129 results about "Manual extraction" patented technology

A 1D-CNN-based ((one-dimensional convolutional neural network)-based) distributed optical fiber sensing signal feature learning and classification method is provided, which solves a problem that an existing distributed optical fiber sensing system has poor adaptive ability to a complex and changing environment and consumes time and effort due to adoption of manually extracted distinguishable event features, The method includes steps of: segmenting time sequences of distributed optical fiber sensing acoustic and vibration signals acquired at all spatial points, and building a typical event signal dataset; constructing a 1D-CNN model, conducting iterative update training of the network through typical event signals in a training dataset to obtain optimal network parameters, and learning and extracting 1D-CNN distinguishable features of different types of events through an optimal network to obtain typical event signal feature sets; and after training different types of classifiers through the typical event signal feature sets, screening out an optimal classifier.

A face emotion recognition method based on dual-stream convolutional neural network uses a multi-scale face expression recognition network to single frame face images and face sequences to perform learning classification. The method includes constructing a multi-scale face expression recognition network which includes a channel network with a resolution of 224×224 and a channel network with a resolution of 336×336, extracting facial expression characteristics at different resolutions through the recognition network, effectively combining static characteristics of images and dynamic characteristics of expression sequence to perform training and learning, fusing the two channel models, testing and obtaining a classification effect of facial expressions. The present invention fully utilizes the advantages of deep learning, effectively avoids the problems of manual extraction of feature deviations and long time, and makes the method provided by the present invention more adaptable. Moreover, the present invention improves the accuracy and productivity of expression recognition.

The embodiment of the invention discloses a malicious traffic detection implementation method and device based on deep learning. The method can include: obtaining the flow session of malicious code through dynamic sandbox technology; mapping the traffic session of malicious code to a gene map and extracting the map features; using the map features of traffic session for clustering and marking theclustering results by a malicious code family; training the preset depth learning model by using the tagged gene map of the malicious code family to establish the malicious traffic detection model; using the malicious traffic detection model to detect real-time network traffic, and realizing malicious traffic detection. Through the embodiment, many problems existing in the conventional detection technology such as artificial feature extraction difficulty, privacy disclosure, encryption and confusion difficult to identify, machine learning feature manual extraction and the like are solved to acertain extent, and the method has the characteristics of high robustness, high speed, high accuracy, low false alarm rate, cross-platform detection and the like.

The invention provides a deep-learning-based partial discharge defective image diagnosing method and system, comprising the following steps: 1) detecting the partial discharge signal of a piece of power equipment; obtaining a partial discharge defective image; 2) creating a partial discharge defective image sample database; extracting the training set and the testing set; 3) creating a deep convolution neural network model; using samples to do deep learning training and testing to obtain the connection weights and bias parameters of the network model; and 4) inputting the partial discharge defective image to be diagnosed into the network model obtained from the step 3; outputting and obtaining the partial discharge defect type of the image. According to the invention, through the utilization of the learning algorithm of the deep learning theory to complete the characteristic extraction task of a partial discharge defective image, it is possible to accurately and effectively identify the defect type of the partial discharge image without the reliance on manual extraction of the characteristic parameters, providing new solutions for insulation state diagnosing of power equipment.

The invention discloses a multiple features fused bidirectional recurrent neural network fine granularity opinion mining method. The method comprises the following steps of: capturing comment data of a specific website through internet and carrying out labelling and preprocessing on the comment data to obtain a training sample set; carrying out training by using a Word2Vec or Glove model algorithm to obtain word vectors of the comment data; carrying out vectorization after carrying out part of speech labeling, dependence relationship labeling and the like; and inputting the vectors into a bidirectional concurrent neural network to construct a bidirectional recurrent neural network fine granularity opinion mining model. According to the method, attribute words in fine granularity opinion mining is extracted and emotional polarity judgement is carried out through the training of a model, so that plenty of model training time is further saved and the training efficiency is improved; no professionals are required to carry out manual extraction on the attribute words, so that a lot of manpower cost is saved; and moreover, the model can be trained by using a plurality of data sources, so that cross-field fine granularity opinion analysis can be completed, thereby solving the problem of long-distance emotional element dependency.

The present invention discloses an improved full-convolutional neural network-based power transmission line insulator state recognition method. The method includes the following steps that: S1, the picture of a power transmission line insulator is collected through an unmanned aerial vehicle; S2, classification regression and position regression are performed on the image through a target detection network Faster R-CNN so as to intercept a separate insulator picture; S3, semantic segmentation is performed on the insulator picture through using a full-convolutional neural network; S4, fine segmentation is performed through a full-connection condition random field; S5, noise points in the image are filtered by using a morphological operation method; and S6, the insulator is classified through a deep learning classification network, and the status of the insulator is determined. According to the method of the invention, training and parameter adjustment and optimization are performed on labeled insulator pictures; the status of the power transmission line insulator can be effectively identified; the subjective influence of manual setting of thresholds and the randomness of manual extraction of features in traditional insulator status recognition can be avoided; the efficiency of line inspection can be significantly improved; and the difficulty of the line inspection can be decreased.

A text semantic analysis method and system can realize semantic analysis of text data base on lexical level and sentence level. Aiming at the semantic analysis at the lexical level, the invention firstly adopts an improved word segmentation algorithm to solve the problem that English words are segmented only by spaces. Secondly, based on word segmentation, TF-IDF modeling is performed to obtain weight value; Then the text is vectorized by weighting and summing the weight value and the word vector trained by Word2Vec, and finally the document similarity is solved. At the same time, the invention considers the contribution degree of the vocabulary to the document content and the semantic status to calculate the similarity degree of the document, the result has higher accuracy, and provide agood foundation for subsequent text clustering. The present invention extracts subject-predicate object structure based on text segmentation, part-of-speech tagging, syntactic analysis and dependencyrelation for sentence level semantic analysis. The invention realizes the extraction of subject-predicate-object structures of various sentence types in all aspects, and realizes the noun expansion function, which is more consistent with the manual extraction result.

A medication carrier for administering individual doses of therapeutic products to a patient, in a non-sequential fashion. The medication carrier comprises a receptacle which stores individually sealed, unit dose packages in random order, allowing each unit dose package to be easily accessed and released in response to automated or manual extraction. The medication carrier includes an array of stalls and retaining means for securing the sealed unit dose packages within the stalls until a scheduled dosing time. The unit dose packages are oriented such that identifiers imprinted thereon can be easily read without removing the packages from the carrier.

The invention relates to a face identification method and an attendance system based on a deep convolution neural network. The system comprises that user information and face sample labels are input into a user management server and then sent to a central server; in the central server, the pre-processed sample labels are used to establish a face identification model based on the deep convolution neutral network; employees carry out the online face identification through a trained neutral network on attendance machines at each site, and face identification results will be returned to the user management server by an internal local area network; and management staffs can carry out checking, modification and other operations to attendance records on the user management server. A face identification algorithm based on the deep convolution neutral network used by the invention can avoid problems such as incomplete and uncertain characteristic description caused by traditional manual extraction, can also make use of advantages of a receptive field and weight sharing, and can increase a face identification rate so as to increase an accuracy rate of the attendance system.

The invention belongs to the crossing field of artificial intelligence and data mining, and particularly relates to a text sentiment analysis method based on a bidirectional long-short term memory neural network. The method comprises the following steps: in a text mapping word vector mapping framework, expressing a text as a word vector matrix; Constructing an improved convolutional neural network, and performing feature extraction on the mapped word vector matrix; Training the extracted features by using a bidirectional long-short term memory recurrent neural network until the loss function is minimum; And adding a global mean pooling layer and a classification layer behind the recurrent neural network so as to output sentiment classification of the text. According to the method, deep learning is applied to text sentiment analysis, the limitation of manual extraction of text sentiment characteristics is avoided, the advantages of a convolutional neural network and a recurrent neural network are combined in the field of deep learning, and the network is improved, so that the method can be better applied to a text processing task.

The invention discloses a fault type identification method of a transmission line based on a convolutional neural network. A convolutional neural network algorithm belongs to a deep learning algorithm. The deep learning algorithm is applied to the field of fault type identification of transmission lines, manual extraction of fault features is not needed for fault type identification, and a conventional line fault type identification method based on an artificial intelligence algorithm needs extraction of fault features in advance, so the invention simplifies the structure of fault type identification. The invention improves the identification efficiency of the line fault type identification, and in the application of the line fault type identification algorithm based on the deep learning,a plurality of parameters will cause the algorithm to be different in the training process, and the invention intends to optimize the line fault type identification algorithm. The method reduces the error rate of line fault type identification, and different activation functions can make the training error completely different. The method uses different activation functions to train the line faulttype identification, and finds the optimal activation function.

The invention provides a road network map generating method, a road network map generating system, road network map generating equipment and a storage medium. The method comprises the following stepsof obtaining collection data of a laser radar on a mobile collecting device to obtain multiframe point cloud data; obtaining collection data of an inertial measurement unit on the mobile collecting device corresponding to each frame, wherein the collection data of the inertial measurement unit includes the linear velocity and the angular velocity of the mobile collecting device at the front frame;generating a high-precision map according to the point cloud data and the collection data of the inertial measurement unit; extracting a road network map from the high-precision map. The invention provides the technical scheme for generating a road network map; the road network map is automatically extracted and generated; the problems of low precision and low efficiency of a manual extraction method in the prior art is solved; the technical scheme can be used for road network map extraction in a specific open industrial region, and has the characteristics of high automation, high robustnessand high precision.

A system is provided for obtaining parameters about the position and orientation of a camera, where the camera has a fixed position and orientation, by using the image coordinates and world coordinates of markers included in an image photographed by the camera. This system uses a live-image display mode for successively obtaining images input to the camera and a still-image display mode for obtaining an image input to the camera at predetermined time. The markers can be extracted by using the image obtained in either mode. The system can select either automatic-extraction mode or manual-extraction mode for extracting the marker in live images, so that the markers can be extracted with a high degree of accuracy under all conditions to obtain the position and orientation parameters of the camera.

The invention discloses a method for automatically extracting a terrain characteristic line according to vector contour line data. The method for automatically extracting the terrain characteristic line according to the vector contour line data comprises the following steps of: determining characteristic points of terrain; classifying the characteristic points to obtain ridge points and valley points; determining judgment factors for generating a ridge line and a valley line; and automatically generating the ridge line and the valley line. The characteristic points of the terrain such as the ridge points and the valley points are determined by using the characteristics of curved lines of a vector contour line and a geometrical morphology analysis method, and then connected to form the terrain characteristic line such as the ridge line and the valley line. The method is high in identification speed, the extracted terrain characteristic line such as the ridge line and the valley line is in accordance with actual terrain change, and errors caused by manual extraction can be effectively eliminated. The method can be used for extracting a terrain contour line from digital terrain data. Furthermore, the method is convenient to operate, simple, and great in practical application value.

The invention discloses a digital line graph mapping method, relating to the fields of mapping and remote sensing. The digital line graph mapping method comprises the following steps of: verifying flight by using onboard LIDAR (Laser Intensity Direction and Ranging) equipment to realize calibration of equipment parameters; establishing a point cloud data error model according to an onboard LIDAR point cloud generation principle, acquiring a parameter correction value according to overall adjustment to realize integral accuracy optimization for point cloud data, and acquiring a point cloud with optimized accuracy; carrying out filtering and sorting treatment on the ground and buildings, acquiring a ground point cloud and a building point cloud, and establishing a digital ground model according to the ground point cloud; carrying out aerial triangulation encryption on an air photo according to the digital ground model and POS (Position and Orientation System) auxiliary positioning information to realize integral accuracy optimization for fixed-position and fixed-attitude data of the air photo, and acquiring the air photo with optimized accuracy; accurately matching the point cloud with the optimized accuracy and the air photo with the optimized accuracy; and establishing a mapping environment of a point cloud and air photo fused mode, realizing semi-automatic extraction of vector lines and total-manual extraction of residual ground feature elements in the fused mode mapping environment, and generating a digital line graph product.

The invention discloses a kit for extracting nucleic acid from bacteria by using a paramagnetic particle method and an extracting method. The kit comprises a bacteria lysate, a magnetic bead binding solution, a magnetic bead scrubbing solution and a nucleic acid eluent. The bacteria lysate comprises sodium dodecyl sulfate, ethylenediaminetetraacetic acid, tri-hydroxymethyl aminomethane and sodium chloride; the magnetic bead binding solution comprises polyethylene glycol-8000 and sodium chloride; the magnetic bead scrubbing solution comprises ethanol; and the nucleic acid eluent comprises tri-hydroxymethyl aminomethane and ethylenediamine tetraacetic acid. The kit comprises the unique bacteria lysate which has a strong lysis function for the bacteria; and the kit can be used for manual extraction and instrument extraction by using a commercially available nucleic acid isolation machine, high-purity and high-concentration bacteria nucleic acid can be extracted.

The present invention discloses an improved deep Boltzmann machine-based pulmonary nodule feature extraction and benign and malignant classification method. The method includes the following steps that: step A, pulmonary nodules are segmented from CT images through using a threshold probability image graph method, so that regions of interest (ROI) are obtained, and the regions of interest are cut into nodule images of the same size; and step B, a supervised deep learning algorithm Pnd-EBM is designed to realize the diagnosis of a pulmonary nodule, wherein the diagnosis of the pulmonary nodule further includes three major steps: B1, a deep Boltzmann machine (DBM) is adopted to extract the features of the ROI of the pulmonary nodule which have deep expression abilities; B2, a sparse cross-entropy penalty factor is adopted to improve a cost function, so that the phenomenon of feature homogenization in a training process can be avoided; and B3, an extreme learning machine (ELM) is adopted to perform benign and malignant classification on the extracted features of the pulmonary nodule. The improved deep Boltzmann machine-based pulmonary nodule feature extraction method is superior to a traditional feature extraction method. With the method adopted, the complexity of manual extraction and the difference of feature selection can be avoided, and references can be provided for clinical diagnosis.

The disclosed medical image template matching method based on microcomputer comprises: pre-processing image, training sample with SVM instead of RNN, extracting feature by complex wavelet, and finding out the optimal matching. This invention overcomes defects in prior art, improves speed and robustness, and has strong adaptability.

The invention discloses a deep learning-based question matching method and apparatus. The method comprises the steps of obtaining an input question of a user, and performing semantic analysis on the input question to obtain words of the input question; obtaining a similarity matching matrix between the words in the input question and words in a standard question, and performing hierarchical convolution based on the similarity matching matrix to obtain a semantic granularity matching feature between the input question and the standard question; based on a convolutional neural network, performing accurate matching on the input question and the standard question according to feature information of the input question and feature information of the standard question to obtain a fine granularity matching feature between the input question and the standard question; and obtaining a matching degree between the input question and the standard question according to the semantic granularity matching feature and the fine granularity matching feature. According to the method, the extracted features better meet actual demands; large cost for manual extraction of the features is omitted, and the accuracy of a matching result is improved.

A cover, carriage, or faceplate for a napkin dispenser is provided. The cover, faceplate or carriage defines one or more pick zones that facilitate manual extraction of different numbers of multiple napkins from the dispenser during each dispensing event. Preferably, the cover, carriage or faceplate cooperates with a container configured to receive a stack of napkins and is maintained adjacent an end or edge of the stack. More preferably, the cover, faceplate or carriage is configured to cooperate with containers having various shapes and / or configured for countertop, hanging, or in-counter implementation of the container to allow retrofitting of currently available single quantity dispensing napkin dispensers.

The invention provides a mobile carrying robot for logistics storage. The robot comprises a four-wheel trolley chassis, a steering engine for steering, a large mechanical arm, a small mechanical arm,a mobile portal frame, a front axle structure, a trolley body structure, a damping device and other assemblies. Firstly, the robot actively moves to the vicinity of objects, the large mechanical arm is used for clamping and extracting the objects, then the objects are placed on a cargo circulating storage frame, a system performs the operation for the next round according to the sizes of the objects, if the objects need to be combined and classified, the objects are firstly moved to the corresponding positions through the circulating storage frame, then the mobile portal frame at the rear carries the small mechanical arm to clamp the objects and place the objects into corresponding lattices of a storage cabinet, if the objects need to be moved to the next station in batches, a push-pull door at the rear portion of the storage cabinet can be manually opened for manual extraction, if the objects need to be supplemented to a warehouse large-scale goods shelf, "reverse operation" is performed, the objects are clamped by the small mechanical arm to be placed on the cargo circulating storage frame, and then the objects are moved and carried by the large mechanical arm.

The invention discloses a method for classifying surface EMG signals based on CNN and LSTM. The method includes utilizing a sliding window to convert a time sequence into a 'data-tag' pair, applying aHamming window to the surface EMG signals in each time window, using the fast Fourier transform to calculate the time-frequency spectrum Spectrogram, superimposing and integrating the time-sequence spectrum data along the time axis direction, sending the data to a convolutional neural network to complete the local spatial high-level feature extraction and obtain high-dimensional features, expanding the high-dimensional features along the data superposition dimension, restoring the data to the time sequence, feeding the data into a long and short time memory network, extracting the sequence features, sending the sequence features into a fully connected network for further feature extraction and integration to obtain the fully extracted high-dimensional features and feeding the fully extracted high-dimensional features into a Softmax function to get a final classification result. The core of the method is based on a deep learning algorithm, and the classification decoding accuracy is obviously improved by further analysis and extraction on the traditional manual extraction features.

The invention provides a ship type identification method and a ship type identification system based on migration learning. The method comprises the following steps: migrating a trained convolution neural network model; adjusting parameters and structure of the convolution neural network model in combination with a ship image to obtain a migration learning model; The ship image is grayed, normalized and annotated, and the data set is obtained. Training the transfer learning model with the data set according to a back propagation method; The image to be measured is input into the trained transfer learning model for type recognition. The invention adopts a transfer learning method, which can actively extract features of warship images without manual extraction of features, reduces workload and leakage possibility of personnel, avoids the problem of poor network classification result caused by de novo network design and lack of a large number of warship data sets, and improves accuracy ofclassification.

The invention relates to a pedestrian re-identification method based on the fusion of attribute features and weighted block features, comprising the following steps: constructing an attribute featureextraction sub-network, which integrates the manually extracted features and the features extracted by a depth neural network; using A weighted cross-entropy loss function to train the attribute feature extraction subnetwork; constructing A block-based feature extraction sub-network, which can fuse the depth features of multiple blocks. Training the sub-network based on block feature extraction, setting the weighted fusion layer of local loss function, learning different weights independently, and then endowing each local loss function; training the whole network to extract the pedestrian feature representation which combines the attribute feature and the depth feature based on the block. The invention is reasonable in design, effectively combines attribute features and depth features, optimizes the loss function calculation method, obtains a good pedestrian recognition result, and greatly improves the overall matching accuracy of the system.

The invention discloses a tumor region segmentation method and system for a liver CT image based on a cascaded full convolutional network. According to the method, automatic segmentation of a liver tumor region is realized by using a full convolutional network model. The method comprises the following steps: performing preprocessing operations such as filtering and sharpening enhancement on a CT image; training a cascade network by using the preprocessed CT data set; achieving liver region segmentation by using two stages of FCN networks and segmenting a tumor region from a liver region of interest. A first-stage FCN network uses a variable pooling method so that more liver features are reserved and liver segmentation precision is improved; and a second-stage FCN uses hole convolution replaces an original convolution layer and a pooling layer, so that the position information of the target can be reserved while a image is reduced and the feeling domain is increased. A large number of features do not need to be manually extracted, and the tumor region can be effectively and accurately segmented from the liver CT image.

The invention discloses a clip defect detection method based on depth learning and machine vision fusion. The invention relates to the field of automatic detection technology, which aims at overcomingthe problem of low detection accuracy caused by the manual extraction of features in the clip quality detection method based on machine vision. The method comprises the following steps: (1) image acquisition of the clip workpiece; (2) tooth defect detection of the clip through depth learning; (3) size defect detection based on the method of machine vision; (4) obtaining of the results of defect detection through statistics. The technical scheme of the invention is simple to realize and can effectively improve the detection accuracy rate of the clip defects.

The present invention relates to a lung sound classification method, system and application based on a convolutional neural network. An image identification convolutional neural network is applied tolung sound classification, a general model is used to solve huge defects that a traditional algorithm has manual extraction of features and join experience limitation conditions and cannot be generally suitable for all the samples, and solve problems that a traditional algorithm cannot be enhanced with a data set and cannot be subjected to autonomic learning enhancement.

The invention discloses a multi-polarization radar high-resolution range profile target recognition method based on LSTM. The invention can fuse HRRPs of different polarization channels, extract the correlation characteristics of different channels, fully mine the polarization characteristics of targets, simultaneously realize automatic extraction of features, avoid manual extraction of features,and improve target recognition performance. As all channel of that multi-polarization radar are echoes of the same target, LSTM is applied to multi-polarized HRRP target recognition by taking different channels as sequences. By using LSTM to automatically fuse HRRP information of different polarized channels and fully mine the polarization characteristics of targets, the depth and correlation characteristics of targets can be extracted automatically, and the limitation of manual extraction of features can be avoided.

The invention discloses a method for diagnosing leading screw faults, belongs to the technical field of machinery fault diagnosis, and especially relates to ball screw fault diagnosis. The problem that a conventional intelligent leading screw fault diagnosis system is difficult for manual extraction of characteristics and limited in applying shallow-layer network non-linear expression capability. A diluted self-coding deep neural network structure is employed, and a network identification model employs a Softmax regression classifier to determine the hidden layer quantity of a network structure. The quantity of the input terminals and the quantity of the output terminals of a fault diagnosis model are determined, a training sample set, pre-training, fine tuned training and a fault diagnosis model test sample set are prepared, and the fault diagnosis performance of the fault diagnosis model is tested. Data sections in the test sample are integrally input continuously and sequentially, and the output quantity of the model is recorded to obtain an actual output list of the model. The ideal output list of the model is compared with the designed output list to obtain the fault performance test and evaluation result of the fault diagnosis model.