1362results about How to "Enhance expressive ability" patented technology

The invention provides a group of relevant enzymes for preparing mannitol by performing anabolism on Chinese caterpillar fungus serving as a multifunctional production fungus and hirsutella sinensis based on glucose, a gene for encoding these enzymes and application thereof. The relevant enzymes include (1) hexokinase: manA1-A6 proteins of which the sequences are SEQ ID No.1-6, (2) phosphoglucoisomerase: manB1-B3 proteins of which the sequences are SEQ ID No.7-9, and (3) mannitol-1-P dehydrogenase: manC protein of which the sequence is SEQ ID No.10. In the invention, detailed researches are performed on the metabolic pathway of mannitol synthesized by using Chinese caterpillar fungus serving as a multifunctional production fungus, hirsutella sinensis and glucose on the aspect of principle, cloned DNA (Deoxyribose Nucleic Acid) comprising a nucleotide sequence provided by the invention can be transferred into engineering bacteria with transduction, conversion and conjugal transfer methods, and host mannitol is endowed with high expression by regulating the expression of a biosynthetic gene of the mannitol.

The invention relates to a front small-region object recognition and vehicle early warning method, and belongs to technical field of early warning for collision in front of vehicles. The front small-region object recognition and vehicle early warning method specifically comprises the steps that an image of a target to be recognized in front of a vehicle is acquired, and a region of interest (ROI)image is obtained; image preprocessing is performed on the obtained region of interest image, the preprocessed region of interest image is inputted into an improved YOLO convolutional network structure model to perform target recognition, and a target regression frame and a target category are outputted; the recognition result is inputted into a vehicle early warning system, if the target regression frame is located beyond the range of a vehicle driving route or the target is recognized to be of a non-dangerous category, the target is marked as a non-dangerous target, and the vehicle does notperform early warning; and if the target regression frame is located within the vehicle driving route and the recognized object belongs to a dangerous category or the object cannot be recognized but has certain speed features, the target is marked as a dangerous target, the vehicle performs early warning to remind a driver to pay attention so as to avoid occurrence of an accident.

A stretched cellulose acylate laminate film having a skin layer containing a cellulose acylate with a total degree of acyl substitution of more than 2.7 and a core layer containing a cellulose acylate with a total degree of acyl substitution of 2.0-2.7 wherein the core layer is thicker than the skin B layer and at least one of these layers contains a retardation-controlling agent, is excellent in high expressibility, little optical unevenness and good releasability from a support.

The invention provides a method for fusing a full-color image and a multispectral image based on a deep neural network. Specific steps are as follows: step 1. constructing a training set of high resolution and low resolution image block pairs; step 2. using an improved sparse denoising self-encoder to learn to train initialization parameters of a first layer in a neural network model; step 3. using the improved sparse denoising self-encoder to perform pretraining of the neural network layer by layer; step 4. performing fine adjustment of parameters of the pretrained deep neural network; and step 5. using the deep neural network to reconstruct a multispectral image of high resolution according to a known multispectral image of low spatial resolution. The method provided by the invention adopts a method of deep learning, and can make full use of a nonlinear neural network to depict complex structural information of a multispectral image, thereby enabling the fused multispectral image to have high spatial resolution, and well keeping spectral information of the multispectral image.

The invention discloses a method and a device for acquiring knowledge graph vectoring expression. The method comprises the following steps of labeling an entity, existed in and belonging to a knowledge graph, in a given auxiliary text corpus by utilization of an entity labeling tool according to a to-be-processed knowledge graph so as to obtain an entity-labeled text corpus; constructing a co-occurrence network comprising words and entities on the basis of the text corpus so as to relate text information of the auxiliary text corpus to entity information of the knowledge graph, and then learning to obtain a text context embedded expression; respectively modeling the embedded expression of the entity and relation in the knowledge graph according to the text context embedded expression so as to obtain an embedded expression model of the knowledge graph; training the embedded expression model by utilization of a stochastic gradient descent algorithm so as to obtain the embedded expression of the entity and relation in the knowledge graph. The method and the device disclosed by the invention have the advantages that not only can the expression capability of the relation be improved, but also the problem of insufficient expression effect caused by sparseness of the knowledge graph can be effectively solved.

A phishing webpage detection method based on Hungary matching algorithm is characterized by firstly extracting the text feature signatures, image feature signatures and general webpage feature signatures of the rendered webpages and more comprehensively depicting the features after access to webpages; and then computing the optimal matching of bipartite graphs by Hungary algorithm to search for the matched feature pairs among different webpage signatures and more objectively measuring the similarity among the webpages on the basis, thereby improving the phishing webpage detection efficiency. The method is also characterized by determining the inside weights of the text features, image features and global image features by utilizing the area under curve and determining the relative weightsamong the text similarity, image similarity and global image similarity during webpage similarity computation by utilizing logarithmic regression analysis. The precision and the recall rate are greatly improved in the method provided by the invention.

The invention discloses a deep learning-based human body motion recognition method of multi-channel image feature fusion. The method comprises: (1) extracting original RGB pictures from videos, and calculating dynamic graphs and optical flow graphs of the segmented videos through the RGB pictures; (2) carrying out cropping operations on the input pictures to expand a training data set; (3) constructing a three-channel convolutional-neural-network, and respectively inputting lastly obtained video segments into the three-channel convolutional-neural-network to carry out training to obtain a corresponding network model; and (4) for a to-be-recognized video segment, extracting original RGB pictures, calculating dynamic graphs and optical flow graphs corresponding thereto, and obtaining a recognition result of a final motion category. According to the method, the three-channel convolutional-neural-network is utilized for learning essential features of data for original input of different morphologies, multi-channel dense fusion operations are carried out on the data of the three morphologies in the middle of the network, expression ability of the features is improved, and purposes of multi-channel information sharing and a high accuracy degree are achieved.

The invention discloses a brain tumor segmentation network and segmentation method based on a U-Net network. The tail of a contraction path of the segmentation network is connected with a spatial pyramid pooling structure; hole convolution of different scales is introduced into a network jump connection part of the segmentation network; an Add operation and original input are adopted to form a residual block with hole convolution; a receptive field of shallow feature information in the contraction path is expanded; fusing with an expansion path of a corresponding stage is carried out. The segmentation method comprises the following steps: cutting and preprocessing a training data set, then constructing a brain tumor segmentation network DCU-Net based on a U-Net network, then inputting a preprocessed two-dimensional image into a segmentation model for feature learning and optimization, obtaining an optimal parameter model of the segmentation model, and finally inputting a to-be-segmented test data set image into the segmentation model for tumor region segmentation. According to the method, the problems of over-segmentation and under-segmentation in brain tumor segmentation can be effectively solved, and the brain tumor segmentation precision is improved.

The invention provides a news recommendation system and method based on an FOLFM model. Based on a content recommendation method, a news content model is expressed abstractly through a latent class model and content characteristics, and an individual latent class preference model is built for each user. Real-time training is carried out on a real-time behavior record of a user to obtain preference, on certain latent class news, of the user, whether the news is recommended to the user is determined through calculation, and a final news recommendation list is obtained after a series of processing processes. The news recommendation system and method based on the FOLFM model deeply excavate the interest of the user, improve recommendation accuracy and satisfaction of the user, avoid a cold starting problem of the news, and guarantee performance under the condition that the recommendation effect is improved as much as possible. The experiment shows that the news recommendation system and method based on the FOLFM model not only guarantee the requirements for high accuracy and high speed, but also realize visual real-time recommendation for the user.

The invention relates to an image description generation method based on a depth LSTM network, comprising the following steps: (1) extracting the CNN characteristics of an image in an image description dataset, and acquiring an embedded vector corresponding to the image and describing the words in a reference sentence; (2) building a double-layer LSTM network, and carrying out series modeling based on the double-layer LSTM network and a CNN network to generate a multimodal LSTM model; (3) training the multimodal LSTM model by means of joint training; (4) gradually increasing the number of layers of the LSTM network in the multimodal LSTM model, carrying out training each time one layer is added to the LSTM network, and finally, getting a gradual multi-objective optimization and multilayer probability fused image description model; and (5) fusing the probability scores output by the branches of the multilayer LSTM network in the gradual multi-objective optimization and multilayer probability fused image description model, and outputting the word corresponding to the maximum probability through common decision. Compared with the prior art, the method has such advantages as multiple layers, improved expression ability, effective updating, and high accuracy.

The invention provides an improved CNN-based facial expression recognition method, and relates to the field of image classification and identification. The improved CNN-based facial expression recognition method comprises the following steps: s1, acquiring a facial expression image from a video stream by using a face detection alignment algorithm JDA algorithm integrating the face detection and alignment functions; s2, correcting the human face posture in a real environment by using the face according to the facial expression image obtained in the step s1, removing the background information irrelevant to the expression information and adopting the scale normalization; s3, training the convolutional neural network model to obtain and store an optimal network parameter before extracting feature of the normalized facial expression image obtained in the step s2; s4 loading a CNN model and the optimal network parameters obtained by s3 for the optimal network parameters obtained in the steps3, and performing feature extraction on the normalized facial expression images obtained in the step s2; s5, classifying and recognizing the facial expression features obtained in the step s4 by using an SVM classifier. The method has high robustness and good generalization performance.

The invention discloses a face multi-area fusion expression recognition method based on depth learning, which comprises the following steps of detecting a face position with a detection model; obtaining the coordinates of the key points by using the key point model; aligning the eyes according to the key points of the eyes, then aligning the face according to the coordinates of the key points of the whole face, and clipping the face region by affine transformation; cutting the eye and mouth areas of the image to a certain proportion; dividing the convolution neural network into one backbone network and two branch networks; carrying out the feature fusion in the last convolution layer, and finally obtaining the expression classification results by the classifier. The method of the inventionutilizes the priori information, besides the whole face, the eyes and mouth regions are also used as the input of the network, and the network can learn the whole semantic features of facial expressions and the local features of facial expressions through model fusion, so that the method simplifies the difficulty of facial expression recognition, reduces the external noise, and has strong robustness, high accuracy, low complexity of the algorithm and so on.

The present invention claims a handwriting numeral recognition method based on a convolutional neural network and a support vector machine. The method organically combines a convolutional neural network model with a support vector machine model. A handwriting numeral recognition model combining the convolutional neural network with the support vector machine can deeply describe correlation between sample data and expected data and automatically learn an image feature from original data, has very good decision plane, and is very strong in discrimination capability of digital pattern classification. The handwriting numeral recognition method based on the convolutional neural network and the support vector machine, provided by the present invention, is simple, easy to implement, and very good in handwriting numeral recognition effect.

The invention discloses a senile dementia monitoring system based on a healthy service robot. The senile dementia monitoring system based on the healthy service robot comprises the healthy service robot, an intelligent terminal and a cloud server. The healthy service robot comprises a robot body, a main control unit, a human-computer interaction unit and a medical detection unit; the human-computer interaction unit is connected with the main control unit and comprises a tablet computer, and the tablet computer is arranged in front of the chest of the robot body; the medical detection unit is connected with the main control unit and comprises a brain electrical detection device which is independent of the robot body, and the brain electrical detection device is connected with the intelligent terminal and the tablet computer through Bluetooth signals; the intelligent terminal and the tablet computer are connected with the cloud server through the mobile Internet, and data interaction between the intelligent terminal and the tablet computer is achieved through wireless signals. By means of the senile dementia monitoring system based on the healthy service robot, automatic auxiliary diagnosis and treatment of senile dementia can be achieved, the diagnostic accuracy is improved, prevention and early detection of the senile dementia are facilitated, the pathogenetic condition is mitigated from exacerbating, and the purpose of healing is achieved.

The invention discloses a pre-fetching-based phishing web page detection method, and relates to the acquisition of website information and the extraction and classification of topological characteristics and mainly aims to solve problems on phishing web page detection capacity. In the method, a user interface module 1 serves as an interface, a master control module 2 serves as a center, and a classifier module 3, a characteristic extraction module 4 and a web page extraction module 5 are scheduled, wherein the classifier module needs training in a training set and adopts an incremental updating mode to ensure that a classifier keeps capacity in the detection of new phishing web pages; the characteristic extraction module mainly extracts the pre-fetched characteristics of topological website structures, saves the characteristics into a training set database and simultaneously transmits the characteristics to the classifier module; and the web page extraction module captures a certain number of web pages of a given website according to an instruction of the master control module and saves the web pages into a web page database. Through the pre-fetching-based phishing web page detection method provided by the invention, both accuracy and recall rate are greatly improved.

The invention discloses an abnormal behavior detection method, which comprises the following steps of: extracting dynamic human skeleton joints in a video by using a neural network human skeleton extraction model to form a skeleton data set; obtaining a higher-level behavior feature map corresponding to the bone, namely surface behavior features, by an ST-GCN network; and inputting the behavior feature map into the abnormal behavior classifier model, and performing matching to identify the behavior type. The invention also discloses an abnormal behavior detection system which comprises a videomonitoring module and a network model integration module. According to the method, various human body behaviors and a large amount of human body skeleton data can be accurately and efficiently processed, and abnormal behaviors appearing in video monitoring can be automatically identified.

The present invention relates to a commodity property characteristic word clustering method. The method comprises the following steps: A1: obtaining comment texts of a target commodity from related e-commerce websites, and performing data preprocessing; A2: selecting a comment text containing commodity property characteristic words, performing manual annotation on the commodity property characteristic words, and using the manually annotated commodity property characteristic words as a training sample of an obtained part-of-speech template; A3: training the part-of-speech template according to the manually annotated data in the A2; A4: using data obtained in the A1 to train a language model, thereby obtaining a word vector representation; and A5: using a word vector obtained in the A4 to perform clustering on the commodity property characteristic words obtained in the A3, thereby obtaining a final property characteristic word set of the target commodity. The commodity property characteristic word clustering method provided by the present invention can be applied to a commodity recommendation system based on a commodity comment text. The number of commodity property characteristic words can be reduced by clustering, so that characteristic dimensions and characteristic sparsity are reduced, and the designed recommendation system is faster and more accurate.

The invention discloses a natural scene text detection method based on full convolution neural network, This method uses CNN network to extract the feature representation of text, and adjusts the feature representation by feature fusion module, at the same time fuses the semantic feature of high-level feature map and the position information of low-level feature map, so that the extracted featurehas stronger representation ability, and combines with text prediction module to directly predict the candidate text object. This method adopts end-to-end training and prediction process, The processing flow is simple, and no multi-step hierarchical processing is needed. Finally, the final detection result is obtained by simple NMS operation, which has high accuracy and strong robustness. The multi-directional and multi-size text object in the natural scene image with complex background can be detected well, and the detection performance of the natural scene text is excellent.

The invention discloses a method for automatically identifying a breast tumor area based on an ultrasound image. The method comprises the following steps of acquiring the ultrasound image of the breast, and preprocessing the ultrasound image; segmenting the ultrasound image subjected to preprocessing through an image segmentation method to obtain a plurality of segmented subareas; extracting a grey level histogram, texture features, gradient features and morphological features of the ultrasound image, and combining the grey level histogram, the texture features, the gradient features and the morphological features of the ultrasound image with two-dimensional position information to obtain high-dimensionality feature vectors; selecting the most effective feature subset of the high-dimensionality feature vectors through feature ordering based on biclustering and a selection method; performing learning classification on the selected most effective feature subset through a classifier, and then automatically identifying the breast tumor area. By means of the method, the breast tumor area can be identified automatically from segment results of the breast tumor ultrasound image, therefore, automation performance of computer-aided diagnosis is improved, manual operation of clinical doctors is reduced, and subjective influence of clinical doctors is reduced.

The invention discloses a facial feature recognition method and system based on the multi-region characteristic and metric learning. The method comprises the steps that convolution neural network parameters of the corresponding location and scale are obtained through the multi-scale facial area training, and corresponding facial area features are extracted according to the neural network parameters; the above features are filtered to obtain the high dimensional facial features; metric learning is conducted according to the high dimensional facial features, the after defined loss function of the feature expression is obtained through the dimension reduction processing of the features, a network model of the metric learning is obtained through the training of the loss function; the images to be recognized are inputted into the network model, the facial features are dimension reduced using the Euclidean distance in order to be recognized. In the method, multiscale is used to select multiple areas, the convolutional neural networks are trained, and the expression skills of the characteristics are improved. Meanwhile, through the selection of the acquired multi-scale features, the efficiency of expression of characteristics is improved, and the accuracy of face recognition is effectively improved.

The invention discloses an optimized face recognition method and a device thereof, which improve the rate of facial image recognition. The technical proposal is that: the method and the device combine principal component analysis and linear discriminant analysis to solve the recognition problem; namely, the principal component analysis is carried out first before the conduction of the linear discriminant analysis so as to obtain relatively low dimensional space, then the space is utilized to carry out the linear discriminant analysis, and therefore, an intra-class dispersion matrix can not be caused and the process of the linear discrimination is effective. The invention firstly adopts the principal component analysis to obtain the best feature description, based on which, optimum identifying features are obtained by adopting the linear discriminant analysis, thereby greatly reducing the space dimension of facial features; finally, a minimum distance method is adopted to carry out classification and identification and therefore the rate of recognizing human faces is evidently increased. The method and the device of the invention are applied to face recognition.

The invention discloses a visual target tracking method based on self-adaptive subject sensitivity, and belongs to the technical field of computer vision. The visual target tracking method comprises an overall process, an offline part and an online part. The whole process includes: designing a target tracking process, and designing a network structure; adjusting the feature map of each stage of the network into an adaptive size to complete the end-to-end tracking process of the twin network; the offline part comprises six steps: generating a training sample library; carrying out forward tracking training; calculating a back propagation gradient; calculating a gradient loss item; generating a target template image mask; and training a network model and obtaining the model. The online part comprises three steps: carrying out model updating; carrying out online tracking; and positioning a target area. The model updating comprises forward tracking, back propagation gradient calculation, gradient loss item calculation and target template image mask generation; the online tracking comprises the steps of performing forward tracking to obtain a similarity matrix, calculating the confidencecoefficient of a current tracking result and returning to a target area. The method can better adapt to target robust tracking of appearance changes.

The application discloses a human face living body detection method and apparatus, and an electronic device. The method includes the steps of training a first deep learning model according to a human face acquisition global image; training a second deep learning model according to a human face region partial image clipped from the human face acquisition global image; and performing human face living body detection according to the trained first deep learning model and the trained second deep learning model. With the embodiments of the application, the extraction of more human face living body image features is facilitated, for the combination of the trained first deep learning model and the trained second deep learning model has a better feature expression capability of the human face living body as compared with a model in the prior art, and thus the improvement in the accuracy of human face detection results.

The invention relates to an RGB-D image classification method and system. The method comprises the steps of: S1, utilizing a convolution neural network (CNN) to process a source RGB image and a Depth image respectively, and extracting low level characteristics; S2, utilizing a recursion neural network (RNN) to carry out feedback learning on the image low level characteristics, and extracting image middle level characteristics; S3, adopting a block interior constraint dictionary learning method, carrying out characteristic set sparse expression on the image middle level characteristics, and obtaining high level characteristics of the RGB-D images; and S4, inputting the high level characteristics of the RGB-D images into a linear SVM to complete the classified identification of the RGB-D images. According to the invention, automatic characteristic extraction of the images is realized, learning RGB-D image characteristic expressions can effectively distinguish classification of noise data from high similarity images, and the classification precision of the RGB-D images is improved; in addition, the linear SVM is utilized, and the image classification speed is improved.

The invention discloses a recurrent neural network (RNN)-based social network message bust detection method and system, and relates to the technical field of popularity prediction of contents in social networks. The method comprises the following steps: acquiring history messages published and forwarded by a user in a social network, and preprocessing the history messages to obtain a history forwarding time sequence; carrying recurrent neural network training on the history messages and the history forwarding time sequence, and generating a prediction model; and acquiring messages published and forwarded by the user in real time, generating a forwarding time sequence according to the messages, inputting the forwarding time sequence into the prediction model to generate feature expressions, inputting the feature expressions into a fully-connected neural network to carry out classification, and outputting the classification result in a softmax manner so as to complete the social network message burst detection.

The invention discloses a multi-scale fast human face detection method for convolutional neural network feature fusion. The method comprises the following steps: step 1, based on a model structure ofan SSD fast target detection method, a feature extraction method in the SSD is transformed, a feature fusion method is added, and a modified detection model is obtained; step 2, human face detection training is carried out on the detection model transformed in the step 1, and a trained deep neural network model is obtained; step 3, the to-be-detected picture is calculated by using a deep neural network model trained in the step 2, to obtain a model output result. According to the invention, the human face in the image can be quickly recognized and the human face can be accurately positioned, so that the human face is separated from the complex background, and a foundation is provided for identity verification and tracking of people in an image.

The invention relates to a multilanguage question and answer interface fast constituting method based on domain ontology and template logics. The method comprises the following steps of (1) multilanguage domain ontology structure building; (2) first-order template logic system building based on the domain ontology; (3) multilanguage question template structure design based on domain ontology and template logics; (4) domain-oriented question template base building; (5) user question preprocessing and question template matching; (6) user question semanteme obtaining and multilanguage intertranslation method. The multilanguage question and answer interface fast constituting method has the advantages that the question template semanteme can be more precisely expressed through the domain ontology and the template logics, the question template expression ability is improved by combining template operation characters, higher representativeness is realized, the template base scale can be reduced, in addition, a multilanguage body is similar to a multilanguage semantic dictionary, and the cross-language information inquiry can be easily realized. According to the method, various multilanguage man-machine interactive interfaces of field-oriented intelligent information retrieval and automatic question and answer systems can be fast constituted.

An embodiment of the invention relates to a deep learning-based road traffic sign detection and identification method. The method comprises the steps of calculating convolutional features of a plurality of layers for training set data by utilizing a convolutional neural network; training a region-of-interest suggestion network by utilizing the convolutional features; and training a classification layer and a regression layer of an output layer by utilizing a random gradient descent method based on the convolutional features and candidate boxes obtained by the trained region-of-interest suggestion network.

The invention discloses a face attribute recognition method and device. A depth residual network is used to train the attribute of a collected face image, and a multi-attribute network training model with multiple face attributes is obtained. The obtained multi-attribute network training model is used to carry out multi-attribute detection on the face attribute of an image to be identified so as to identify the multiple face attributes in the image to be identified. The invention provides the face attribute recognition method and device, the depth residual network is used to train the attribute of the collected face image, the multi-attribute detection of the face attribute of the image to be identified is carried out according to the multi-attribute network training model obtained through specific task training, since a multi-attribute network model can extract a face characteristic with strong expressive ability, the multiple face attributes in the face image can be rapidly and accurately identified.

The invention provides an underwater autonomous robot fixed depth control method based on reinforcement learning and belongs to the underwater robot control field. The method comprises steps that firstly, a Markoff decision process model for underwater autonomous robot fixed depth control is constructed, and expressions of a state variable, a control variable, a transfer model and a one-step lossfunction for underwater autonomous robot fixed depth control are respectively acquired; the decision network and the evaluation network are respectively established; through reinforcement learning, the decision network and the evaluation network can be continuously updated whenever an underwater autonomous robot moves forwards for each step during fixed depth control training till convergence; thefinal decision network for fixed depth control is acquired. The method is advantaged in that fixed depth control on the underwater autonomous robot under the condition of a completely-unknown underwater autonomous robot dynamic model is realized, and the practical value is high.