2536 results about "Negative sample" patented technology

An exemplary embodiment of the present invention provides a method of verifying an identity of a person-to-be-identified using biometric signature data. The method comprises creating a sample database based on biometric signature data from a plurality of individuals, calculating a feature database by extracting selected features from entries in the sample database, calculating positive samples and negative sampled based on entries in the feature database, calculating a key bin feature using an adaptive boosting learning algorithm, the key bin feature distinguishing each of the positive samples and negative samples, and calculating a classifier from the key bin feature for use in identifying and authenticating a person-to-be-identified.

The invention discloses a vehicle detection method fusing radar and CCD camera signals. The method includes the steps of inputting the radar and CCD camera signals, correcting a camera to obtain a projection matrix of a road plane coordinate and a projection matrix of an image coordinate, converting a road plane world coordinate into an image plane coordinate, building positive and negative sample sets suitable for a vehicle HOG describer, carrying out batch feature extraction on the vehicle sample sets to build an HOG sample set, building an SVM classification model of a linear support vector machine to train an SVM, extracting regions of interest of barriers detected by radar in a video image, inputting the regions of interest into an SVM classifier to judge a target category, outputting an identifying result, and measuring the distance of a target which is judged to be a vehicle by means of the radar. Combination detection is carried out by means of the radar and CCD camera signals, depth information of the vehicle is obtained, meanwhile, profile information of the vehicle can be well detected, and reliability and accuracy of vehicle detection and positioning are improved.

The invention discloses a goods recommendation method based on scores and user behaviors. First of all, a latent factor model is established for user score data, goods are automatically clustered, latent classes or feature factors are found, user interest is decomposed into preference degrees of the multiple latent classes, the goods are expressed by use of weights comprising latent features, and the scores of the users for the goods are inner products of the user interest and the goods. Then for the purpose of solving the score data sparsity problem, by use of the user behaviors, negative samples are introduced, the features are extracted, and a possibility that the users buy the goods is estimated through a logic regression model. Finally, candidate sets of the two are combined and weighed for ordering, and top goods are recommended to the users. According to the invention, diversified interest of the users is discovered from the single scores by use of the latent factor model, information of the multiple features of the goods is mined, the method better accords with actual application, the negative samples are introduced, distinctiveness of the user interest is enabled to be larger, the quality of a recommendation result is higher, demands of the users can be better satisfied, and the method can be applied to recommending the goods.

The embodiment of the invention discloses a target user orientation method and device. The method comprises: generating a virtual user basing on the user characteristic of a seed user; taking the seed user and the virtual user as positive sample users; determining negative sample users; respectively extracting the user characteristics of the positive sample users and the negative sample users; training basing on the user characteristics of the positive sample users and the negative sample users, determining orientation parameters; and orientating the target user satisfying the predetermined similarity condition with the seed user basing on the orientation parameters.

The invention discloses a deep learning-based vehicle detection method which is used for conducting vehicle detection in a complex environment by combining Edge Boxes and an improved Faster R-CNN model. The method comprises the steps that an image is processed by using the Edge Boxes, and an accurate vehicle candidate area is extracted primarily; and then the candidate area is input to the improved Faster R-CNN model to achieve precise positioning on vehicles, and a final detection result is obtained through classification and discrimination. In order to enhance the detection capacity on the small-size vehicles by the model and the discrimination capacity of the model, different layers of convolutional features are combined, some detail information of the vehicles is supplemented, a difficult negative sample mining strategy is added in the training stage, therefore, the model can pay more attention on difficult samples, and the vehicles can be well distinguished with the backgrounds ofsuspected vehicles.

Provided is a pedestrian detection method. The pedestrian detection method comprises the following steps that a pedestrian positive sample set and a pedestrian negative sample set needed for training a convolutional neural network are prepared; the sample sets are preprocessed and normalized to conform to a unified standard, and a data file is generated; the structure of the convolutional neural network is designed, training is carried out, and a weight connection matrix is obtained during convergence of the network; self-adaptive background modeling is carried out on videos, information of moving objects in each frame is obtained, coarse selection is carried out on detected moving object regions at first, the regions with height to width ratios unsatisfying requirements are excluded, and candidate regions are generated; each candidate region is input into the convolutional neural network, and whether pedestrians exist is judged.

The invention discloses an analog circuit dynamic online failure diagnosing method based on GSD-SVDD, belonging to the technical field of analog circuit failure diagnosis. In an offline test process, a KFCM algorithm is adopted to calculate a failure resolution value of each testable node and an optimal test node set is selected according to the failure resolution value. In an online diagnosis process, a failure diagnosis model is established by adopting an SVDD single classification approach based on a map spatial distance positive and negative sample weighting, test samples are diagnosed by a layered diagnosis method, and a failure class library and the diagnosis model are renewed dynamically. The method effectively reduces the drill and online diagnosis time of the diagnosis model, guarantees the real-time property of the online diagnosis and improves the precision of the failure diagnosis and can dynamically renew parameters of the diagnosis model so as to enable the a diagnosis system to have the self-adaption capability.

The invention discloses a generative adversarial network improved CT medical image pulmonary nodule detection method. The method includes: 1), acquiring a section of a pulmonary CT image; 2), separating according to image morphological properties to acquire a ROI pulmonary parenchyma area; 3), acquiring different suspected pulmonary nodule candidate sets according to a connected domain formed by abinarized image; 4), building a model of an assistant classifier generative adversarial network to generate positive samples overcome the circumstance that positive-negative sample number is unbalanced; 5), building a convolution neural network to classify suspected pulmonary nodule parts to acquire pulmonary nodule areas; 6), using a non-maximum suppression algorithm to acquire a final area of pulmonary nodule. By the method, efficient processing performance of a computer can be fully utilized, certain expandability is provided, and data processing efficiency is improved; through a convolution neural network algorithm, classifying accuracy is improved, CT image data processing performance is improved, and pulmonary nodule images can be built and analyzed more efficiently.

The invention discloses a triple loss-based improved neural network pedestrian re-identification method. The method comprises the following steps of constructing a sample database, establishing positive and negative sample libraries based on the sample database, and randomly selecting two positive samples and one negative sample to form a triple; constructing a triple loss-based neural network, and performing training, wherein the neural network is formed by connecting three parallel convolution neural networks with a triple loss layer; inputting a to-be-tested picture and each sample picture in the expanded sample database, which serve as a group of inputs, to the trained neural network in sequence, wherein another input of the neural network is zero or zero input; and calculating a distance of eigenvectors of two input pictures output by the neural network by utilizing a Euclidean distance, and querying and arranging first 20 Euclidean distances in an ascending order, and then performing simple manual screening to obtain a final identification result. The method has the beneficial effects that the identification method can be suitable for a picture scene with a relatively great change, can ensure robustness, and has relatively high identification accuracy.

The invention discloses a video smoke detecting method based on multi-feature fusion. The video smoke detecting method includes extracting motion targets from a training video sequence, judging whether or not a suspected smoke area according to color features, extracting shape irregularity features, area growth features and K background fuzzy features from the suspected smoke area to combine into feature vectors to combine into feature vector, subjecting the feature vector into positive and negative sample division to obtain training samples; using the feature vector of the training samples as output and sample types as output and training a classifier; extracting the feature vector from images to be detected by the same method, and inputting the feature vector to judge whether smoke is or not. The features can be extracted from the suspected smoke area, smoke can be comprehensively judged and detection efficiency of smoke is improved.

The invention relates to a surface defect detection method based on positive case training. The method comprises two steps of image reconstruction and defect detection, image reconstruction is to reconstruct an inputted original image into an image without defects, reconstruction steps are as follows, artificial defects and noise are added to the positive case image during training, a self-encoderis utilized for reconstruction, the L1 distance between the reconstruction result and the noise-free original image is calculated, the distance is minimized as a reconstruction target, in cooperationwith the generative adversarial network, the reconstruction image effect is optimized; defect detection is performed after image reconstruction, LBP features of the reconstructed image and the original image are calculated, after difference between the two feature images is made, the two images are binarized based on the fixed threshold, so the defects are found. The method is advantaged in thatthe depth learning method is utilized, the method can be sufficiently robust to be less susceptible to environmental changes when positive samples are enough, moreover, based on regular training, themethod does not rely on a large number of negative samples and manual annotation, the method is suitable for being used in real-world scenarios, and the surface defects can be better detected.

The invention relates to an integrated transfer learning method for classification of unbalance samples, which comprises the following steps of: in the initializing process, giving different weights to positive and negative samples to ensure that the negative samples which account a small ratio for the total samples and have a large amount of information have large initial weights; in the training process in each round, extracting part of samples according to a certain ratio and using the selected samples as a training subset to carry out training, and after finishing the training, selecting the classifier with the smallest error from a plurality of simple classifiers as a weak classifier and regulating the training dataset according to a redundant data dynamic eliminating algorithm; and obtaining a weak classifier sequence after T rounds of iteration and overlaying and combining a plurality of weak classifiers into a strong classifier. According to the invention, the classification law of novel data which is distributed similarly with old data is found by effectively utilizing the classification law of the old data; particularly, a novel method is provided for solving the problem of classification of the data which is classified in an unbalance mode; the effect of a small amount of the negative samples in the classification process in the classification training process is ensured; the contribution rate of the negative samples is effectively improved; and the classification efficiency and accuracy are improved.

A method for determining the extent of recovery of materials injected or otherwise introduced into oil wells or subsurface formations is practiced using a portable device. The portable device can also be used to determiner the occurrence of a predetermined condition in an oil well such as water break through in a production zone, or the opening or closing of a sliding sleeve. When, for example, water breakthrough is detected, the zone producing too much water can be plugged, using, for example, a flow-through bridge plug, if there are other producing zones further downhole. The portable device may be use to make quantitative measurements or it may be used to do pre-screening of samples thereby avoiding further testing of negative samples.

The invention discloses a self-adaptive cascade classifier training method based on online learning, which comprises the following steps: (1), preparing a training sample set with a small quantity of samples, and training an initial cascade classifier HC(x) in a cascade classifier algorithm; (2), using the HC(x) for traversal of image frames to be detected, extracting areas with sizes identical to the sizes of the training samples one by one, calculating a feature value set, classifying the areas with the initial cascade classifier, and judging whether the areas are target areas, thereby completing target detection; (3) tracking the detected targets in a particle filtering algorithm, verifying the target detection results through tracking, marking detection with errors as a negative sample for online learning, obtaining different attitudes of real targets through tracking and extracting a positive sample for online learning; and (4) carrying out online training and updating for the initial cascade classifier HC(x) in a self-adaptive cascade classifier algorithm when an online learning sample is obtained, thereby gradually improving the target detection accuracy of the classifier.

The present invention discloses a driving model training method, a driver identification method, a driving model apparatus, a driver identification apparatus, a device, and a medium. The driving modeltraining method includes the following steps that: the training behavior data of a user are acquired, wherein the training behavior data are associated with a user identifier; training driving data associated with the user identifier are obtained on the basis of the training behavior data; positive and negative samples are obtained from the training driving data on the basis of the user identifier, and the positive and negative samples are divided into a training set and a test set; the training set is trained by using a bagging algorithm, so that an original driving model can be obtained; and the test set is adopted to test the original driving model, so that a target driving model can be obtained. With the driving model training method adopted, the generalization of the driving model can be effectively enhanced; the problem of poor recognition results of current driving recognition models can be solved; and the accuracy of identifying the driving of drivers is improved.

The invention provides a method for constructing a fracture recognition model and application. The method comprises: step A, acquiring X-ray images of different fracture types of different fracture parts; step B, carrying out image preprocessing on the X-ray images of different fracture types of different fracture parts; step C, carrying out image feature extraction on the preprocessed images; step D, generating a candidate region according to the image features; step F, carrying out fracture type target detection and region localization on the candidate region; step G, according to a facturetype target detection score, carrying out difficult sample mining to obtain negative samples for training; step K, expanding positive samples for training based on an adversarial learning strategy; and step H, training is carried out by using the positive samples and the negative samples so as to obtain all fracture recognition models for different fracture parts. Therefore, the fracture type andthe regional position can be predicted accurately by using the model, so that the diagnosis time of the doctor is reduced substantially and the missed diagnosis and misdiagnosis rates are reduced.

The invention discloses a state prediction method and device. The method comprises the following steps of: sampling target users; respectively generating a negative sample, a positive sample and a verification sample according to account information of lost and unlost users according to a recognized sampling moment; training a decision-making tree which is used for predicting user loss state afterthe sampling moment; inputting the verification sample into the trained decision-making tree model to obtain a predicted loss state; and if a correct recall rate obtained through carrying out calculation according to the predicted loss state and a practical loss state is not smaller than a threshold value, determining that the training of the decision-making tree model is completed, and carryingout user loss state prediction. According to the state prediction method and device, the decision-making tree model is trained by a training sample generated through sampling the target users, and user loss state prediction is carried out according to the trained decision-making tree model, so that the technical problems of relatively recognition efficiency low and reusing difficulty caused by user loss state prediction carried out through artificial experiences or established rules in the prior art are solved.

The invention relates to a target detection method based on high resolution optical satellite remote sensing images, and a system thereof. The target detection method based on high resolution opticalsatellite remote sensing images includes the steps: obtaining a marked target positive sample and a marked background negative sample to form a training sample; for the training sample, extracting a plurality of different weak characteristic channels, and according to the plurality of different weak channels, obtaining a candidate region; obtaining the context scene of the candidate region, performing characteristic extraction on the candidate region and the context scene of the candidate region, and fusing the extracted characteristics to form the characteristics of the candidate region; training the training sample, and obtaining a classifier; classifying the characteristics of the candidate region by means of the classifier, and obtaining a target region including the target; and performing duplicate removal on the target region to obtain a detection target. The target detection method based on high resolution optical satellite remote sensing images realizes target detection on remote sensing images with enlarged formats, optimizes the targets with very short interval and the target detection effect with uncommon length breadth ratio.

The invention discloses a deep-learning-based image identification method of the melanoma of a skin cancer. The method comprises: establishing a skin lesion dermatoscope image database; carrying out data preprocessing and quality assessment and screening; carrying out cascaded connection of deep convolution neural networks, and introducing transfer learning and a classifier. At a training stage, enhancement or screening is carried out on original data; and after inputting of positive and negative samples, sample expansion and overfitting prevention are carried out. At the preprocessing stage,data enhancement is added, cascaded connection of two deep convolution neural networks is carried out; transfer learning of existing pre-trained features at a natural image to an identification network is carried out; and then classification prediction is carried out by using the classifier and fine adjustment of network parameters is carried out based on network convergence and prediction situations. Therefore, the accuracy of skin lesion classification is improved; the restriction of manual feature selection is avoided; the adaptability is improved; and thus the deep-learning-based image identification method has the certain significance in a medical skin disease image analysis.

Embodiments of the invention provide a target user determination method and apparatus. The method comprises the steps of obtaining historical advertisement data of a transmitted advertisement meeting a preset matching condition with a to-be-transmitted advertisement; analyzing the historical advertisement data to determine a seed user of the to-be-transmitted advertisement; marking the seed user as a positive sample user and extracting a positive influence characteristic of the positive sample user; determining a negative sample user and extracting a negative influence characteristic of the negative sample user; and matching user characteristics of alternative users with the positive and negative influence characteristics to determine a target user from the alternative users.

The invention relates to an image block deep learning characteristic based infrared pedestrian detection method, and belongs to the technical fields of image processing a computer vision. According to the method, a data set is divided into a training set and a test set. In a training stage, firstly, small image blocks are extracted in a sliding manner on positive and negative samples of the infrared pedestrian data set, clustering is carried out, and one convolutional neural network is trained for each type of image blocks; and then feature extraction is carried out on the positive and negative samples by using the trained convolutional neural network group, and an SVM classifier is trained. In a test stage, firstly, a region-of interest is extracted for a test image, then feature extraction is carried out on the region-of-interest by using the trained convolutional neural network group, and finally prediction is performed by using the SVM classifier. The infrared pedestrian detection method achieves a purpose of pedestrian detection via a mode of detecting whether each region-of-interest belongs to a pedestrian region or not, so that pedestrians in an infrared image can be detected accurately under the conditions such that the detection scene is complicated, the environment temperature is high, and the pedestrians vary greatly in scale attitude, and the method provides support for research in follow-up related fields such as intelligent video.

The invention discloses a method and device for recommending cash goods, equipment and a storage medium. The method comprises the steps of obtaining multiple pre-determined positive samples and negative samples composed of cash goods purchased by multiple users; collocating preset user feature information with the positive samples and the negative samples correspondingly and generating a target training set; adopting a binary classifier algorithm, and training a preset mathematical model according to the target training set to obtain a training model; obtaining user feature information of a user to be predicted, inputting the user feature information as an input variable into the training model, and outputting an ordering result of cash goods corresponding to the user to be predicted. According to the method and device for recommending the cash goods, the equipment and the storage medium, the problem is solved that the cash goods recommended for the user are largely deviated from the cashed goods actually purchased by the user, the recommendation accuracy is improved, the profit is increased, and the user experience is improved.

The invention relates to the field of video monitoring, in particular to a safety helmet identification method integrating HOG human body target detection and an SVM classifier. The method includes the steps that parameter values Sigma of HOG positive and negative sample characteristics, an SVM classification function and a Gaussian kernel function are obtained; a monitoring box is extracted; a moving target is detected; HOG characteristic matching is conducted to judge whether a safety helmet is worn. By means of the safety helmet identification method, whether staff in a construction site wear safety helmets as required can be accurately monitored, the algorithm principle is simple, the real-time performance and a high accuracy rate are achieved, a human body target and a non-human-body target can be effectively distinguished, interference factors in a background are overcome, adaptability to outdoor changing light conditions and changes of colors of the safety helmets is achieved, and high robustness is achieved.

The invention discloses an image retrieval algorithm based on a free-hand sketch. According to the method, a metric learning algorithm is used for obtaining a mapping function, the free-hand sketch and a real image are mapped to the same metric space, an L1 normal form in the metric space can directly represent the 'semantic similarity' of the sketch in an original space and a real photo, and the problem of cross-domain matching of the free-hand sketch and the real photo is well solved. Specifically speaking, a positive and negative sample pair is used for training the convolutional neural network, feature vectors of the photo and the free-hand sketch input in the space and labeled to be similar are put together, the feature vectors of images labeled to be not similar are pushed away, and an integrated retrieval method is put out finally. Experiments show that the retrieval performance of the method is greatly improved in the large-scale free-hand sketch retrieval data set.

Systems and methods for classifying a object as belonging to an object class or not belonging to an object class using a boosting method with a plurality of thresholds is disclosed. One embodiment is a method of defining a strong classifier, the method comprising receiving a training set of positive and negative samples, receiving a set of features, associating, for each of a first subset of the set of features, a corresponding feature value with each of a first subset of the training set, associating a corresponding weight with each of a second subset of the training set, iteratively i) determining, for each of a second subset of the set of features, a first threshold value at which a first metric is minimized, ii) determining, for each of a third subset of the set of features, a second threshold value at which a second metric is minimized, iii) determining, for each of a forth subset of the set of features, a number of thresholds, iv) determining, for each of a fifth subset of the set of features, an error value based on the determined number of thresholds, v) determining the feature having the lowest associated error value, and vi) updating the weights, defining a strong classifier based on the features having the lowest error value at a plurality of iterations, and classifying a sample as either belonging to an object class or not belonging to an object class based on the strong classifier.

The invention discloses a method for automatically detecting the failure of an insulator. The method comprises the following steps: establishing positive and negative sample banks of insulator images, training an insulator classifier by a convolutional neural network algorithm, and splitting a collected image by utilizing the image salient area detection algorithm to split an equipment area and a background area in the image so as to obtain an insulator searching candidate area; training the insulator classifier by utilizing the convolutional neural network algorithm to locate the insulator, and controlling image collection equipment according to the location information of the insulator string in a current image to complete the image collection of the insulator string; and carrying out insulator surface stain detection, insulator string cracking detection and insulator string breakage or mixed foreign substance detection by utilizing an image identification technology so as to determine abnormal insulators. According to the method, insulator stains, cracking and breakage can be detected by adopting different image processing algorithms, insulator states can be comprehensively detected, and potential safety hazards of the insulators in transformer substations / converter stations can be reduced.

The invention provides a video high-level characteristic retrieval system based on a plurality of bottom-level characteristics of color, edge, texture, characteristic point and the like and a supportvector machine (SVM). Shot boundary detection is firstly carried out on a video clip, and then a plurality of representative frames are extracted at equal intervals from a shot to be as key frames. For the extracted key frames, a plurality of robust bottom-level characteristics of color, edge, texture and characteristic point are extracted. The adoption of the bottom-level characteristics provides description in different respects for video high-level semantic characteristics, because the low-level characteristics has strong complementarity and can respectively present strong differentiatingcapability for different semantic concepts, so that the detection performances of the system for different concepts can be ensured. The extracted characteristics are respectively sent to the support vector machine (SVM) for classification to form a multi-branch subsystem. In the concept classification stage, the support vector machine (SVM) is selected as a classifier, a method based on condensednearest neighbor is firstly used for selecting training parameters, so that the ubiquitous problem in training process of imbalance of positive and negative samples is effectively solved. In order tofully utilize the description information provided by a plurality of subsystems, a two-grade integrating strategy is adopted for the classification scores of the multi-branch subsystem, and a method of logistic regression is introduced to learn the optimal integrating strategy, so that the accuracy and the recall ratio of an integrating system are greatly improved.

An optical remote-sensing image, GIS (Geographic Information System) automatic registration and water body extraction integrated method comprises the following steps: 1, segmenting input optical remote-sensing images to obtain initial segmentation results of water systems; 2, performing local level set evolution segmentation to obtain a set R; 3, matching basic geographic information water system layer vector objects with objects in the set R, registering the images, executing the step 4 if the registration is successful, otherwise, returning to the step 2; 4, obtaining unchanged water bodies, suspected newly-added water bodies and suspected changed water bodies through buffer detection, further filtering out suspected water body objects, and confirming real changed water bodies and unchanged water bodies; 5, taking the multispectral values of corresponding pixels in unchanged water body objects as positive samples, randomly selecting the multispectral values of pixels on the optical remote-sensing image except the area in the set R as negative samples, training an SVM (Support Vector Machine) classifier, verifying whether the objects in a filtered suspected water body object set are water bodies or not, and obtaining water body results of segmentation, registration and extraction.

The invention discloses a rapid face and face key point joint detection method, comprising the following steps: 1, constructing a teacher network and a student network; 2, inputting a batch of training images, and carrying out data enhancement; 3, dividing positive and negative anchor frame samples according to a self-adaptive scale matching strategy; 4, mining positive and negative samples, calculating a multi-task loss function, and updating network parameters; 5, turning to the step 2 until the training converges to obtain a teacher network model; 6, repeating the steps 2 to 5, adding a transfer learning loss function by using the teacher network model, and training to obtain a student network model; and 7, in a test stage, inputting a test image to the student network model to obtain adetection result. The rapid face and face key point joint detection method can obtain face and key point detection results at the same time, and can increase the speed of the face recognition preprocessing process. The lightweight network provided by the invention is high in reasoning speed and can be deployed in embedded equipment with limited computing power.

The invention discloses a color and singular value feature-based face in-vivo detection method, and mainly aims at solving the problem, that the existing face authenticity identification technology is complicated in calculation and low in identification rate. The method is realized through the following steps: 1) marking positive and negative samples of a face database, and dividing the samples into a training set and a testing set; 2) segmenting face images in the training set into blocks and extracting the color features and singular value features of the small blocks in the training set in batches; 3) normalizing vectors of the extracted features and sending the normalized vectors into a support vector machine classifier to train so as to obtain a training model; and 4) carrying out feature extraction on data in the testing set and predicting the features by utilizing the training model so as to obtain a classification result. The color and singular value feature-based face in-vivo detection method is capable of improving the classification efficiency and obtaining higher classification result, and can be used for the face authenticity detection in social networks or real life.