503 results about "Sequence Feature" patented technology

The invention provides methods for sorting polynucleotides from a population based on predetermined sequence characteristics. In one aspect, the method of the invention is carried out by extending a primer annealed polynucleotides having predetermined sequence characteristics to incorporate a predetermined terminator having a capture moiety, capturing polynucleotides having extended primers by a capture agent that specifically binds to the capture moiety, and melting the captured polynucleotides from the extended primers to form a subpopulation of polynucleotides having the predetermined sequence characteristics. In another aspect, the method of the invention is carried out on a population of tagged polynucleotides so that after a subpopulation is selected, the members of the subpopulation may be simultaneously analyzed using the unique tags on the polynucleotides to convey analytical information to a hybridization array for a readout.

Disclosed are methods for detecting and/or prioritizing phenotype-causing genomic variants and related software tools. The methods include genomic feature based analysis and can combine variant frequency information with sequence characteristics such as amino acid substation. The methods disclosed are useful in any genomics study; for example, rare and common disease gene discovery, tumor growth mutation detection, personalized medicine, agricultural analysis, and centennial analysis.

The invention claims to protect a character recognition system and method based on the combination of a neural network and an attention mechanism, the system comprising: a convolution neural network feature extraction module, which is used for spatial feature of character image; The spatial features extracted by the convolution neural network are input to the bi-directional long-short memory network module, and the bi-directional long-short memory network can extract the sequence features of characters. The extracted feature vectors are semantically encoded, and then the attention weights of feature vectors are assigned through the attention mechanism, so that the attention is focused on the feature vectors with higher weights. In the decoding part of the model, the features extracted fromattention and the prediction information of the previous time are used as the inputs of the nested long-short memory network. The purpose of using the long-short memory network is to keep the temporal characteristics of the eigenvectors and make the attention points of the model constantly change with time. In the decoding part, the features extracted from attention and the prediction informationof the previous time are used as the inputs of the nested long-short memory network. The invention can more accurately detect the text area in the natural scene, and has good detection effect on thesmall target text and the text with small tilt angle.

The invention discloses a motion history image and convolutional neural network-based behavior identification method. The method comprises the following steps of S1, obtaining an input original videoimage, and processing the input original video image through a motion history image-based behavior sequence feature extraction method; and S2, performing behavior identification on a local motion history image by adopting a deep convolutional neural network-based method to obtain a behavior type classifier, and finally outputting a behavior identification result through the behavior type classifier. The motion history image is calculated in an original video sequence, so that the to-be-processed information amount is reduced and key time-space information in behavior identification is extracted; and by taking the motion history image as an input, a deep convolutional neural network is established, then the network is trained by utilizing a stochastic gradient descent method and a Dropout policy, and finally behavior type identification is realized. The method can be effectively applied to online real-time behavior identification.

The present invention provides a face identification method in a video and a face identification device in the video. The face identification method in the video comprises the steps of carrying out the feature extraction on the target faces in multiple frames of images of the video to generate a plurality of feature vectors corresponding to the target faces in the multiple frames of images respectively; carrying out the timing sequence feature extraction on the plurality of feature vectors to transform the plurality of feature vectors into a feature vector of a predetermined dimension; utilizing a classifier to judge the feature vector of the predetermined dimension to identify the target faces. The face identification method and device in the video adopts a dynamic identification mode, and the information of the frames of images are complemented by utilizing the feature that the frames of images in the video have relevance on a time dimension, thereby improving the face identification accuracy.

Methods for rapidly detecting single or multiple sequence alleles in a sample nucleic acid are described. Provided are all of the oligonucleotide pairs capable of annealing specifically to a target allele and discriminating among possible sequences thereof, and ligating to each other to form an oligonucleotide complex when a particular sequence feature is present (or, alternatively, absent) in the sample nucleic acid. The design of each oligonucleotide pair permits the subsequent high-level PCR amplification of a specific amplicon when the oligonucleotide complex is formed, but not when the oligonucleotide complex is not formed. The presence or absence of the specific amplicon is used to detect the allele. Detection of the specific amplicon may be achieved using a variety of methods well known in the art, including without limitation, oligonucleotide capture onto DNA chips or microarrays, oligonucleotide capture onto beads or microspheres, electrophoresis, and mass spectrometry. Various labels and address-capture tags may be employed in the amplicon detection step of multiplexed assays, as further described herein.

The invention discloses an interactive video retrieval method based on unsupervised learning and semantic matching features, which comprises the steps of extracting image bottom layer features and model matching sequence features from the video image frame level of a video database; extracting semantic matching features from the high-level semantic level of image bottom layer features; performing unsupervised learning on the extracted model matching sequence features and semantic matching features, establishing retrieval and direct retrieval based on unsupervised learning, and forming an interactive interface by related feedback. The middle layer features, high layer features, unsupervised retrieval mechanism and interactive mechanism of an integrated video compose a set of new complete video retrieval system, which precisely measures the time-space sequence information of a video object, therefore better retrieval effect is achieved, the semantic understand of a sports video theme is developed, the online calculating complexity and retrieval time of a system are reduced, and the interactive interface greatly improves the retrieval performance of the system.

The invention relates to an instruction-set-irrelevant binary code similarity detection method based on a neural network. The method mainly includes the steps that binary files are reversely analyzed, and 24 features on the 9 aspects of call relation features, character string features, stack space features, code scale features, path sequence features, path basic features, degree sequence features, degree basic features and map scale features of functions are extracted. Based on expression forms of the features, the similarity degrees of the 24 features of the two to-be-compared functions are calculated through 3 similarity calculation methods and serve as input vectors of an integrated neural network classifier, and predicted values of the overall similarity between the two functions are acquired and ranked. Compared with the prior art, dependence on specific instruction sets is avoided, similarity detection of binary files of different instruction sets can be achieved, accuracy is high, the technology is simple, and popularization is easy.

The invention discloses a method and a device for detecting malicious software of an intelligent mobile terminal. The user operation is compared with a software behavior and according to behavior characteristics, the malicious software is identified. For continuously emerging new viruses or virus mutations, compared with a conventional detection method, the method and the device do not depend on a malicious code section feature library and a malicious function calling sequence feature library and are also limited to known malicious software, and thus, the method and the device have high universality and high detection rate, malicious behaviors such as malicious charging, privacy stealing and remote control can be well identified and reliable guarantee is provided for safety use of the intelligent mobile terminal.

The invention discloses a sign language recognition method based on a space-time attention mechanism. The method comprises the following steps: firstly, sampling a sign language video into a continuous sign language sequence with a uniform length as input of a model; and then, inputting the video frame sequence into a spatial attention network formed by 3D residual blocks, so that the network canautomatically pay attention to a salient region in a space. Analyzing the extracted convolution features through a ConvLSTM convolution long-short-term memory network, extracting long-time sequence features, and distributing time attention weights of different video frames to generate feature representation of the video; and finally, the generated feature representation passes through a Softmax classifier, and classification categories are output in a vector form. According to the invention, the interference of redundant information on recognition can be reduced, and the recognition accuracy is improved.

The invention relates to a recommendation sorting method based on a wide and deep gate loop joint model, and belongs to the technical field of natural language processing. According to the method, Sina Weibo data are first crawled for preprocessing to obtain a topic feature set; then generalized cross-feature conversion is used to memorize topic features, and the same is input into a linear module; then an embedding vector is learned for each classification feature, all the embedding vectors are connected with dense features, dense vectors generated by connection are input to a deep module formed by gate loop units; and finally, parameters in linear and deep loop processes are simultaneously optimized, and a recommendation sorting result is obtained through joint training on the model. According to the method, the gate loop units are used for feature generalization, the problem that conventional methods mostly do not consider sequence features of dynamic time sequences is alleviated, abetter recommendation result is achieved as a whole, and recommendation efficiency is also improved to a certain extent.

The invention discloses a malicious code detecting system and method based on dynamic instrumentation. The method comprises the steps that a software to be detected is uploaded on a system simulator of a mobile terminal; a framework of the software to be detected of the system simulator is ordered again by an instrumentation manger of a server, an instrumentation strategy file is implanted into the system simulator, and one or more probe functions monitor an API function of the framework according to the instrumentation strategy file; a sound monitor of the server communicates with the mobile terminal and receives the operation behaviors and content of the software to be detected which are conducted to the API function and obtained by the probe functions; the detected content is compared with a malicious API sequence feature library, if the detected content exists in the malicious API sequence feature library, the software is marked to have malicious behavior operation, and if the detected extent does not exist in the malicious API sequence feature library, the software is marked to be safe, so that the safety of the software to be detected is determined. By means of the system and method, it is achieved that the software to be detected is installed and operated in the system simulator of the mobile terminal, and malicious codes are detected through instrumentation, so that it is avoided that safety hazards or hidden dangers are caused to the mobile terminal.

The invention discloses an emotion recognition method of multi-channel electroencephalogram (EEG) data and an electronic device. The method includes the steps: sending frequency domain features of channels of extracted EEG data into a sequence feature extractor so as to obtain hidden-layer feature vectors of the channels of the EEG data, capturing association between the EEG channels and emotion categories, and sending the association and the hidden-layer feature vectors to an automatic encoder so as to obtain probability value vectors of the emotion category corresponding to the EEG data. Long-term and short-term memory networks are utilized to learn the non-linear relationship between the EEG channels on the EEG channel sequence, the high-dimensional EEG features related to the emotion categories are extracted, the key information in the multi-channel EEG signals are selected automatically, and EEG feature field differences which are caused by instability of the EEG data and environmental change are reduced.

The invention discloses an entity identification method, which comprises the following steps: obtaining an LSTM-based entity identification model after training is completed; wherein, the LSTM-based entity identification model is trained by using the annotated training corpus; the LSTM-based entity identification model is trained by using the annotated training corpus. Inputting the text to be identified into the LSTM-based entity recognition model after the training is completed, and obtaining the probability that each character in the text to be identified belongs to a label label; Inputtingthe probability into the CRF model to obtain marks of each character; LSTM networks rely heavily on data, The size and quality of the data will also affect the training results of the model, Combining LSTM model and CRF model, The LSTM model is used to solve the problem of extracting sequence features, and the CRF model can effectively utilize the sentence-level markup information. The LSTM + CRFmodel improves the execution efficiency of the dialogue system, and at the same time, the entity recognition and word segmentation are realized, which improves the accuracy and efficiency of entity recognition.

The invention discloses a method for quickly counting and automatically examining and verifying target contents of long video. The method includes steps of extracting key frame sequences of the long video and target video; extracting local binary-pattern histogram features and spatial gray sequence features of the key frame sequences; quickly searching the target video in the sequence of the long video; roughly positioning the target video; precisely positioning the target video in the sequence of the long video by a temporal-spatial constraint maximal connected region process; judging whether positioning for the target video in the long video is completed or not; counting the contents of the long video and automatically examining and verifying suspicious targets according to positioning results. The method has the advantages that by the method, the large-scale long video can be automatically processed, the contents of the long video can be quickly counted, and contents of the suspicious targets can be automatically examined and verified.

The invention discloses a protein-ligand binding site prediction algorithm based on deep learning. For protein to be predicted, the algorithm comprises the steps of: firstly, extracting sequence features and a distance matrix; distributing the sequence features to each residue through adoption of a sliding window method; and inputting the features corresponding to the residues into a residual neural network and a hybrid neural network one by one, and inputting output results of the residual neural network and the hybrid neural network into a Logistic regression classifier to obtain a final result, namely the binding probability corresponding to each residue in the protein. According to the method, a classic bidirectional long-short-term memory network and a residual neural network are fused, the fused network can process heterogeneous protein sequences and structural data at the same time, and complementarity of sequence features and structural features is mined. Compared with an existing method, the protein-ligand binding site prediction algorithm has higher prediction precision, and has good generalization performance for data sets of different ligands.

The invention discloses a writing feature and sequence feature combined Chinese sentiment new word recognition method and system. The method comprises the following steps: for input text clauses, expressing the text clauses as sequences of various features (such as word, part of speech and the like) on the basis of writing features of writers of sentiment words and sequence features of the sentiment words; in allusion to the feature expressed text clauses, outputting sentiment word label sequences corresponding to the text clauses by utilizing a linear chain condition random field model, wherein the linear chain condition random field model is obtained on the basis of a traditional sentiment word-containing text through training; on the basis of sequences of words and label sequences of the sentiment words in the text clauses, recognizing the sentiment words in the text clauses by utilizing a finite state machine so as to form a sentiment word set; and finally filtering the sentiment word set by utilizing a Chinese old word bank, and taking the sentiment words which do not appear in the Chinese old word bank as Chinese sentiment new words. Through the embodiments of the invention, the technical problem of how to improve the sentiment new word recognition precision and the recall rate is solved.

The present invention relates to a high-voltage circuit breaker mechanical failure diagnosis method based on vibration signal analysis. The high-voltage circuit breaker mechanical failure diagnosis method based on the vibration signal analysis comprises the steps of measuring the time sequences and the time difference sequences of the occurrence moments of vibration events during the opening and closing processes of a circuit breaker under a normal working condition experiment and a typical fault experiment, and creating a time sequence feature library and a time difference sequence feature library; measuring a time sequence and a time difference sequence of the occurrence moments of the vibration events of an actually operated circuit breaker; comparing the time difference sequence of the occurrence moments of the vibration events during the opening and closing processes of the actually operated circuit breaker with the time difference sequences in the time difference sequence feature library to determine the types of faults; comparing a time sequence of the fault types under the faults of varying degrees in the time sequence feature library with the measured time sequence to determine the degrees of the faults. The method more visually solves the problems that the judgement of the conventional circuit breaker diagnosis method based on the vibration is too simple, and the accuracy is low, and can determine the degrees of the faults.

The invention relates to a resume extraction method based on a deep neural network, and the method comprises the steps: data preprocessing: obtaining a resume data text, carrying out the word division, obtaining a word vector feature and a word sequence feature, and obtaining a word vector data set and a word sequence data set; training a deep neural network: training to obtain a deep neural network training model, simultaneously taking the word vector data set and the word sequence data set as feature input of the deep neural network training model, taking semantic features obtained through training as output features, and performing entity labeling by utilizing the output semantic features to obtain entity labels; label matching analysis: according to a corresponding extraction rule in apre-trained information element extraction rule base, matching information element phrases in the marked resume data text, and returning information element labels and information element phrase pairs of the resume data text; According to the method, the word vectors and the word sequences are used as input characteristics, and the recognition accuracy is improved by combining a deep neural network and a rule-based text analysis technology.

The invention relates to the field of natural language processing, and particularly relates to an enterprise industry classification method based on fully-automatic learning. The method of the invention uses a recurrent neural network to carry out fully-automatic feature learning on a business scope of a to-be-classified enterprise and automatically divides the to-be-classified enterprise to a corresponding industry scope. The method breaks domain barriers of the natural language and the neural network technology, and automatic enterprise industry classification based on natural language analysis is realized. The method does not need to manually select features, the defect that manual feature selection in the traditional method deviates from a specific sample is solved, the developer pays more attention to the task itself and data reserve, the working efficiency is enhanced, and as the recurrent neural network is used, longer sequence features can be captured by the model, the industry classification accuracy can be greatly enhanced, and more accurate basic data are provided for business and finance analysis.

The invention relates to a character-level nested deep network-based text classification method. The method comprises the following steps of S1, constructing a character vector matrix table; S2, performing short text preprocessing; S3, extracting high-dimensional sequence features by improved Resnet; and S4, performing LSTM network classification. The character level-based text conversion can effectively perform conversion on all texts; compared with a conventional vector space model, the dimension is reduced remarkably; all the texts can be effectively converted; and low-frequency words are not ignored. In addition, the improved Resnet can self-learn a feature extraction method; and compared with conventional methods such as a TF-IDF formula, a mutual information quantity, an informationgain, an x2 statistical quantity and the like, the extracted features are more effective and abstract. Finally, the LSTM network classification can consider a sequence relationship between the words,so that the text classification can be performed more accurately.

The embodiment of the invention provides an iris recognition method and terminal, and relates to the technical field of terminals. The method and terminal can improve the accuracy and robustness of iris recognition. The method comprises the steps: enabling the terminal to carry out the contour detection of an iris region of eyes, determining the rotation angle of the eyes through an angle between the connection line of the pupils of the eyes and the horizontal direction, and obtaining the parabola of an eyelid based on the rotation angle and the position of an external contour; removing an eyelid region according to the position of the external contour and the parabola of the eyelid, carrying out the eyelash detection and highlight interference detection, so as to remove an eyelash region and a highlight point and obtain an iris image; generating a mask image according to the iris image, and obtaining an iris unfolded image and a mask unfolded image; extracting local weight histogram characteristics according to the iris unfolded image, calculating the exterior product tensor sequence features, and taking the exterior product tensor sequence features as the iris texture of the iris image; carrying out the feature matching according to the iris pattern obtained through modeling and the mask unfolded image, so as to determine a matching result. The method is suitable for the iris recognition.

The invention discloses a method for predicting an N6-methyladenosine modification site in RNA based on stacking integration, belonging to the field of systems biology. The method comprises the following steps: extracting RNA sequence features of three species, namely saccharomyces cerevisiae, homo sapiens and arabidopsis thaliana through six feature extraction methods, and conducting feature fusion to obtain an initial feature space of an original data set; performing dimensionality reduction on the initial feature space by using an elastic network, eliminating redundant and noise features, and reserving important features related to model classification so as to obtain an optimal feature set; inputting optimal feature subsets and corresponding category labels into stacking integration for model training, and evaluating the prediction performance of a model in combination with evaluation indexes to obtain a prediction model; and inputting a to-be-predicted RNA sequence in a test set into the prediction model, predicting the m6A site and outputting the m6A site. The prediction accuracy of the model on the test set reaches 92.30% and 87.06% respectively, and the model has good development potential in the aspect of cross-species prediction and is expected to become a useful tool for identifying the m6A site.