracting a background from the image which is acquired, and extracting a foreground from the image; (3) acquiring a foreground contour by using a binary image contour extraction algorithm; (4) performing convolution computation on the foreground contour and Gaussian functions in different scales, mapping a contour image to a curvature scale space (CSS) to form a CSS image, extracting peak points of the CSS image, and thus obtaining a video word bag based on curvature scale characteristics by the peak points in different scales; and (5) training an extreme learning machine classifier by using the video word bag which is obtained. By adoption of the method, under the condition of few training samples, an accurate falling detection result can be obtained within a short training time.

A fault diagnosis method for power electronic circuits based on optimizing a deep belief network, including steps. (1) Use RT-LAB hardware-in-the-loop simulator to set up fault experiments and collect DC-link output voltage signals in different fault types. (2) Use empirical mode decomposition to extract the intrinsic function components of the output voltage signal and its envelope spectrum and calculate various statistical features to construct the original fault feature data set. (3) Based on the feature selection method of extreme learning machine, remove the redundancy and interference features, as fault sensitive feature data set. (4) Divide the fault sensitive feature set into training samples and test samples, and primitively determine the structure of the deep belief network. (5) Use the crow search algorithm to optimize the deep belief network. (6) Obtain the fault diagnosis result.

The invention relates to a fault diagnosis method of a power transformer based on an extreme learning machine, which can be applied to a transformer monitoring / detection device or system. Fault characteristics are extracted based on data collected by the monitoring / detection device or system, and fault diagnosis module leaning of the extreme leaning machine of the transformer is carried out by selecting state samples of the transformer. The method comprises the steps of dividing the operating state of the transformer; selecting monitoring / detection data comprising the operating state of the transformer as a data source; extracting characteristics of the data source of the transformer, and determining characteristic variables; determining target vector expression manner of the extreme learning machine of the transformer in various operating states; selecting sample data of the transformer in various operating states; determining training sample data and testing the sample data; determining an input layer, a hidden layer, node number of an output layer and an excitation function of the fault diagnosis module of the extreme learning machine of the transformer; and learning and verifying the fault diagnosis module of the extreme learning machine of the transformer.

The invention discloses an eye-tracking method and device and belongs to the technical field of image processing. The eye-tracking method comprises the following steps: determining an observation area where the iris center of an iris image to be tested is according to a target model which is obtained according to target parameters and an extreme learning machine neural network; adopting the observation area to correct a predication area to obtain a target area, wherein the prediction area is the area, where the iris center of the iris image to be tested is located, determined by the Kalman filtering method; determining a screen falling position watched by an eye according to the target area. The eye-tracking method and device solve the problems that the eye watching accuracy and speed of the screen falling point are relatively low, achieve the effect of improving the eye watching accuracy and speed of the screen falling point, and are used for eye tracking.

The present invention belongs to the field of power prediction of wind power generation and particularly relates to a method for predicting a wind power interval based on a particle swarm optimization nucleus limit learning machine model. The method comprises: carrying out data preprocessing, i.e. preprocessing historical data in SCADA according to correlation between a wind speed and power; initializing a KELM model parameter and carrying out calculation to obtain an initial output weight betaint; initializing a particle swarm parameter; constructing an optimization criterion F according to an evaluation index and carrying out particle swarm optimization searching to obtain a model optimal output weight betabest; and bringing test data into a KELM model formed by betabest to obtain a wind power prediction interval and evaluating each index of the prediction interval. The method is easy for engineering realization; a good prediction result can be obtained; not only can a future wind power possible fluctuation range be described, but also reliability of the prediction interval is effectively evaluated, possible fluctuation intervals of wind power at different confidence levels are given out and reference is better provided for a power system decision maker.

The invention discloses a performance prediction method applicable to dynamic scheduling for a semiconductor production line. An extreme learning machine (ELM) is applied to prediction and modeling in the performance prediction method. Feeding control and scheduling rules are considered in a unified manner in the method, short-term scheduling key performance indexes such as an equipment utilization rate and a movement step number are predicted on the basis of a real-time state of a system, and a foundation is provided for dynamic real-time scheduling. A novel feed-forward neural network of the ELM is introduced into the semiconductor manufacturing system, and a prediction model is built by the aid of available data of the production line. As shown by test results, ideal prediction results can be quickly acquired by the method implemented by the aid of the ELM, the method has obvious advantages and an obvious application prospect in the aspects of parameter selection and learning speed as compared with the traditional neural network modeling method, and a new idea is provided for online optimal control.

The invention provides a one-dimensional deep convolution network underwater multi-target recognition method. A 6dB / frequency doubling one-order digital filter is adopted to enhance a high-frequency part to enable a signal spectrum to be flattened; a window function is selected to intercept signals, and signals with the duration of 170 ms are acquired as the best input frame length for a convolutional neural network; the one-dimensional deep convolution network is adopted to carry out feature extraction on sound signals; a training sample set is used for learning the convolution network to obtain network structure parameters with the best features; and an extreme learning machine is selected to carry out classification and recognition on output features of the convolution network. the deep convolution network in the deep learning is used for carrying out feature extraction on the sound signals, the traditional manual feature extraction is replaced, the automatically-extracted sound features contain richer recognition information, the extreme learning machine is used for carrying out classification and recognition on features automatically extracted by the convolution network, features different from those obtained in the traditional manual analysis mode can be effectively found out, and the underwater sound signal recognition rate is improved.

The invention is applied to the technical field of computers and provides a generation method and a generation device of a strategy network model for automatic drive of a vehicle. The method comprises the steps of enabling vehicle state information collected at each test moment to form a state information set, processing the state information set, and acquiring a characteristic state information set; finding a vehicle movement which acquires the maximum return value in each piece of state information of the characteristic state information set according to a preset return value function and a pre-constructed vehicle movement set; training a network model of an extreme learning machine according to the characteristic state information set, the maximum return value corresponding to each piece of state information in the characteristic state information set and the vehicle movement which acquires the maximum return value; and generating the strategy network model for automatic drive of the vehicle according to a training result of the network model of the extreme learning machine. Therefore, the consumption of computing resource is effectively reduced and the generation efficiency of the strategy network model for automatic drive of the vehicle is effectively improved.

The invention discloses an intelligent detection and quantitative recognition method for the defect of concrete. According to the method, a concrete test piece is subjected to impact echo signal sample acquisition, signal noise reduction treatment and characteristic value extraction so as to construct a recognition model for analysis components including feature extraction, defect inspection, defect diagnosis and defect quantification and positioning; and the model is used for detecting and recognizing to-be-detected concrete. The intelligent detection and quantitative recognition method provided by the invention is directed at disadvantages of conventional detection technology for concrete defects and based on theoretical analysis, value simulation and model testing, employs advanced signal processing and artificial intelligence technology and fully digs out characteristic information of a testing signal, thereby establishing the model for intelligent rapid detection and classified recognition based on wavelet analysis and an extreme learning machine; and the model has good classified recognition performance, realizes intelligent rapid quantitative recognition and evaluation of the variety, properties and scope of the defect of concrete and further improves the innovation and application level of non-destructive testing technology for the defect of concrete.

The invention discloses an extreme learning machine-based hyperspectral remote sensing image ground object classification method. An original extreme learning machine network is expanded into a hierarchical multi-channel fusion network. In terms of network training, the method is different from the least squares algorithm-based output weight solving strategy of the original ELM (extreme learning machine) and the global iterative optimization strategy of a deep learning network; according to the method of the invention, a greedy layer-by-layer training mode is adopted to train a hierarchical network layer by layer, and therefore, the training time of the network is greatly shortened; and in the layer-by-layer training process, a l1 regular optimization item is added into the training solving model of each layer of the network separately, so that parameter solving results are sparser, and the risk of over-fitting can be lowered. In terms of network functions, A single-hidden layer ELM network focus on solving the fitting and classification problems of simple data, while the different levels of the network model provided by the invention achieve target data feature learning or feature fusion, the network model of the invention integrates the advantages of high training speed and strong generalization capacity of the single-hidden layer ELM network, and therefore, the in-orbit realization of the model is facilitated, and the requirements of emergency response tasks can be satisfied.

The invention discloses a rolling bearing fault diagnosis method based on improved variational model decomposition and an extreme learning machine. The method comprises: vibration signals of a rollingbearing under different types of faults are collected, the vibration signals are filtered by means of maximum correlation kurtosis deconvolution, parameter optimization is carried out on the maximumcorrelation kurtosis deconvolution method by using a particle swarm algorithm, and an enveloped energy entropy after signal deconvolution is used as a fitness function; the mode number of variationalmodel decomposition is improved by an energy threshold and improved variational model decomposition of the filtered vibration signals is realized to obtain mode matrixes of the corresponding vibrationsignals; singular value decomposition is carried out on the mode matrixes to obtain a singular value vector and a rolling bearing fault feature set is constructed; and the fault feature set is trained by using an extreme learning machine and a rolling bearing fault diagnosis model is established. Therefore, stable feature extraction of the complex vibration signal of the rolling bearing is realized, so that the diagnostic accuracy is improved.

The invention discloses a power system short-term load forecasting method based on improving extreme learning machine (IELM) methods, which adopts an extreme learning machine (ELM) as the basic structure of a forecasting model and is an iteration-analysis learning algorithm which puts forward BFGS (Broyden, Fletcher, Goldfarb, Shanno) simulated Newton's method to give optimization, training and adjustment to network left metrics and analyze to get right metric parameters. The invention establishes the short-term load forecasting model based on improving extreme learning machine (IELM) method, puts forward extreme learning machine network reference hidden node concepts, trains the equidimensional extreme learning machine networks which have the same numbers of hidden nodes and samples, orderly clusters module values of equidimensional network right metrics vector, finds out a plurality of corresponding module value break points, and regards the break points as the reference hidden nodes of predicting networks. The method of the invention is also greatly improved in the aspect of prediction precision and speed.

The invention provides a method for diagnosing faults of bearings on the basis of extreme learning machines, and belongs to the field of technologies for diagnosing mechanical faults. The method includes carrying out variation mode decomposition on vibration acceleration signals to obtain IMF (intrinsic mode function) mode components; acquiring singular values of the IMF mode components by the aid of singular value decomposition (SVD) algorithms; dividing the singular values of each IMF mode component into a training sample and a test sample; utilizing the singular values of the training samples as input values of extreme learning machine (ELM) neural network models and determining input connection weights, offset values and optimal output connection weights of the ELM neural network models; utilizing the singular values of the test samples as input values of the ELM neural network models with the determined input connection weights, the determined offset values and the determined optimal output connection weights and acquiring output results which are bearing fault diagnosis results. The method has the advantages that the signals can be accurately effectively separated, component signal modes are fast in convergence and high in robustness, and the method is high in fault recognition speed and accuracy; model building can be omitted, professional requirements can be lowered, and accordingly the method is suitable for industrial application.

The invention discloses a multiclass image classification method based on a semi-supervised extreme learning machine. A training sample set is divided into an already marked sample set L, an unmarked sample set U and a test set T; training subsets of N differences are constructed by carrying out replaceable resampling on samples in the U; semi-supervised extreme learning machine models are respectively trained with the already marked training sample L on each subset, wherein all together there are N classifiers; the sum of output of corresponding nodes of N extreme learning machines is solved and an average is taken; after normalization processing on the output, uncertainty evaluation is carryed out on the samples by use of the active learning technology of a best mark and second best mark algorithm BvSB, most uncertain samples are taken from the U for manual marking and are transferred to the L; and a classifier model is re-updated until iteration ends. Through the method provided by the invention, the problems of low classification correction rate and low learning speed existing in image classification in related arts are solved, and a certain foundation is laid for accurate, rapid and stable image classification.

The present invention discloses an improved extreme learning machine method based on artificial bee colony optimization, which includes the following steps: Step 1, generating an initial solution for SN individuals: Step 2, globally optimizing a connection weight ω and a threshold b for the extreme learning machine; Step 3, locally optimizing the connection weight ω and threshold b of the extreme learning machine; Step 4, if food source information is not updated within a certain time, transforming employed bees into scout bees, and reinitializing the individuals after returning to Step 1; and Step 5, extracting the connection weight ω and threshold b of the extreme learning machine from the best individuals, and verifying by using a test set. With the method provided by the present invention, the defect of worse results of the traditional extreme learning machine in classification and regression is overcomed, and effectively improves the results of classification and regression.

The invention discloses an on-line fault diagnosis method of weighted extreme learning machine sewage treatment on the basis of a kernel function. The on-line fault diagnosis method comprises the following steps: 1) rejecting data with an incomplete attribute in sewage data, and then, carrying out data normalization processing to determine a historical data set and an update test set; 2) selecting a kernel function and a weighting scheme, and then, determining model parameters according to an optimal model; 3) according to the selected weighting scheme, endowing a weight for each sample of the historical data set; 4) training the model, and calculating a kernel matrix according to the kernel function; 5) adding a new sample into the model from a new test set for testing, and updating the historical data set; and 6) returning to 3), training the model again, and continuously repeating the above process until the on-line data test is finished so as to realize the identification of the on-line operation state of the sewage treatment process. The method has the advantages of short update time and high classification accuracy rate and has an important meaning for diagnosing operation faults in real time, guaranteeing the safe operation of sewage treatment works and improving the operation efficiency of the sewage treatment works.

The invention discloses a fault diagnosis method based on local mean decomposition (LMD), singular value decomposition (SVD) and an extreme learning machine (ELM), and aims to increase the accuracy of bearing fault diagnosis under a variable condition. The method comprises the following steps: firstly, decomposing a nonlinear and unstable original vibration signal into a series of product functions (PFs) by adopting an efficient adaptive signal processing method LMD, wherein each PF component is a product of an envelope signal and a pure frequency modulated signal having physical meaning; secondly, processing the PF components by adopting the SVD to compress a feature vector scale and obtain a more stable feature vector value; lastly, classifying a bearing fault state by applying the ELM having higher application computation efficiency and higher classification accuracy based on an extracted feature vector. By adopting the fault diagnosis method based on LMD-SVD-ELM, a set of complete and valid variable-condition fault diagnosis scheme is provided for a bearing. The method has a very good practical engineering application value.

Provided is an extreme learning machine based on a length-changing particle swarm optimization algorithm. The method comprises the steps that (1) the position and the speed of a particle swarm are initialized randomly, each particle in the particle swarm represents an ELM classifier; (2) an adaptation value f(p) for an evaluation function of each particle is computed; (3) the magnitude relationship of the row number of particles and the row number of globally optimal solution is compared, different updating formulas are selected to update the speed and the positions of the particles, a next generation of particle swarm is generated; (4) the optimum hidden unit number and corresponding input weight and hidden unit offset are selected; (5) output weight is computed, and the ELM classifier with the highest cross validation accuracy is obtained. The hidden unit number is automatically selected through the length-changing particle swarm optimization algorithm, and meanwhile the corresponding input weight and hidden unit offset are selected, so that the generalization performance of the ELM classifier is maximum, the ELM classifier with a small number of the hidden units can obtain the maximum generalization performance, testing time is short, and efficiency is high.

The invention discloses a zero sample classification method based on an extreme learning machine, and the method is used for image classification. The method comprises the following steps: extracting the visual features of a training image at a training state, and extracting the training semantic features corresponding to the visual features of the training image; randomly generating a first input weight and a first threshold value for L junctions, and calculating a first output matrix of a hidden layer through employing a hidden layer mapping function; calculating the output weight of a network through the training semantic features and the first output matrix of the hidden layer; extracting the visual features of a test sample at a test stage, randomly generating a second input weight and a second threshold value for L junctions, and calculating a second output matrix of the hidden layer through employing the hidden layer mapping function; calculating an embedded vector, correspondingly located in a semantic space, of the second output matrix through the output weight, and judging the type of the test sample according to the similarity of the embedded vector with the semantic features in a semantic feature space. The method reduces the training time, and improves the classification speed of the image.

The invention discloses a rapid facial expression recognition method based on an ELM self-encoding algorithm. The rapid facial expression recognition method comprises the steps of: step 1, training an Adaboost-based face region detection classifier and performing face detection; step 2, preprocessing a detected face region, including clipping, size normalization and histogram equalization processing; step 3, adopting an ELM-AE algorithm based on an auto-encoder and an extreme learning machine as a feature extraction algorithm, and performing feature extraction on a facial expression image after preprocessing; step 4, and constructing a facial expression classifier based on the extreme learning machine, inputting vectors of feature extraction into the facial expression classifier, and regarding an output result as a mood of a face. The rapid facial expression recognition method can extract primary information and perform dimensionality reduction more quickly and efficiently. When in facial expression recognition classification, the ELM just needs to adjust a parameter of a neuron, the recognition is short in operating time and high in accuracy rate, and the rapid facial expression recognition method is efficient and rapid in learning speed.

The invention provides a detection method and apparatus for a malicious android application, and relates to the technical field of android systems. The method comprises: obtaining a training sample set consisting of a malicious application sample and a normal application sample; performing fusion feature extraction on the training sample set, and performing selection on the extracted fusion features with a principal component analysis method to obtain android application related fusion features; according to the android application related fusion features, establishing a malicious application detection model based on an extreme learning machine; and according to the malicious application detection model based on the extreme learning machine, detecting a to-be-tested android application and determining whether the to-be-tested android application is a malicious application or not. According to the detection method and apparatus, the problem of incapability of correctly determining a malicious application due to detection failure of the malicious android application possibly caused by a current detection method based on a feature code, a static source code, a dynamic behavior or the like can be solved.

Bearing fault diagnosis based on pseudo-tag semi-supervised kernel local Fisher discriminant analysis is provided. The method is characterized in comprising the following steps: (1) collecting vibration signals of bearings under different working conditions to form a training sample; (2) performing feature extraction on the training sample obtained in (1); (3) performing normalization processing on features obtained in (2); (4) obtaining a clustering tag set by using density peak clustering for the entire feature set obtained in (3); (5) using clustering pseudo-tags obtained in (4) to construct local inter-cluster divergence and intra-cluster divergence regularization terms, and combining the regularization terms with the inter-class divergence and intra-class divergence with tag samples in the FDA to determine a final projection vector; (6) using the final projection vector obtained in (5) to solve a projection set of the tagged feature set in the dimensionality reduction space; (7) using the projection set obtained in (6) to train an extreme learning machine; and (8) performing processing of (2), (3) and (5) on the collected vibration signals, and inputting the processed vibration signals to determine the working conditions. The technical scheme of the present invention is applied to the problem of fault identification of bearing equipment.

The present invention relates to an image automatic marking method based on Monte Carlo data balance. The method comprises the steps of carrying out the region segmentation on the training sample images in a public image library, enabling the segmented regions possessing different characteristic description to correspond to one marking word, then carrying out the Monte Carlo data balance on the different types of image sets, extracting the multiscale characteristics of the balanced images, and finally inputting the extracted characteristic vectors in a robustness least squares increment limit learning machine to carry out the classification training to obtain a classification model in the image automatic marking; for the to-be-marked images, carrying out the region segmentation on the to-be-marked images, adopting the same multiscale characteristic fusion extraction method and inputting the extracted characteristic vectors in the least squares increment limit learning machine to obtain a final image marking result. Compared with a conventional image automatic marking method, the method of the present invention enables the images to be marked more effectively, is strong in timeliness, can be used for the automatic marking of the large-scale images, and possesses the actual application meaning.

The invention discloses a short-term electric load prediction method based on improved genetic algorithm for optimizing extreme learning machine. A hill climbing method is used to perform preferentialselection again in the progeny population, an initial individual is selected, another individual in a close area is select, their fitness values are compared, and one individual which has good fitness values is leaved. If the initial individual is replaced or a better individual cannot be found in several iterations, iteration is stopped, the search direction of the genetic algorithm through thehill climbing method is optimized, obtaining an optimal weight value and a threshold value, a network optimization prediction model are obtained, a network optimization prediction model is obtained, the network optimization prediction model and prediction results of BP network and the extreme learning machine are comparative analyzed, including selection of input and output of the prediction network model, algorithm of improved genetic algorithm for optimizing extreme learning machine, and analysis of prediction results. The short-term electric load prediction method based on improved geneticalgorithm for optimizing extreme learning machine has faster training speed and more accurate prediction results, and is suitable for modern short-term electric load prediction with plenty of influence factors and huge data volume.

The invention is applicable for the field of robots and intelligent control, and provides a robot imitation learning method and device, a robot and a storage medium. The method comprises the steps ofacquiring the pose of an end-effector at the current time when receiving movement instructions, and checking whether the pose at the current time is the target pose; determining that a preset imitation learning task is finished by the end-effector if the pose at the current time is the target pose; or else generating the predicted pose of the end-effector at the next time according to the pose atthe current time and a dynamic prediction model, and adjusting angles of joints according to the predicted pose; and setting the pose after adjustment of the end-effector to the pose at the current time, and skipping to the step of checking whether the pose at the current time is the target pose. The dynamic prediction model is obtained through a limit learning machine model in combination with preset stability constraint condition training, so that the stability, repetition precision and model training speed of robot imitation learning are ensured, and the humanization degree of robot motionis improved effectively.

The invention discloses a vehicle license plate recognition method based on extremal regions and an extreme learning machine. The method includes the steps that color images to be processed are preprocessed, vehicle license plate regions are roughly positioned, and multiple vehicle license plate candidate regions are obtained; based on the vehicle license plate candidate regions, the extremal regions of RGB color channels are extracted from the color images to be processed, the extremal regions according with the geometric attributes of vehicle license plate character regions are selected from a classifier, and the vehicle license plate character regions are obtained; a single implicit strata feedforward neural network based on the extreme learning machine is established through supervised learning, characteristic vectors of the character regions are extracted as input, and vehicle license plate characters are automatically recognized through the neutral network. The method has the advantages of being high in speed and precision and the like and can well deal with adverse factors such as complex backgrounds, weather changes, illumination influence and the like particularly in complex traffic environments. The defects of a traditional vehicle license plate recognition method in real time performance and robustness are overcome, and the method has significant application value.

The invention relates to a wavelet transformation and improved firefly-optimized extreme learning machine-based short-term load prediction method. The method includes the following steps that: noise reduction is carried out on an original load sequence through wavelet decomposition and reconstruction; (2) in a model training stage, improved firefly algorithm optimized extreme learning machine parameters are utilized to obtain the optimal model of sub sequences; and (3) the final prediction values of the sub sequences, which are obtained superposition, are predicted. According to the method, numerical value calculation is carried out on data sequences of two kinds of time scales, and therefore, problems in short-term load prediction can be solved effectively. Compared with a plurality of classic prediction models such as a traditional ARMA, a BP neural network, a support vector machine and an LSSVM, the model adopted by the method has a prediction effect.

The invention relates to an extreme learning machine-based pipeline magnetic flux leakage defect detection method. The method comprises the following steps: establishing an extreme learning machine model according to data of the length, the width and the depth of a known pipeline magnetic flux leakage defect and a magnetic flux leakage signal waveform characteristic value; training the data of the length, the width and the depth of the known pipeline magnetic flux leakage defect in sample data, wherein the data is taken as the input of the model; selecting the number of nodes of a hidden layer by a cut-and-trial method; and calculating an output matrix and an output weight value of the hidden layer, wherein the magnetic flux leakage signal waveform characteristic value is taken as the output of the model; when a pipeline suffers from magnetic flux leakage, acquiring the waveform of a magnetic flux leakage signal of an unknown magnetic flux leakage shape, and performing pipeline magnetic flux leakage defect detection by the extreme learning machine model. According to the method, the pipeline defect shape is subjected to intelligent inversion by the extreme learning machine model; the method has the advantages of high learning speed, high generalization performance and the like; the shape of the defect can be constructed quickly and accurately by virtue of the waveform of the detected defect, so that the severity of the defect is learnt, a pipeline risk can be foreknown, and the leakage of the pipeline is prevented.