1035results about How to "Fast training" patented technology

Dynamically filtering and classifying messages, as good messages, bulk periodicals, or spam. A regular expression recognizer, and pre-trained neural networks. The neural networks distinguish “likely good” from “likely spam,” and also operate at a more discriminating level to distinguish among the three categories above. A dynamic whitelist and blacklist; sending addresses are collected when the number of their messages indicates the sender is good or a spammer. A dynamically selected set of regular expressions input to the neural networks.

The invention discloses a gesture identification method and a gesture identification system, wherein the method comprises the following steps: step S1, a gesture detecting step: detecting a gesture in video stream in real time, marking a region in which the gesture is detected as an interested region; step S2, a gesture segmenting step: processing the interested region by utilizing skin color segmentation, and then performing edge detecting and outline extracting to obtain a point sequence of a hand-shaped outline; step S3, a gesture identifying step: firstly extracting a Fourier descriptor of the hand-shaped outline and then mapping the Fourier descriptor to a new vector of a characteristic space by utilizing PCA (principal components analysis), comparing the distance between the new vector and a gesture clustering center obtained by training, and judging a gesture type represented by the vector. The gesture identification method and the gesture identification system disclosed by the invention can improve the gesture identification accuracy and the gesture identification efficiency, and can effectively avoid background color interference.

A computer-implemented method for an object and activity query language wherein an object is a data type representing a thing or a being with a visual shape in an image or video frame and said activity is a data type representing an action or an event visually shown in an image or video or video frame, the method comprising the steps of storing a plurality of items in a raw data storage, said items comprising images and/or videos, processing said items in a processor and to generate and/or segment annotated information from said items and to extract object, activity and/or metadata information from said items in said first data storage, storing said annotated information in a secondary data storage, storing said extracted object, activity, and/or metadata information and said annotated information in a primary data storage, executing on a processor an identify function, wherein given a query item said identify function identifies said query item and/or finds a list of items similar to said first item; and a said query item comprises a video, a video frame, an image, a set of images, a template extracted from a video or image or images, an object, an activity, or annotated information, and displaying results of said identify function.

A computer-based training system and related methods utilizing a simulated offshore environment. The training system allows users to safely simulate various job tasks in which they are expected to perform prior to actually performing the job tasks in an offshore environment. Conditions can be altered each time to make the training session easier or more difficult, which also enables users to experience a different training scenario each time. The system allows users to make mistakes in a safe environment, before there are any real consequences, such as an injury or an environmental upset. The training system has reporting and updating capabilities to enable companies to determine the status of employee training and ensure that all of the procedures are current and consistent company-wide.

The invention provides a deep learning face verification method based on mixed training. The method comprises the steps that a face data set is prepared; face and face key point detection is conducted on all images; all faces are normalized to obtain a face image training set, the face image training set is partitioned into a training data set and a verification data set, a mean image of all face images is calculated; the mean image is subtracted from all the face images to obtain a mean training data set and a mean verification data set; a deep convolutional neural network is trained; a corresponding triad is generated for each face image, and a triad training data set and a triad verification data set are formed; the deep convolutional neural network is trained again; face and face feature point detection is conducted on two given images to be verified, the mean image is subtracted from the images, the images are input into the deep convolutional neural network, a network feedforward operation is conducted, and features are extracted; according to a selected threshold value, when the distance between the extracted features of the two images is larger than the threshold value, it is judged that the faces in the two images belong to a same person, and otherwise, it is judged that the faces belong to different persons.

The invention discloses an orchard pedestrian detection method based on a YOLOv3 algorithm. The method specifically comprises the steps of collecting pedestrian images in an orchard; Preprocessing thecollected images, and constructing a standard pedestrian detection data set; Placing a training set into the modified Darknet-53 network structure to extract pedestrian features; Generating a predicted pedestrian boundary frame by using a K-means clustering method, performing category prediction by using a binary cross entropy loss function, and performing multi-scale fusion prediction by using asimilar FPN network; Finally, removing redundant prediction boundary frames through the Soft-NMS, and outputting final prediction boundaries and categories. The pedestrian detection accuracy is high,the real-time performance is good, the robustness of a training model to a complex background is enhanced by aiming at data augmentation methods such as Random Eraging proposed in an orchard environment, and through the adopted Soft-NMS algorithm, the recall rate of detection can be increased, and the introduced group normalization Group Normalizations can reduce the requirements of a trained model on hardware.

The invention is applicable to the field of computer technology and provides a generation method and device for a decision network model of vehicle automatic driving. The method comprises the steps that a sample triad corresponding to each test moment is generated according to vehicle state information collected at each test moment, a preset vehicle movement set and a preset return function, all the sample triads are stored as sample data in a pre-established experience database, and all the sample data is subjected to clustering analysis; training samples are uniformly collected from each cluster obtained after clustering analysis of the experience database according to a preset sampling scale value, and a return accumulated value of each training sample is calculated; and according to all the training samples, the return accumulated value of each training sample and a preset deep learning algorithm, training is performed to obtain the decision network model of vehicle automatic driving. Therefore, the training efficiency of the decision network model and the generalization ability of the decision network model are effectively improved.

The invention discloses a wavelet transformation and deep learning-based rolling bearing weak fault diagnosis method. The method comprises the following steps: a rolling bearing vibration signal is obtained, and the acquired vibration signal is subjected to continuous wavelet transformation to obtain a time-frequency diagram; autocorrelation operation is performed on wavelet coefficients corresponding to each frequency on the time-frequency diagram to filter out noise interference and extract periodic fault components; the Hilbert transformation is used to perform envelope demodulation to obtain a fault characteristic frequency; a processed time-frequency diagram is input as a feature diagram, and categories of early faults are determined by training a deep learning classification model. Theoretical and experimental results show that fault categories can be determined accurately at an early stage when weak fault of a rolling bearing occur based on an improved wavelet time-frequency diagram used as the classification model for input training, right determination results can be given when the method is applied to different bearings, high training speed can be realized, and high actual application value can be achieved.

Embodiments of the invention relate to improvements to the support vector machine (SVM) classification model. When text data is significantly unbalanced (i.e., positive and negative labeled data are in disproportion), the classification quality of standard SVM deteriorates. Embodiments of the invention are directed to a weighted proximal SVM (WPSVM) model that achieves substantially the same accuracy as the traditional SVM model while requiring significantly less computational time. A weighted proximal SVM (WPSVM) model in accordance with embodiments of the invention may include a weight for each training error and a method for estimating the weights, which automatically solves the unbalanced data problem. And, instead of solving the optimization problem via the KKT (Karush-Kuhn-Tucker) conditions and the Sherman-Morrison-Woodbury formula, embodiments of the invention use an iterative algorithm to solve an unconstrained optimization problem, which makes WPSVM suitable for classifying relatively high dimensional data.

The utility model discloses a robot navigation system and a navigation method. The robot navigation system comprises a navigation network which is formed by a plurality of wireless access points, a wireless communication module which is used for transferring data and collecting the intensity sequence communicated with the wireless access points, a sensor which is used for checking that the robot meets barriers or not, and a position server which is used for storing the referenced intensity sequence and running the intricate position arithmetic, and is characterized in that the position server is connected with the wireless communication module and interacts with the navigation network. The navigation method is characterized in that the robot judges the next target position until reaches the destination by comparing the intensity sequence collected in real time with stored reference intensity sequence of the position points; when the robot meets barriers, the robot records and demarcates the intensity sequence of the position in order to avoid entering the position again, therefore achieving intellectual learning; the robot can upload the correlative information to the position server and achieve the assistant navigation position by the help of the database of the position server and the position arithmetic. The utility model is not likely to be affected by the environment and also has the advantages of low cost of maintenance.

The invention discloses a multilayer neural network language model training method and device based on knowledge distillation. The method comprises the steps that firstly, a BERT language model and amulti-layer BILSTM model are constructed to serve as a teacher model and a student model, the constructed BERT language model comprises six layers of transformers, and the multi-layer BILSTM model comprises three layers of BILSTM networks; then, after the text corpus set is preprocessed, the BERT language model is trained to obtain a trained teacher model; and the preprocessed text corpus set is input into a multilayer BILSTM model to train a student model based on a knowledge distillation technology, and different spatial representations are calculated through linear transformation when an embedding layer, a hiding layer and an output layer in a teacher model are learned. Based on the trained student model, the text can be subjected to vector conversion, and then a downstream network is trained to better classify the text. According to the method, the text pre-training efficiency and the accuracy of the text classification task can be effectively improved.

The invention discloses a microblog sentiment analysis method based on large-scale corpus characteristic learning. The method is characterized in that microblog texts is used to train a neural network based on Word2Vec, the neural network completes training by inputting the microblog texts, the neural network maps the input microblog to a corresponding word vector according to grammar contents, emotional tendency of current contents, and whether words are denied or not, and a matrix formed by word vectors corresponding to the words in the microblog is obtained. Synthesis of word vectors is performed on the matrix, so as to further obtain a feature vector corresponding to the microblog. Through applying the feature vector to training and prediction of a sentiment classification device, a relatively accurate microblog sentiment analysis is obtained. Compared with the prior art, the method is low in cost and high in analysis accurate rate, and is especially suitable for large-scale corpus, and is fast in training speed. Combining with grammar contents and emotional tendency, the method effectively processes negation relations, and improves accurate rate of sentiment analysis.

The invention discloses a word vector training method and system fusing word class information and position information. The method includes the steps that data is preprocessed to obtain a target text; the target text is subjected to word segmentation and word class tagging; the word class information and the position information are modeled; the word class information and the position information are fused on the basis of a skip-gram model based on the negative sampling strategy for word vector learning to obtain target word vectors, and the target word vectors are used for word analog task and word similarity task evaluation. The word class information and the position information of words are considered, on the basis of modeling the word class information and the position information of the words, the word class information of the words and the position information between word classes are fully used for helping training of the word vectors, and parameters are updated more reasonably in the training process.

The invention discloses a two-dimensional recursive network-based recognition method of Chinese text in natural scene images. Firstly, a training sample set is acquired, and a neural network formed bysequentially connecting a deep convolutional network, a two-dimensional recursive network used for encoding, a two-dimensional recursive network used for decoding and a CTC model is trained; test samples are input into the trained deep convolutional network, and feature maps of the test samples are acquired; the feature maps of the test samples are input into the trained two-dimensional recursivenetwork, which is used for encoding, to obtain encoding feature maps of the test samples; the encoding feature maps of the test samples are input into the trained two-dimensional recursive network, which is used for decoding, to obtain a probability result of each commonly used Chinese character in each image of the test samples; and clustering searching processing is carried out, and finally, the overall Chinese text in the test samples is recognized. According to the method of the invention, space/time information and context information of the text images are fully utilized, the text imagepre-segmentation problem can be avoided, and recognition accuracy is improved.

The invention discloses a domain adaptive semantic segmentation method based on similarity space alignment. The segmentation output of a source domain and the segmentation output of a target domain are respectively transformed into a similarity space, and the similarity space distribution of the source domain and the target domain is aligned to reduce the inter-domain difference, so that a semantic segmentation model with a better segmentation effect on an unsupervised target domain can be obtained. According to the method, the concept of similarity space is introduced into a cross-domain semantic segmentation task, the correlation between categories in a segmentation scene is better coded, and the discriminator is used for discriminating the similarity space of different domains, so thatthe segmentation network pays more attention to the structure, category coexistence and other information of an image, and the whole network can be trained end to end. The unsupervised domain self-adaptive semantic segmentation method based on similarity space alignment is innovated on the basis of the existing technical thought, the correlation space information of categories in a segmentation scene is fused, the segmentation performance is better, and the method has very high practical application value.

The invention provides an engineering standardized construction process control method on the basis of BIM (building information modeling). The method comprises a construction scheme analysis step, a BIM model association step, an on-site task prompt step, and an on-site data collection step. The invention also provides a corresponding system. According to the engineering standardized construction process control method and system on the basis of the BIM, each construction procedure is created or searching is performed to obtain a corresponding subtask, the subtask is associated with building components in a BIM model, according to instructing information such as subtask prompt, operation instruction and the like, the engineering which is completed and executed step by step is namely standard engineering; retention of on-site data collection can be performed in a engineering process, is convenient for tracing in the future, and remote control on live on-site construction is also convenient to be performed.

The invention discloses a fiber optic gyroscope temperature drift modeling method by optimizing a dynamic recurrent neural network through a genetic algorithm. The fiber optic gyroscope temperature drift modeling method by optimizing the dynamic recurrent neural network through the genetic algorithm comprises the following steps of (1) initializing network parameters, and establishing an improved Elman neural network model; (2) obtaining a training and testing sample; (3) training an improved Elman neural network, and optimizing model parameters through the genetic algorithm; (4) outputting forecasts of an fiber optic gyroscope, and compensating errors. The output of the fiber optic gyroscope processed through a denoising algorithm is trained by introducing the improved Elman neural model with self-feedback connection weight, constant iterative optimization is carried out on the model parameters through the genetic algorithm, and the optimal model is obtained according to the magnitude of the errors of the model under different parameters. According to the fiber optic gyroscope temperature drift modeling method by optimizing the dynamic recurrent neural network through the genetic algorithm, the complexity of the algorithm is taken into consideration, the accuracy of the fiber optic gyroscope temperature drift model is improved, the application of the fiber optic gyroscope temperature drift model in engineering is expanded, and certain practical significance is achieved.

The invention discloses a scene text end-to-end recognition method based on boundary point detection, and the method comprises the steps: extracting text features through a feature pyramid network, and generating candidate textboxes through a region extraction network; detecting a more accurate multi-directional bounding box of the text instance through a multi-directional rectangular detection network; secondly, detecting an upper boundary point sequence and a lower boundary point sequence of the text in the multi-directional bounding box; and finally, converting the text in any shape into ahorizontal text by utilizing the detected boundary point sequence for the subsequent attention mechanism-based sequence recognition network to performing recognizing, and finally, finding out the mostmatched word of the prediction sequence in the given dictionary by utilizing a cluster search algorithm to obtain a final text recognition result. According to the method, the scene text in any shapein the natural image can be detected and recognized at the same time under the condition that character-level labeling is not needed, the scene text comprises the horizontal text, the multi-directiontext and the curved text, and end-to-end training can be completely carried out.

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.