63results about How to "Improve generation accuracy" patented technology

The embodiment of the invention discloses a nerve network modeling method and device used for forming titles; the method comprises the following steps: obtaining a plurality of sample documents and standard titles corresponding to each sample document; inputting each sample document into a to-be-optimized model, and allowing the to-be-optimized model to output reference titles corresponding to each sample document according to the received sample documents; coupling the reference titles with standard titles corresponding to each sample document, thus obtaining semantic gap between the reference title and the standard title; optimizing target parameters in the to-be-optimized model according to the semantic gaps corresponding to each sample document. The method and device can optimize the parameters in the to-be-optimized model in a sentence level according to the semantic gaps between the reference titles and the standard titles; compared with a conventional method capable of optimizing the to-be-optimized model only in the vocabulary level, the nerve network modeling method and device can improve model stability, and can improve the title forming accuracy.

The invention discloses a laser altimeter vertical control point generating method in combination with a stereopair. The method comprises the steps that dense elevation point cloud data is obtained through a stereo camera; then point cloud data within a footprint is classified according to elevations, and the surface feature classification number at different elevations within a light spot is obtained; finally, large-spot laser radar waveform decomposition is guided through the classification number, and a generalized vertical control point is obtained. The aim of cooperative mapping of the stereo camera and satellite-borne laser radar is achieved, the result of three-dimensional positioning of a laser altimeter is combined with the mapping result of the stereo camera, and automatic extraction of the generalized vertical control point is achieved.

The present invention discloses a Chinese complete sentence generating method and device. The method of the present invention includes: a candidate word generated the last time and a candidate in a pinyin string are obtained; a candidate directed graph is constructed and a candidate with a biggest conditional probability is selected which corresponds with the candidate generated the previous time from candidates which correspond with the initial arc of the directed graph. Based on the candidate with the biggest conditional probability, a complete sentence result of the pinyin string is obtained. The embodiment of the present invention also provides a corresponding device. When computing probability of the candidate which corresponds with the initial arc of the candidate word directed graph, the embodiment of the present invention applies word frequency of the candidate combination and of the candidate word generated the previous time, that is, applies the context information to generalize a complete sentence, thus improving the complete sentence generating accuracy and the candidate word generating accuracy.

The invention discloses a method for extracting road characteristics in a high-resolution image. A generator G capable of accurately identifying road characteristics is obtained by optimizing structures of a generator G, a discriminator D and a generative adversarial network V (D, G) and training parameters. In the training process, the difference between the generated sample Pg (x) and the real sample PData (x) is evaluated by using the Waserstein distance, so that the problem of gradient non-convergence of a loss function in the existing KL divergence or JS divergence evaluation can be effectively avoided. The road characteristics obtained through identification of the generator G obtained through the method are closer to the actual situation, and the accuracy is higher.

The invention provides a robot welding device and a method for acquiring a welding seam track. The robot welding device comprises a welding robot, an XY workbench, a calibration reference block and acomputer; the welding robot comprises a welding gun, a camera and a 3D scanner which are fixedly arranged on a mechanical arm; the calibration reference block is fixedly arranged on the upper end surface of the XY workbench; the welding robot firstly performs photographing on the calibration reference block which is fixedly arranged on the XY workbench through the camera so as to obtain current spatial position information of the robot; 3D scanning and synchronous photographing is carried out on a to-be-welded part to obtain a three-dimensional model and a picture of the to-be-welded part; processing and template matching identification are carried out on the picture of the to-be-welded part to obtain a welding seam picture; and the welding seam picture is attached to a corresponding position on the surface of the three-dimensional model of the to-be-welded part, and if the intersecting line of the two intersecting planes on the three-dimensional model coincides with the welding seam,a welding seam track is automatically generated according to the spatial coordinate of intersecting line of the two intersecting planes on the three-dimensional model.

The invention provides a remote sensing image style conversion method based on text data, and the method comprises the steps: constructing a data set, and obtaining a text data set and a to-be-converted image data set; generating a low-resolution image, extracting sentence features according to the text data, and generating a low-resolution remote sensing image and corresponding image features incombination with noise; generating a high-resolution image, extracting word features according to the text data, and generating a high-resolution remote sensing image and image features of the next layer by combining the low-resolution features of the previous layer; calculating a loss function, detecting the matching degree of the generated image and the text, and generating a corresponding lossfunction; and image style conversion: taking the generated high-resolution image as a reference style image, and performing style conversion according to a cyclic consistency principle and an adversarial loss function. The method has the beneficial effects that the high-resolution images are generated layer by layer from the text data, so that the generation precision from the text to the images is greatly improved, and the vacancy of style conversion of the text data is made up.

The invention relates to a data generation method and device, electronic equipment and a storage medium, and relates to the technical field of big data, and the method comprises the steps: obtaining adata table, and extracting all fields contained in the data table; determining a plurality of data generation rules for all the fields, and associating the data generation rules with the types of thefields to obtain a mapping relationship; determining a target rule for a to-be-processed field from a plurality of data generation rules in combination with the mapping relationship according to thetype of the to-be-processed field in at least one data table; and generating corresponding data for the to-be-processed field in at least one data table by utilizing the target rule. According to themethod, the target rule for the to-be-processed fields is determined through the data generation rule based on the big data, the data corresponding to each to-be-processed field can be quickly and accurately generated according to the target rule, and the data generation efficiency and accuracy are improved.

The embodiment of the invention discloses a user portrait generation method and device, a storage medium and equipment, and the method comprises the steps: obtaining the user feature data of a first user and the user feature data of at least two second users, and enabling the at least two second users and the first user to have a social association relation; clustering the at least two second users to obtain at least two user sets, the similarity between the user feature data of the second users in each user set satisfying a similarity condition; determining key user feature data correspondingto each user set according to the user feature data of the second user in each user set; and generating a user portrait of the first user according to the key user feature data corresponding to eachuser set and the user feature data of the first user. Through the embodiment of the invention, the generation accuracy of the user portrait can be improved.

An information processing apparatus includes a processor. The processor is configured to receive first image data representing a document, and generate, by processing corresponding to appearance characteristics of the document, second image data not representing information of a deletion target out of information represented in the first image data but representing information other than the information of the deletion target.

The invention relates to the field of big data processing, and provides ancient poem keyword extraction and poem line generation method, device and equipment, and a medium. The method comprises the following steps: obtaining poem information in a poem extraction instruction; inputting the poem information into a poem annotation model based on BERT and classical characters in an ancient poem keyword extraction model, and splitting and annotating the poem information through the poem annotation model to obtain poem sequence information; identifying a poem type through a poem type identificationmodel; inputting the poem sequence information into a CRF-based type constraint model corresponding to the poem type in an ancient poem keyword extraction model, and performing marking and constraintprediction processing on the poem sequence information to obtain a keyword result; and determining ancient poetry keywords according to keyword results. According to the method, the ancient poetry keywords reflecting the poetry in the poetry information are automatically extracted through the poetry labeling model based on BERT and classical characters and the type constraint model based on CRF.

The invention discloses a time sequence behavior capture box generation method and device based on a self-attention network. The method comprises the steps of obtaining a target video; obtaining an input feature matrix according to the target video, and inputting the input feature matrix into a first self-attention model to obtain an action probability sequence, a starting point probability sequence and an ending point probability sequence; wherein the first self-attention model is obtained by connecting a plurality of groups in series, each group comprises at least one self-attention unit, and the target feature matrix corresponding to each group forms feature offset in time sequence; generating a candidate capture box set according to the action probability sequence, the start point probability sequence and the end point probability sequence; and extracting relative position information among the candidate capture boxes, and inputting the relative position information into a second self-attention model to obtain a target capture box set. According to the method, the target capture frame set is obtained by arranging the two self-attention models, and the generation precision of the target capture frame set is improved.

The invention relates to artificial intelligence, and provides a case analysis method and device, equipment and a storage medium. According to the invention, the method comprises the steps: performing element extraction on case information to obtain feature information, wherein the feature information comprises appeal information, deblurring information, evidence information and fact information; carrying out classification processing on the appeal information and the deblurring information to obtain an appeal category and a deblurring category, and carrying out disambiguation processing on the appeal category and the deblurring category to obtain an appeal entity and a deblurring entity; generating a dispute focus according to the appeal entity and the deferential entity, obtaining a graph path corresponding to the dispute focus from a preset legal knowledge graph, generating a trial element according to the dispute focus, the evidence information and the fact information, and obtaining a result corresponding to the trial element from the graph path as a judgment prediction result. According to the invention, the prediction judgment result of the case can be accurately analyzed. In addition, the invention also relates to a block chain technology, and the judgment prediction result can be stored in a block chain.

The invention relates to the field of computers, in particular to the field of artificial intelligence, and provides a video retrieval method, device and equipment and a storage medium to solve the problems of low quantification efficiency and low accuracy. The method comprises the steps of performing feature extraction on image features of a to-be-retrieved video to obtain a first quantitative feature, then obtaining a second candidate video with high category similarity with the to-be-retrieved video based on the first quantitative feature, and finally taking the second candidate video with high content similarity with the to-be-retrieved video as a target video. According to the texture feature loss value corresponding to each training sample, a quantization control parameter is adjusted, so that the target quantization processing sub-model learns the sorting capability of the target texture feature sub-model, it is ensured that the sorting effects of the two sub-models tend to be consistent, and the end-to-end model architecture has high efficiency. The target quantization processing sub-model can obtain the corresponding quantization features based on the image features, and the generation accuracy of the quantization features and the recall performance of video retrieval are improved.

The embodiment of the invention discloses a detection method, a device and an electronic device of a stationary object. The method comprises the following steps: acquiring point cloud data of a sceneand characteristic information of each data point in the point cloud data; connecting the data points in the point cloud data with triangulation network, the triangulation network model being generated, and the preset camera being used to photograph the triangulation network model, and the image being obtained; acquiring a first data point corresponding to each pixel point in the image, and obtaining a feature map of point cloud data according to the feature information of each first data point; the feature map being inputted into the classification model, and the data points corresponding tothe stationary objects in the point cloud data being obtained. That is, in this embodiment, by acquiring the feature map of the point cloud data of the scene and taking the feature map as the input ofthe classification model, the still object in the scene is accurately detected, and the generation accuracy of the still map is improved.

The embodiment of the invention provides a construction error information obtaining method and device, and relates to the field of building construction. The method comprises the steps of: obtaining abuilding information model; obtaining a building construction video; generating a construction local model according to the building construction video; and generating construction error informationaccording to the building information model and the construction local model. Visibly, by implementing the implementation mode, the purpose of construction monitoring can be achieved, and then consumption of a large number of human resources and time cost is avoided.

The invention discloses a universal fingerprint generation method and device based on artificial intelligence and a cloud server, a storage medium and an electronic device. The method comprises the following steps: obtaining a plurality of original fingerprint images output by a target fingerprint generation model, wherein the target fingerprint generation model is a model obtained by training anoriginal generative adversarial network by using a target sample fingerprint sub-image, the target fingerprint generation model is used for outputting a plurality of original fingerprint images afterinputting random data, the target sample fingerprint sub-image is an image screened from a plurality of first sample fingerprint sub-images and used for training an original generative adversarial network, and the first sample fingerprint sub-image is an image intercepted from the sample fingerprint image; obtaining a target image corresponding to each original fingerprint image in the plurality of original fingerprint images; selecting a target fingerprint image from the plurality of original fingerprint images; and determining a fingerprint indicated by the target fingerprint image as a universal fingerprint. The technical problem of low universal fingerprint generation accuracy is solved.

The invention discloses an all-sky aurora image classification method and system. The method comprises the steps that: all-sky aurora images to be classified are preprocessed; the preprocessed all-sky aurora image is input into a pre-trained classifier, a classification result is output, and the classification result is in an arc shape, a valance slow coronal shape, a radiation coronal shape or a hot spot shape; and the classifier, the encoder and the decoder form an integral model for joint training. According to the method, the aurora images can be trained and classified under the condition that the data size is huge and only a small number of labels exist, and high accuracy is obtained; and, according to the method, all-sky aurora image classification learning is carried out under the condition that only a small number of labels exist, the semi-supervised generation model based on the variational auto-encoder is used, and all-sky aurora image classification can be carried out with high accuracy.

The embodiment of the invention discloses an image processing method and device, an electronic terminal and a storage medium, and the method comprises the steps: inputting a received CT image of a first dose into a first generator to generate a predicted CT image of a second dose, wherein the second dose is greater than the first dose, the first generator is included in a preset generative adversarial network in a training process, and is trained along with the preset generative adversarial network based on an adversarial loss function and a cyclic loss function, and the loop loss function comprises an image-based style loss function and / or a perception loss function. And the first generator in the trained network is utilized to generate a high-dose prediction CT image according to the low-dose CT image, so that noise and artifacts in the low-dose CT image are inhibited, and the image quality is improved.

The invention provides a robot welding device and a method for acquiring a welding seam track. The robot welding device comprises a welding robot, an XY workbench, a calibration reference block and acomputer; the welding robot comprises a welding gun, a camera and a 3D scanner which are fixedly arranged on a mechanical arm; the calibration reference block is fixedly arranged on the upper end surface of the XY workbench; the welding robot firstly performs photographing on the calibration reference block which is fixedly arranged on the XY workbench through the camera so as to obtain current spatial position information of the robot; 3D scanning and synchronous photographing is carried out on a to-be-welded part to obtain a three-dimensional model and a picture of the to-be-welded part; processing and template matching identification are carried out on the picture of the to-be-welded part to obtain a welding seam picture; and the welding seam picture is attached to a corresponding position on the surface of the three-dimensional model of the to-be-welded part, and if the intersecting line of the two intersecting planes on the three-dimensional model coincides with the welding seam,a welding seam track is automatically generated according to the spatial coordinate of intersecting line of the two intersecting planes on the three-dimensional model.

The embodiment of the invention discloses a multi-sensor data fusion method and device, electronic equipment and a medium. One specific embodiment of the method comprises the following steps: acquiring a point cloud data set in a target area and an image shot by a vehicle-mounted monocular camera; performing feature extraction on the image to generate an image feature map; meshing the target areato generate a sub-target area set; determining a mean value of the point cloud data in each sub-target area to generate point cloud data mean value information; generating a point cloud data feature information set; determining mapping coordinates of the point cloud data corresponding to each piece of point cloud data feature information in the image feature map; generating splicing feature information; generating fusion feature information; generating obstacle detection information; and sending the obstacle detection information to a target display terminal and displaying the obstacle detection information on the target display terminal. According to the embodiment, the waste of computing resources is reduced, and the accuracy of obstacle information generation is improved.

A system and method for generating a stained image including the steps of obtaining a first image of a key sample section; and processing the first image with a multi-modal stain learning engine arranged to generate at least one stained image, wherein the at least one stained image represents the key sample section stained with at least one stain.

The invention discloses a text generation model training method and device, a storage medium and computer equipment, and relates to the technical field of machine learning. The method comprises the following steps: acquiring question texts and a plurality of context texts in multiple rounds of dialogues; generating a first text vector matrix corresponding to the plurality of context texts and a reply text vector matrix corresponding to the question text by using an initial text vector generation model; utilizing an initial topic vector extraction model to respectively extract a context text topic vector corresponding to the first text vector matrix and a reply text topic vector corresponding to the reply text vector matrix; constructing a loss function based on the context text topic vector and the reply text topic vector; and jointly carrying out iterative training on the initial text vector generation model and the initial topic vector extraction model based on the loss function, and constructing a text generation model. The method is suitable for training the text generation model.

The invention discloses a road element vector generation method and system. The method comprises the following steps: acquiring point cloud data and an original image collected in the same area; projecting the point cloud data to a two-dimensional plane to obtain a target image corresponding to the point cloud data; performing semantic segmentation on the road elements in the original image and the target image to obtain semantic segmentation results; and vectorizing the road elements based on the point cloud data and the semantic segmentation result to obtain a road element vector. The technical problem that the generation error of the road element vector is large in the prior art is solved.

The invention discloses an amplified license plate batch generation method and system, a storage medium and equipment, and the method comprises the steps: obtaining a license plate recognition image; acquiring a character image forming an amplified license plate based on the license plate recognition image; based on the obtained character images, combining the character images to generate a plurality of amplified license plate images; based on the generated amplified number license plate image, performing fuzzy processing on the amplified number license plate image to obtain a fuzzy processed license plate image, and storing the fuzzy processed license plate image in a database. According to the scheme, image processing can be performed on the amplified number license plate image, batch amplified number license plate images are generated, the sample size of the amplified number license plate is increased, and the identification rate of the amplified license plate can be improved.

The invention relates to a category label generation method and device, electronic equipment and a storage medium, and belongs to the technical field of computers. The method comprises the following steps: based on the target number, selecting a target image group from the alternative image set, generating a first category label for each target image in the target image group; based on the presetnumber, selecting a reference image group from the alternative image set, and obtaining a first accuracy rate according to the category label marked for each reference image in the reference image group, adjusting the target number according to the first accuracy rate, continuing to select a target image group from other alternative image sets based on the adjusted target number, and generating afirst category label for each target image in the target image group. According to the method, the first category label is automatically generated for the target image, manual labeling is not needed,the cost is reduced, and the generation efficiency is improved. Moreover, the target images are selected from the alternative image set by adjusting the number of the targets, so that the accuracy isimproved.

The present invention discloses a Chinese complete sentence generating method and device. The method of the present invention includes: a candidate word generated the last time and a candidate in a pinyin string are obtained; a candidate directed graph is constructed and a candidate with a biggest conditional probability is selected which corresponds with the candidate generated the previous timefrom candidates which correspond with the initial arc of the directed graph. Based on the candidate with the biggest conditional probability, a complete sentence result of the pinyin string is obtained. The embodiment of the present invention also provides a corresponding device. When computing probability of the candidate which corresponds with the initial arc of the candidate word directed graph, the embodiment of the present invention applies word frequency of the candidate combination and of the candidate word generated the previous time, that is, applies the context information to generalize a complete sentence, thus improving the complete sentence generating accuracy and the candidate word generating accuracy.

The invention relates to a video description text generation method and device, equipment and a storage medium. The method comprises the steps that at least two kinds of modal data corresponding to video data to be processed are acquired; inputting the at least two types of modal data into a modal association network to obtain modal association results corresponding to the at least two types of modal data; the modal association result represents association degrees between at least two types of modal data and theme contents of the to-be-processed video data; filtering the at least two types of modal data based on the modal association result to obtain filtered modal data; inputting the filtered modal data into a description text generation network to obtain a description text of the to-be-processed video data; the description text is used for describing the theme content. According to the embodiment of the invention, the interference of noise of some modal data can be eliminated, so that the description text better fitting the theme content of the to-be-processed video data can be generated, and the generation precision of the description text is improved.

The embodiment of the invention provides an artificial defect graph generation and model training method and a related device, and the provided artificial defect graph generation model training method comprises the following steps: training a matrix model of an artificial defect graph generation model and a graph authenticity discrimination model through an alternate training mode; under the condition of ensuring that the graph authenticity discrimination model has relatively high discrimination capability, the artificial defect graph generated by the matrix model of the artificial defect graph generation model is difficult to identify by the graph authenticity discrimination model, so that the matrix generation precision of the matrix model of the artificial defect graph generation model obtained by training can be improved; and a large number of artificial defect matrixes meeting quality requirements can be generated by utilizing the matrix model of the artificial defect graph generation model obtained by training.