857results about How to "Avoid subjectivity" patented technology

The invention discloses a physiological information-based depressive disorder evaluation system, which comprises an information acquisition module, a signal processing module, a parameter calculating module, a characteristic selecting module, a machine learning module and a result output module. The invention further discloses a depressive disorder evaluation method based on multiple pieces of physiological information, which comprises the following steps: 1, processing one or more than one signal of an electrocardiosignal, a pulse wave signal, an electroencephalographic signal, a galvanic skin response signal, an electrogastrogram signal, an electromyographic signal, an electrooculogram signal, a polysomnogram signal and a temperature signal, and calculating signal parameters; 2, carrying out normalization on the obtained signal parameters, and carrying out characteristic selection on a parameter set consisting of the signal parameters subjected to the normalization, so as to obtain a characteristic parameter set; 3, carrying out machine learning on the characteristic parameter set, and establishing a depressive disorder evaluation mathematic model according to the relationship between the characteristic parameter set and the levels of the depressive disorder, and evaluating the levels of the depressive disorder. The physiological information-based depressive disorder evaluation system and method have the advantages that subjectivity of scale evaluation and the like can be avoided.

A visual detection device for parts mainly comprises an imaging system, an optical lens component, a lighting system, a mechanical adjusting system, a displacement detection system and a computer, wherein the imaging system consists of a CCD camera and an image acquisition card; the optical lens component consists of a lens body, an amplifying objective lens, and the like; the lighting system is respectively used for detecting surface defects and measuring topographic sizes; the mechanical adjusting system consists of displacement platforms used for adjusting the positions of the parts; the displacement detection system is used for recording the displacement distance of the displacement platforms; and the computer is used for recording and processing images. By shooting edge images of theparts through the imaging system and recording the displacement distance of the displacement platforms through the displacement detection system, the detection device calculates the topographic sizessuch as diameter, and the like, detects the surface defects of the whole spherical surface by the combined rotation of rotary tables, measures the defect sizes accurately, records the number of the defects and meets the two detection requirements of topography measurement and surface detection. By utilizing the visual microscopic detection technology, the device has high system resolution, good integrated performance and simple measurement process.

The invention discloses a depression evaluating system and method based on a heart rate variability analytical method, particularly a depression evaluating system and method in a perinatal period. The depression evaluating system comprises a data storage module used for storing electrocardio and pulse wave signals, a peak detecting module used for acquiring a peak point sequence of recording data, a data correcting module used for acquiring sinus beat NN interval sequence, a heart rate variability curve acquiring module used for acquiring a heart rate variability curve, an HRV analyzing module conducting time domain analysis, frequency domain analysis and nonlinear analysis, a feature parameter selecting module used for selecting a feature parameter from HRV parameters, a modeling module used for obtaining a perinatal period depression classification model, and a model applying module used for inputting the data of a testee into the classification module to obtain a depression degree. By means of the depression evaluating system and method based on the heart rate variability analytical method, degree quantitative evaluation of perinatal period depression is achieved, a scientific research method of the depression based on the technical field of physiological information examination is enriched, a test is simple and practicable, medical resources can be effectively saved, and good clinical practicality is achieved.

The invention discloses an image retrieval method based on color and shape features. The method comprises that: a sample image is converted and quantized on color and space, the quantized image is divided into blocks, the color complexity of each pixel point of each subblock image is calculated and vision weight is obtained, the percentage of the vision weight of different colors in the vision weight of the subblock image is calculated for each subblock image and a weighting color histogram is obtained. The color feature of each subblock is obtained according to the weighting color histogram. Contour extraction is carried out on the sample image through grey processing. Curvature scale space is adopted to describe the image shape feature of operator extraction after the contour extraction. The extracted color feature and shape feature are normalized and the normalized image feature is obtained. The normalized image feature is matched in an image feature database through an index according to a similarity measurement formula and an index result is obtained. The index method of the invention is more accurate.

The invention provides risk control system and method, wherein the system and the method are used for carrying out risk monitoring on transaction data in an electric power transaction platform. The method comprises the steps that (1) an evaluation index system is determined according to properties, characteristics and evaluation purposes of an evaluated object, wherein classification and hierarchy of the evaluation index, specific setting of the evaluation index and the determination of an evaluation standard are comprised; (2) basic data are collected, wherein relevant data of the evaluated object are collected according to the setting of the evaluation index, and the index is pretreated; (3) weight determination is carried out, and weight is given to the index; (4) and comprehensive evaluation is carried out, wherein scoring and grading are carried out on all aspects and the overall credit condition of the evaluated object, so as to reflect the credit risk condition of the evaluated object. The system and the method solve the problems of high cost and poor effect of the risk monitoring method of the existing electric power transaction platform.

The tissue imaging system, incorporates an integrated X-Ray and Ultrasound Imaging system. The X-Ray system is used for imaging and generating a 3D Volume of the extracted tissue sample (specimen). The Ultrasound imaging system is used for imaging the body area of interest of the patient from which the tissue sample is to be extracted. On the display of the system, the physician draws contours on the area of interest and the same is displayed as a 3D Volume. A 3D volume of the extracted tissue is generated using the x-ray imaging sub-system. Quantitative analysis is performed on the two 3D volumes (extracted tumor x-ray imaging and the contoured ultrasound imaging) to determine the difference between the contoured and the surgically extracted specimen to assist the physician in determining if further surgical intervention is required.

The invention provides a power utilization abnormal behavior recognition method based on a fuzzy neural network, and the method comprises the steps: extracting original data of a part of users as sample data from a power utilization database; carrying out the data preprocessing; designing a power utilization abnormal behavior evaluation index system on the basis of the analysis of power utilization abnormal behavior cases; constructing an expert sample based on the preprocessed data; taking an power utilization abnormal behavior mark as an input item, taking a power utilization abnormal behavior suspicion coefficient as an output item, and constructing a fuzzy neural network model; inputting test data into a built fuzzy neural network model, and carrying out the diagnosis of the power utilization abnormal behavior; evaluating a power utilization abnormal behavior diagnosis result, and a setting a target evaluation and optimization model. The method provided by the invention achieves the automatic recognition and diagnosis of the power utilization abnormal behavior, achieves the automatic training learning and modeling of a system, achieves the quick and precise positioning of a suspected user, and facilitates the obtaining of various illegal behaviors of power utilization.

The invention discloses an intelligent optimization method of a tumour radiotherapy plan. The method includes: 1) intelligent delineation of a biological target region, biological sub-target regions and a multi-modal(mode) image fusion target region of a tumor PET image, wherein an intelligentized image analysis processing method is adopted to delineate the biological target region, the biologicalsub-target regions and the multi-modal image fusion target region of a tumor according to multi-modal image information such as tumor PET / CT / MRI / CBCT / ultrasound; 2) optimization calculation of optimal prescription doses of biological intensity modulated tumor radiotherapy, wherein the optimal prescription dose of each biological sub-target region is determined through the intelligentized and automated optimization method according to biological characteristics of the tumor; and 3) tumor clinical-radiotherapy plan big-data analysis-based optimization calculation of the multi-objective radiotherapy plan, wherein the 1) and the 2) are combined to determine the self-adaptive biological intensity modulated clinical precise tumor radiotherapy plan of multi-objective optimization through an intelligent big-data analysis method and a deep learning, machine learning or artificial intelligence method according to clinical tumour radiotherapy plan data.

The invention discloses a mobile internet based pregnant and lying-in woman health surveillance system. The mobile internet based pregnant and lying-in woman health surveillance system comprises an intelligent terminal, a cloud server, an electrocardiogram detection device and a blood oxygen saturation degree detection device. The electrocardiogram detection device and the blood oxygen saturation degree detection device are respectively used for acquiring an electrocardiosignal and a blood oxygen saturation degree signal of a pregnant and lying-in woman and sending the electrocardiosignal and the blood oxygen saturation degree signal to the intelligent terminal. The intelligent terminal is used for uploading the electrocardiosignal, the blood oxygen saturation degree signal, a scale screening result and clinic information to the cloud server. The cloud server is used for receiving biochemical analysis information and the information uploaded by the intelligent terminal, accordingly automatically completing health state analysis on the pregnant and lying-in woman, generating corresponding nursing and health-care guidance and then feeding a health state analysis result and the nursing and health-care guidance back to the intelligent terminal. The mobile internet based pregnant and lying-in woman health surveillance system utilizes a mobile internet technology and facilitates real-time monitoring of the mental stress condition and the body health state of the pregnant and lying-in woman so as to prevent antenatal and postpartum depression.

The invention discloses an evaluating system of depressive disorder degree quantization. The system comprises an electrocardio and pulse wave integral detection device, a data transmission device and a data processing platform. The invention further discloses an evaluating method of depressive disorder degree quantization. The method comprises the following steps that firstly, human physiological information under different states is acquired through a multi-state comprehensive test platform; secondly; HRV characteristic parameters under the different states are obtained on the basis of the heart rate variability analysis principle; thirdly, the function balance states of sympathetic nerves and pneumogastric nerves in the automatic nervous system are evaluated; fourthly, a depressive disorder degree quantization evaluation model is built, and the depressive disorder degree level of a tested person is evaluated fast and objectively. The system and method belong to the technical field of computer-aided diagnosis, depressive disorder degree quantization evaluation is achieved, the blank of the technical field of depressive disorder inspection is filled up, and the system and method are easy and convenient to implement, save medical resources and have the good clinic practicality.

The invention relates to a key node recognition method for a regional power grid, and the basic flow of the method comprises the steps: inputting a power system topological structure and electrical characteristic parameters; calculating the values of evaluation indexes of a node; obtaining a node importance evaluation decision making matrix according to the values of the evaluation indexes; generating a node importance evaluation decision making matrix of a standard regional power grid; giving consideration to a Gini coefficient and a Kendall correlation coefficient, and solving the objective weights of the indexes through a CRITIC method; and solving an evaluation result of the node importance through a TOPSIS method. According to the technical scheme of the invention, the method gives full consideration to the physical features of the network topological structure and the operation state parameters of the power grid. The method can achieve the better recognition of the key node of the power grid, avoids the subjectivity of manual determination of the weights, and enables an evaluation result to be more suitable for the actual operation conditions of the power grid.

The invention discloses a coupling-bionics-based food crispness tester which comprises a chewing simulator, a cracking signal acquisition system and a data processing and analysis system, wherein the chewing simulator is used for simulating the food chewing of a human chewing system; the cracking signal acquisition system is used for acquiring a pressure signal and a sound signal, which are generated during the chewing of the chewing simulator; the data processing and analysis system is used for receiving the signals, and performing data processing and analysis to acquire a predetermined food crispiness result. The coupling-bionics-based food crispness tester is established by simulating the human chewing system and an auditory system by a bionic technology, and is used for acquiring mechanical and acoustic signals when the food is chewed, establishing a sensory crispness evaluation and mechanical and acoustic signal prediction model and comprehensively determining the crispness of the food by utilizing the model; an integrated manual technical system with the characteristic of achieving maximum environmental adaptability with low energy consumption is constructed; a test is carried out from a crispness formation mechanism, and is superior to crispness evaluation from other indexes such as cracks and simply from the mechanical signals.

The invention discloses a face image quality evaluation model construction method, a face image screening method, a face identification method, equipment and a medium. The construction method comprises the following steps that: obtaining a training set, wherein the training set comprises a plurality of face sample images and scoring true values corresponding to the face sample images one by one; according to the face sample images and the scoring true values, training a randomly initialized neural network model; carrying out an accuracy test on the trained neural network model; judging whetherthe result of the accuracy test meets an accurate condition or not; and if the result of the accuracy test meets the accurate condition, storing the trained neural network model as a face image quality evaluation model, wherein the scoring true value is obtained according to the face sample images and the standard image, and the face sample images and the standard image are related. By use of themethod, the constructed face image quality evaluation model can more favorably define face image quality, the subjectivity and the deviation of the manual definition and analysis of face image quality can be avoided, and the method is favorable for screening images in a face identification task.

The invention relates to a method and a system for evaluating the degree of application of an enterprise information technology. The evaluation method comprises the step of establishing an evaluation index system, the step of giving values to evaluation indexes, the step of determining the relative importance of each evaluation index, the step of determining the weight of each index in the system, the step of calculating an evaluation exponent of the degree of the application of the enterprise overall information technology and other steps. The invention realizes the evaluation of the degree of the application of the enterprise-oriented information technology, avoids the subjectivity caused by scoring by experts in the traditional method and ensures the relative objectivity and the accuracy of the evaluation result by the method of being combined with actual quantitative data. The invention enables enterprises to carry out reasonable and accurate positioning of the own information level and avoid the waste caused by blind investment and is simultaneously conductive to finding weak links during the own information construction for the enterprises and leading the enterprises to carry out the targeted improvements.

The invention discloses a steel coil end portion quality detection method. The detection method comprises the following steps: arranging an imaging device at each of two sides of a steel coil convey chain, respectively shooting the end portions at two sides of the pass-through steel coil through controlling the imaging devices by PLC to obtain images, and processing and comparing the images through an image processing server to realize defect determination; and querying historic image information and the defect detection result in real time through an image operation terminal. The invention also discloses a steel coil end portion quality detection system. The detection system and the detection method can mitigate the labor intensity and damages of high temperature radiation to human bodies, and effectively avoid the subjectivity and false negatives caused by visual inspection, so various defects existing in the end portions of the steel coil are effectively and objectively detected and assessed, and the qualified rate of finished steel coil is guaranteed.

A fine-grained convolutional neural network-based clothes recommendation method. First clothes pictures of various types and pictures that do not contain clothes are collected from an e-commerce website, and pictures of the two categories are labeled. Then a convolutional neural network is initialized, the labeled pictures are used to train the convolutional neural network. A recommendation process first uses a segmentation-based target area selection method to select a candidate shape area that is possible to contain clothes in a picture input by a user, and searches for the closest picture in a training picture database through a specific distance function and indexes the corresponding clothes type labeling. Depth characteristics of all clothes areas are extracted, linking and recommendation grading are performed, and finally a recommendation result is returned to the user. The extracted depth characteristics have very good invariance for rotation, translation and scale changes, and in addition, the recommendation method based on the segmentation-based target area and the adjacent picture improves recommendation accuracy to some extent.

The invention discloses a newborn pain degree assessment method and system based on facial expression recognition, and the method comprises the steps: building a newborn pain facial expression image data set which comprises a preprocessed newborn facial expression image and a corresponding expression class label; constructing a DCNN (deep convolutional neural network) for estimating the pain degree of a newborn, employing a disclosed large-size data set with a label for the pre-training of a network, obtaining an initial weight parameter value, carrying out the fine tuning of the network through the facial expression image data set, and obtaining a trained network model; inputting a to-be-tested newborn facial image into the trained network for facial expression classification and recognition, and obtaining a pain degree assessment result. The method can make the most of the features extracted through the DCNN, can obtain a better pain degree assessment result through a small-size newborn pain degree facial expression image data set, and is a new method for the development of a system for automatic assessment of the pain degree of the newborn based on the facial expression recognition.

The invention discloses a fault diagnosis method and a fault diagnosis device of a power transformer. The method comprises the following steps: establishing a state characteristic data table based on an in-oil dissolved gas sample with a definite fault type; carrying out normalized treatment on the state characteristic data table and establishing a normalized fault table; calculating based on the normalized fault table to obtain various fault type clustering centers; based on the clustering centers, establishing a state standard spectrum matrix; calculating through an improved main component analysis method to obtain a characteristic value, a characteristic vector and a main component contribution rate; setting a threshold value and correspondingly selecting a main component; and calculating an Euler distance between a sample to be detected and the main component of a state characteristic sample main component and taking a state characteristic sample corresponding to a minimum distance value as a diagnosis result. The fault diagnosis method and device of the power transformer have the following advantages that a state standard spectrum is calculated by utilizing fuzzy clustering, and subject data removal and sample quantity restriction are avoided; meanwhile, the dimension of the data can be reduced and main characteristics for representing fault types are refined; and the accuracy of latent fault diagnosis in the power transformer is effectively improved.

The disclosure discloses a spliced-video generation method and device, terminal equipment and a storage medium. The method includes: acquiring a target image set matching with a video theme of a spliced video, and respectively acquiring feature information of each target image in the target image set; screening out target pictures from the target picture set, generating a picture sequence, and determining a transition special-effect, which matches with every two adjacent target pictures, according to feature information of any two adjacent target pictures in the picture sequence; and splicingthe picture sequence to generate a spliced video according to a position order of each target picture in the picture sequence, a picture special-effect matching with each target picture in the picturesequence and the transition special-effect matching with the any two adjacent target pictures. The embodiment of the disclosure can improve generation efficiency of the spliced video, and can also improve flexibility and diversity of generating the spliced video at the same time.

The invention discloses a method for measuring the stability and imaging indexes of a nuclear magnetic resonance imaging device and is characterized in that: the images of a water phantom are obtained by adopting the nuclear magnetic resonance imaging device to scan the water phantom, the images are transmitted to a specific nuclear magnetic resonance imaging quality assurance and analysis work station by a DICOM protocol, and then the quality assurance and analysis work station is self started and enters a corresponding measurement process (referring to the figure 1 in the abstract of the process) according to DICOM control elements so as to process and calculate the images of the water phantom; the measuring content includes common nuclear magnetic resonance imaging quality assurance indexes, the shifting indexes which have important significance to Functional MRI, and other indexes; meanwhile, an automatic early warning mechanism is established for auto prompting when an abnormal condition occurs and gives the possible reason, and all measured results are sent into a database for long-term analysis of data and evaluation of running state of the device, so as to offer guidance to relevant technicians to master device state and carry out equipment testing.

The invention provides a human motion function auxiliary evaluation method, comprising the steps: obtaining a to-be-evaluated motion video which is a video obtained through the photographing of a to-be-evaluated object and the execution of a specified motion; performing human body key point detection on each frame of image in the motion video to obtain human body joint coordinates, and selecting key point coordinates corresponding to the specified action from the human body joint coordinates in each frame of image; determining motion features of each frame of image according to the key point coordinates in each frame of image in the motion video; and generating a motion curve according to the motion characteristics of each frame of image in the motion video, wherein the motion curve is used for carrying out human body motion function auxiliary evaluation processing. The invention further provides a human motion function auxiliary evaluation device and system, equipment and a storage medium.

An automatic optimization dispatching system and a method of an iron and steel energy resource on the basis of integrated data source belonging to the data source field of information resource, which comprises a PCS layer which is constituted by a sensor, a PLC, and a DCS which are mounted on site, an in-site monitoring terminal HMI, a scheduling system data bank server, an ESP data bank server, an applied system server, a client end, a fire wall, an anti-virus server, an apparatus for connecting each computer, a controller, and a computer network of a sensor. The method can analyze energy supplying and allocation of businesses according to the specific energy utilizing condition of specific iron and steel businesses. The invention has the advantages that energy effective utilization rate of the iron and steel businesses can achieve the best effect.

The invention discloses a machine vision detecting system for detecting defects of a rolling roll grinding surface, belonging to the field of grinding processing surface defect detection. The system comprises a cleaning device, an image acquisition device, an image acquisition control unit and an image processing and analysis unit, wherein the cleaning device can be used for forming a uniform laminar flow liquid layer on the surface of a rolling roll to be detected; the image acquisition device is arranged around the rolling roll to be detected; the image acquisition control unit is respectively connected with the rolling roll to be detected and the image acquisition device; the image processing and analysis unit is electrically connected with the image acquisition device and is used for realizing defect classification identification, defect evaluation and over-limit alarm as well as fault diagnosis. The machine vision detecting system can be used for carrying out qualitative and quantitative detection on the defects such as short scratches, micro-vibration, spots, spiral patterns and reticulate patterns on the rolling roll grinding surface, overcomes the defects that the existingartificial naked eye detection method is high in subjectivity and lower in accuracy, easily causes inspection missing, can not realize quantitative detection, and the like, and improves detection efficiency, accuracy and reliability.

The invention provides a collision accident loss estimating method and system based on an expert system. The method includes that the collision data obtained by an acquisition device and a collision simulation analysis module is subjected to feature extraction by a data processing system to determine the collision data, the loss estimating information and the maintenance plan of the collision accident, and the comprehensive database of the system is established; by means of the intelligent recognition and image gray-level matching technology, the loss is checked in a vehicle type damage estimating database; the feature parameters of the collision data is studied by artificial intelligence, the intelligent recognition mechanisms of the accident vehicle and the collision data are sorted, and an accident loss and maintenance expert system is established by combining an accident loss estimating expert knowledge base and a vehicle maintenance expert knowledge base; and a knowledge acquisition module performs classification and arrangement through the combination with the knowledge model established by artificial intelligence learning and the direct knowledge given by the accident loss and maintenance expert system. The subjectivity of a loss estimator in loss estimating is prevented and the probability of joint fraudulent insurance is reduced, and the disassembling cost of accident vehicles can be prevented.

The invention discloses a method for exactly characterizing and evaluating sweet aroma characteristics in cigarette aroma. Aroma substance quantitative data are acquired through an HS-GC-MS / MS detecting method, and a cigarette sweet aroma note index T is obtained by calculating and establishing a threshold value of a corresponding aroma substance. The method is easy and convenient to operate, not easily disturbed by impurities, and accurate in detecting results. The selectivity and the sensitivity of GC-MS / MS are higher than those of GC-MS; the problems that the sensitivity of the GC-MS is low, the resolution is insufficient, the selectivity is low, false positive results are easily generated, and enough qualitative / quantitative information cannot be given and the like are avoided. According to the method, the sweet aroma characteristics of a cigarette are described and expressed correspondingly from the perspective of material basis; the subjectivity of sensory evaluation is avoided; the sweet aroma style of a cigarette product is characterized and evaluated objectively and comprehensively.

The invention discloses an effective time sequence and electrode recombination based electroencephalograph signal categorizing system and a method, which can realize the identification to three types of character expression (happiness, normality and sadness) by acquiring and analyzing the electroencephalograph signals of the human brain, and mainly comprises an electroencephalograph signal acquisition process and an electroencephalograph signal analyzing process. According to the invention, the electroencephalograph signals are acquired by stimulating different expression of a testee; the feature space of the effective electroencephalograph signals is determined by the strong energy distribution of the full field of the electroencephalograph signals; then, the PCA (Principal Component Analysis) dimension reduction is carried out to the original electroencephalograph signals corresponding to the feature space, and the electroencephalograph signals with categorizing advantages are reconstructed; and finally, a linear discriminant function categorizer is selected to for categorizing. According to the invention, during expression identification, only target features are extracted from the acquired electroencephalograph signals, then, the acquired electroencephalograph signals are categorized, and the identification result can be determined; and the identification of the electroencephalograph signals can be realized based on the character expression stimulation. In the invention, the cognition of the human being is introduced, and the advantages of objectiveness and high efficiency are provided.

The invention relates to an operating state assessment method of a capacitor voltage transformer. Data collection of an operating state of the capacitor voltage transformer is performed; collected data are classified and processed and taken as bases for assessing the operating state of the capacitor voltage transformer; the operating state of the capacitor voltage transformer is assessed, and a comprehensive state quantity assessment system of the capacitor voltage transformer is built with an correlation analysis method; the operating state of the capacitor voltage transformer is graded with a single state quantity score calculation method and a capacitor voltage transformer overall score calculation method; constant weight coefficients of each single state quantity and each comprehensive state quantity are provided; and variable weight coefficients of all comprehensive state quantities are calculated; the overall score of the capacitor voltage transformer is calculated according to scores and the variable weight coefficients of all comprehensive state quantities; and then the operating state of the capacitor voltage transformer is judged according to the overall score situation of the capacitor voltage transformer.

The present invention provides an information providing method and apparatus, which have advantages of being objective, high in efficiency, wide in application scope, and good in scalability. The method comprises: extracting a sample feature vector and a corresponding sample decision from historical consultancy session data, wherein an element of the sample feature vector is an attribute value extracted from the historical consultancy session data according to a preset attribute, and the sample decision is a user questioning sentence; performing training by using a plurality of sample feature vectors and corresponding sample decisions to obtain a probabilistic classification model; extracting a to-be-tested feature vector from a current client consultancy request, wherein the to-be-tested feature vector is the same as the sample feature vector in format; inputting the to-be-tested feature vector into the probabilistic classification model, then receiving one or more candidate decisions and corresponding probabilities output from the probabilistic classification model, wherein the candidate decisions are candidate user questioning sentences; choosing K candidate decisions of greatest probabilities as forecasted decisions, then providing a client with a standard reply corresponding to the forecasted decisions.

The invention discloses an intrusion intension recognition system and method based on hidden markov and probability inference, and belongs to the technical field of computer security. The method comprises the steps that firstly, due to the causal relationship existing among attack events, subgoals and intentions, a hierarchical bayesian network (HBN) is built for representing the causal relationship between the layers; secondly, the subgoals have the obvious time-order characteristics, and the hidden relationship for a target layer and an intention layer is learned based on a hidden markov model (HMM); finally, a multi-output hidden markov model (MO_HMM), a loop believable propagation algorithm (LBP) is introduced for optimizing the problem of single-probability output of a forward algorithm and a backward algorithm in HMM, and a plurality of possible intentions and probability can be reported.

The invention discloses a method and device for extracting highlights of a video, and the method comprises the steps: obtaining the number of bullet screens in a target video in each preset time period, adaptively determining a bullet screen number threshold value according to the number of bullet screens in the target video in each preset time period, and avoiding the subjectivity of threshold value selection; determining the preset time periods when the number of bullet screens in the target video is greater than the bullet screen number threshold value as highlight time periods; carrying out the clustering of all highlight time periods, and obtaining a plurality of clusters and the start and end moments of each cluster; carrying out the extraction of the target video according to the start and end moments of each cluster, and obtaining a target highlight segment. Through the clustering of the highlight time periods, the method achieves the discretization and fragmentization of highlight time points, guarantees that the target highlight segment has a certain time duration, and guarantees the wonderfulness of the target highlight segment.