571 results about "Discriminative model" patented technology

In language evaluation systems, user expressions are often evaluated by speech recognizers and language parsers, and among several possible translations, a highest-probability translation is selected and added to a dialogue sequence. However, such systems may exhibit inadequacies by discarding alternative translations that may initially exhibit a lower probability, but that may have a higher probability when evaluated in the full context of the dialogue, including subsequent expressions. Presented herein are techniques for communicating with a user by formulating a dialogue hypothesis set identifying hypothesis probabilities for a set of dialogue hypotheses, using generative and/or discriminative models, and repeatedly re-ranks the dialogue hypotheses based on subsequent expressions. Additionally, knowledge sources may inform a model-based with a pre-knowledge fetch that facilitates pruning of the hypothesis search space at an early stage, thereby enhancing the accuracy of language parsing while also reducing the latency of the expression evaluation and economizing computing resources.

A dialog state tracking system. One aspect of the system is the use of multiple utterance decoders and/or multiple spoken language understanding (SLU) engines generating competing results that improve the likelihood that the correct dialog state is available to the system and provide additional features for scoring dialog state hypotheses. An additional aspect is training a SLU engine and a dialog state scorer/ranker DSR engine using different subsets from a single annotated training data set. A further aspect is training multiple SLU/DSR engine pairs from inverted subsets of the annotated training data set. Another aspect is web-style dialog state ranking based on dialog state features using discriminative models with automatically generated feature conjunctions. Yet another aspect is using multiple parameter sets with each ranking engine and averaging the rankings. Each aspect independently improves dialog state tracking accuracy and may be combined in various combinations for greater improvement.

Iconic images for a given object or object category may be identified in a set of candidate images by using a learned probabilistic composition model to divide each candidate image into a most probable rectangular object region and a background region, ranking the candidate images according to the maximal composition score of each image, removing non-discriminative images from the candidate images, clustering highest-ranked candidate images to form clusters, wherein each cluster includes images having similar object regions according to a feature match score, selecting a representative image from each cluster as an iconic image of the object category, and causing display of the iconic image. The composition model may be a Naïve Bayes model that computes composition scores based on appearance cues such as hue, saturation, focus, and texture. Iconic images depict an object or category as a relatively large object centered on a clean or uncluttered contrasting background.

The present invention requires to protect a conditional generative adversarial network-based online handwriting identification method. The method comprises the steps of 101 using a user registration module to register the basic information of a user; 102 using a reception module to receive a section of character information inputted by the user, wherein the information comprises the character writing style, the character writing strength and the character writing spacing; 103 training a conditional generative adversarial network on a handwriting signature data set by taking the category labels as the conditions, and being able to generate the corresponding directional digital features according to the information of the category labels; 104 using a handwriting identification module, using the conditional generative adversarial network to mine the personalized handwriting of the user and using an adversarial network signature discrimination model D which is a dichotomy device to discriminate whether the inputted data is the real handwriting data or a generated sample; 105 using an application module to apply the handwriting identification to an access control system and a plurality of user document signing scenes. The conditional generative adversarial network-based online handwriting identification method of the present invention has higher stability, safety and convenience, at the same time, can identify the handwriting style, strength and spacing information of the users by combining a conditional generative adversarial network method, and avoids the problem that the character features are not extracted completely.

Embodiments of a dialog system employing a discriminative action selection solution based on a trainable machine action model. The discriminative machine action selection solution includes a training stage that builds the discriminative model-based policy and a decoding stage that uses the discriminative model-based policy to predict the machine action that best matches the dialog state. Data from an existing dialog session is annotated with a dialog state and an action assigned to the dialog state. The labeled data is used to train the discriminative model-based policy. The discriminative model-based policy becomes the policy for the dialog system used to select the machine action for a given dialog state.

A system and method for building a language model for a translation system are provided. The method includes providing a first relative ranking of first and second translations in a target language of a same source string in a source language, determining a second relative ranking of the first and second translations using weights of a language model, the language model including a weight for each of a set of n-gram features, and comparing the first and second relative rankings to determine whether they are in agreement. The method further includes, when the rankings are not in agreement, updating one or more of the weights in the language model as a function of a measure of confidence in the weight, the confidence being a function of previous observations of the n-gram feature in the method.

The invention relates to a discriminant analysis-based high road real-time traffic accident risk forecasting method. The method comprises the following steps of: building a high road accident risk discrimination model for a detection area; substituting real-time traffic flow characteristic parameters into the high road accident risk discrimination model; and judging whether the risk of traffic accident exists or not. According to the method, traffic accidents can be forecasted in real time by using the real-time traffic flow characteristic parameters acquired by high road traffic detection equipment, the method has relatively high forecasting precision, and technical defects and shortages in the prior art for analyzing traffic safety by using aggregated statistics are overcome. The methodhas practical engineering application value in the aspects of discrimination of the risk of the high road traffic accidents and forecast of the traffic accidents.

The invention relates to a device and a method for detecting hatching egg incubation quality based on computer vision, belonging to the technical field of agricultural product detection. The device comprises a CCD camera, an optical chamber, an object stage, a light source, an image acquisition card and a computer. The method comprises the following steps: adopting the CCD camera to acquire a perspective image of an egg, and then transmitting the image into the computer through the image acquisition card; processing the image to extract a color feature parameter; and discriminating whether theegg is fertilized or not according to a Bayes discrimination model for egg incubation quality. The invention detects the hatching egg incubation quality by applying a computer vision method and can reduce the labor intensity and the interferences of artificial subjective factors and obtain a more objective and accurate result. Proved by experiments, when the method is used for detecting the hatching egg incubation quality, the detection accuracy on the sixth day of a white-shell hatching egg reaches 100 percent, and the accuracy for verifying the model reaches 100 percent; the detection accuracy on the sixth day of a brown-shell hatching egg reaches 97.1 percent, and the accuracy for verifying the model reaches 100 percent.

The invention discloses a traffic jam judging method based on a video detection technology. By adopting a digital image processing technology, the background model of a traffic video image is established, foreground extraction and foreground de-noising are carried out on the background model, road occupancy is calculated, and a traffic jam judging model is established, thus finishing the judgment of the traffic jam state by the four steps. The traffic jam judging model comprises a jam fuzzy clustering judger and an auxiliary judger, the video image processing technology is utilized to obtain one parameter, i.e. the road occupancy, thus calculating the occupancy variance and the absolute value of occupancy variation, and being capable of finishing the judgment of the traffic jam state by using the three finite parameters.

A probabilistic boosting tree framework for computing two-class and multi-class discriminative models is disclosed. In the learning stage, the probabilistic boosting tree (PBT) automatically constructs a tree in which each node combines a number of weak classifiers (e.g., evidence, knowledge) into a strong classifier or conditional posterior probability. The PBT approaches the target posterior distribution by data augmentation (e.g., tree expansion) through a divide-and-conquer strategy. In the testing stage, the conditional probability is computed at each tree node based on the learned classifier which guides the probability propagation in its sub-trees. The top node of the tree therefore outputs the overall posterior probability by integrating the probabilities gathered from its sub-trees. In the training stage, a tree is recursively constructed in which each tree node is a strong classifier. The input training set is divided into two new sets, left and right ones, according to the learned classifier. Each set is then used to train the left and right sub-trees recursively.

The invention discloses an orchard pest and disease damage general investigation system and method based on UAV (unmanned aerial vehicle) remote sensing. The method comprises steps as follows: hyperspectral image data and RGB images of fruit tree canopies are collected by a UAV, and the collected hyperspectral image data is corrected; spectral reflectance, crown breadth, blade density and textural features of corresponding band images of the fruit tree canopies are calculated according to the corrected hyperspectral image data; size, shape and color features of fruits on fruit trees, color distribution law of the fruit trees and a three-dimensional view of orchard earth surface are calculated according to the RGB images; the spectral reflectance, the blade density and the textural features of the spectral images of the fruit tree canopies as well as the size and the color features of the fruits are input into a discrimination model, and identification of pest and disease damage of the fruit trees in the orchard is realized. Based on the hyperspectral images and the RGB images with more abundant information, the demand of high-frequency regular monitoring for insect situation information of the orchard can be met, and the general investigation efficiency of the pest and disease damage of the orchard is improved.

The invention relates to the field of computer vision and discloses a computer vision-based vehicle damage distinguishing method so as to solve a problem that a manual distinguishing approach adopted in conventional technologies is incomplete, inaccurate and inefficient in distinguishing operation. The computer vision-based vehicle damage distinguishing method comprises the following steps: in step a, a binocular image collection system is calibrated; in step b, the binocular image collection system is used for collecting images in a monitored area, and a depth map for the collected images is obtained; in step c, a convolutional neutral network is used for subjecting the depth map to characteristic extraction training operation, and a vehicle damage condition discrimination model is trained; in step d, the vehicle damage condition discrimination model is used for determining a damage degree based on collected vehicle images. The computer vision-based vehicle damage distinguishing method is suitable for distinguishing vehicle damage.

The invention discloses a method for identifying producing area of tea by using element content in the tea. Spectroscopy analysis technique is used for determination and establishment of a tea element content database; based on the partial least squares, VBA code of EXCEL is used to establish a multielement content tea producing area discrimination model; through the determination of the contents of 21 elements in the tea, data is input into a prediction module of a piece of software for calculation; if a judgment value of a certain tea is greater than 0.8, the tea can be considered as belonging to the tea producing area, and otherwise the tea doesn't belong to the tea producing area. The analysis results of the invention are only decided by software prediction value, so as to rule out subjectivity and uncertainty of human judgment; and the method with high determination accuracy can be used to identify different teas from different producing areas and has a broad application prospect.

This invention relates to a method for identifying near infrared spectrum of rosewood including the following steps: applying rosewood and non-rosewood samples to collect several times of near infrared spectrums at different positions on the surface of the sample by a near infrared spectrum device and collect spectrums of 3-19 positions at a same sample to set up distinguishing models of true and false rosewood and its kind through spectrum pre-process such as smoothing, base line correction, a first stage derivative, a second stage derivative, multiple dispersion correction or pre-process of dimensionality reduction of data by a multivariable data analysis method of soft independent modeling sorting or deflection of the least square differential analysis so as to realize quick and harmless identification to true and false of rosewood and their kinds in several minutes.

The invention provides a diabetic retinopathy fundus photography standard image generation method, which comprises the following steps: 1) enabling a collected non-standard fundus image to be generated into a new sample image through a generation model; 2) carrying out local feature extraction on the new sample image; and 3) comparing local features of the non-standard image with local features of a standard image in a discrimination model, if the local features are consistent, outputting the new sample image, that is, the generated standard image, and if the local features are not consistent, adjusting the new sample image. The provided method is simple and effective; the definition of the generated standard image reaches requirement of an intelligent aided diagnosis system; and accuracy of diagnosis is improved.

The application provides a training method and device of a voice processing model, a voice recognition method, system and device, and relates to the technical field of information related to artificial intelligence and machine learning. The training method of the voice processing model comprises the following steps of performing iterative joint training on a voice enhancement model, a voice recognition model and a voice discrimination model; for each training, obtaining a joint loss function of the voice enhancement model, the voice recognition model and the voice discrimination model, and a voice discrimination loss function of the voice discrimination model; and adjusting model parameters of the voice enhancement model and/or the voice recognition model according to the joint function after each training, and adjusting the model parameter of the voice discrimination model according to the voice discrimination loss function, and the trained voice processing model is acquired until thejoint loss function and the voice discrimination loss function simultaneously satisfy a convergence condition. The robustness of the voice processing model is improved.

The invention discloses a Microblog advertisement user detection method. The method includes the steps that user information is collected, and classifying identification is added for each user; feature extraction is performed on the acquired used information, and content of advertisement users and common users and a feature attribute set of action are obtained on the basis of a data mining method; according to the feature attribute set with the classification identification, model training is performed, and an advertisement user discrimination model is obtained. Due to the fact that all features of the Microblog users are comprehensively analyzed, accuracy and regression of the trained discrimination model are high, comprehensive performance is good, and the defect that a detection tool built in Sina is not flexible on the advertisement users is overcome.

The embodiment of the invention provides an image processing method, device and system and a computer storage medium. The method comprises the step of generating a generative adversarial network, which comprises a generative model and a discrimination model, through the following way that: inputting an input image into the generative model to obtain an output image, wherein the input image and theoutput image have different styles; inputting the output image into the discrimination model to obtain an output value; according to a pre-defined loss function, training the generative model and thediscrimination mode until convergence is realized; and using the trained generative model to obtain a target image which has different styles with the input image of the trained generative model. Therefore, by use of the embodiment of the invention, the generative model can be obtained through training, on the basis of the generative model of the generative adversarial network, the target image which has different styles with the input image can be obtained, so that a dataset used for vehicle detection can be expanded, and therefore, a vehicle detection effect is guaranteed.

The invention relates to an extremely low illumination image enhancement method based on a generative adversarial network. The method comprises the following steps of: obtaining original image data ofa shot image through an imaging sensor of shooting equipment, and preprocessing the original image data; wherein the original image data is Bayer array data Bayer array s; inputting the preprocessedimage data into a generative adversarial network; wherein the generative adversarial network comprises a generation model and a discrimination model, the generation model is used for image enhancement, and the discrimination model is used for training learning, so that the generated image is enhanced to an optimal image; and processing an output result of the generative adversarial network, and storing the output result as an image. According to the invention, under-exposure and darker images shot in an extremely low illumination environment or a night environment can be enhanced into clear and bright pictures through the method provided by the invention.