62results about How to "Save labeling time" patented technology

The invention relates to a point cloud labeling method and device, computer equipment and a storage medium. The method comprises the steps of obtaining first annotation information of a first object in first frame point cloud data, wherein the first annotation information comprises size information and direction information of an annotation box and the type and number of the first object; Obtaining a second object in the to-be-labeled second frame point cloud data; And when the type of the second object is the same as that of the first object, configuring the first annotation information to the second object. According to the method, the marking efficiency of point cloud marking can be improved.

The invention discloses a method for automatically establishing artificial intelligent image recognizing training materials and annotation files. The method is characterized by comprising the following steps of S1, designing a 3D model and scene, and conducting three-dimensional modeling through CAD type software numbers; S2, synthesizing training images, reading the 3D model, materials and background information in the step S1 through tool software, conducting simulating camera shooting on the 3D model, materials and background to obtain shot images with different distances, angles, time durations and scenes, combining the images obtained through simulation and the object materials, synthesizing and outputting the training images, and recording the model space positions of all focused objects under current states during outputting; S3, establishing the object annotation files, and conducting annotation on the model space position information, recorded in the step S3, of all the focused objects to generate the annotation files; S4, storing the annotation files. By means of the method, the high-quality training materials and annotation files can be rapidly generated.

The invention discloses an image processing method, image processing equipment and a computer readable medium, and is applied to the technical field of image processing. The method comprises the following steps: inputting a target image into a preset field detection model to obtain a target field level image corresponding to the target image; determining a plurality of target crowdsourcing users from the crowdsourcing users according to the annotation information of the crowdsourcing users, and allocating the target field level image to the plurality of target crowdsourcing users for annotation; respectively obtaining labeling results of the plurality of target crowdsourcing users on the target field level image to obtain a plurality of labeling results; And determining a target labeling result from the plurality of labeling results according to a preset determination rule, and performing training according to the target field-level image and the target labeling result to obtain an image recognition model. By adopting the method and the device, the data annotation efficiency is improved, the annotation time is saved, and the cost is reduced.

The invention relates to the field of classification models of artificial intelligence, and provides a pre-labeling model training method and device, a certificate pre-labeling method and device, equipment and a medium. The method comprises the steps: obtaining a target labeling category, target description, model performance parameters and an image sample set; crawling a to-be-migrated category in a target classification and identification library by using a text similarity technology; searching a to-be-migrated model from the target classification and identification library through a simulation target identification technology, and identifying a target region of each image sample; performing target fine tuning to obtain a fine tuning area, and inputting the image sample, the fine tuning area and the target labeling category into the to-be-migrated model; acquiring and marking a target labeling area by using a transfer learning technology; determining a loss value according to the target labeling area and the fine tuning area; and training the to-be-migrated model until the training is completed to obtain a pre-labeled model. According to the invention, automatic training of a zero-labeling image sample set is realized, the pre-labeling model is obtained, and the manual labeling time and workload are reduced.

The invention belongs to the field of intelligent identification of articles, and relates to an article identification method for efficiently labeling samples. The method comprises the following process steps: S11, preparing static pictures of corresponding categories as training samples according to requirements for training to form a training data set; S12, performing image annotation, and combining all the sample annotation files to obtain final training sample data for training an article detection model; s13, performing model training by adopting a resnet-101 backbone network; performingmodel training operation for multiple times on the basis of the existing model by modifying the training parameters until a model meeting the requirements of a user is obtained; s14, performing targetdetection by adopting a mask rcnn algorithm to obtain a prediction category, contour information of a segmented target area and a bounding box. According to the method, background interference is reduced, the accuracy of target matching is effectively improved, the calculated amount is reduced, and the target matching speed is increased; meanwhile, the sample labeling mode can greatly reduce thesample labeling time, and manpower and time are saved.

The invention discloses a corpus labeling method, device and equipment, relates to the technical field of artificial intelligence, and aims to generate text corpora of different violation types in batches and save corpus labeling time. The method comprises the steps of performing sentence segmentation processing on text data in different business scenes, and storing text corpora formed after sentence segmentation processing in a corpus database; dividing preset standard violation descriptions into different violation categories by taking the semantic points as units; according to the entity concepts contained in the semantic points and the logic relation between the entity concepts, building keyword semantic rules, and the keyword semantic rules being violation expressions mapped on different violation categories for standard violation description; and matching target text corpora containing different violation categories from the corpus database by utilizing the violation expression, and labeling the target text corpora based on the violation categories.

The invention discloses an image annotation method, device and equipment and a storage medium, and belongs to the technical field of artificial intelligence. According to the invention, the content ofdifferent images in the same image set is generally related; due to the fact that the annotation results of the different images in the same image set are generally related, when the next image is annotated, the annotation result of the previous image is automatically loaded, time expenditure caused by reannotation of the next image from the beginning is avoided, and the annotation efficiency ofthe next image is improved. Through the method, the remaining images can be labeled by utilizing the labeling result of the labeled images in the image set, so that the overall labeling time of the image set is saved, and the overall labeling efficiency of the image set is improved.

The embodiment of the invention discloses a method and device for outputting information. A specific embodiment of the method comprises the steps of obtaining a fundamental frequency curve corresponding to a to-be-labeled sample syllable; extracting a fundamental frequency sequence from the fundamental frequency curve; converting the fundamental frequency sequence into a sample value sequence; clustering the sample value sequence and a reference sequence with a known boundary tone type to obtain the boundary tone type of the sample value sequence as the boundary tone type of the to-be-labeledsample syllable; and outputting the boundary tone type of the to-be-labeled sample syllable. According to the embodiment, automatic marking of the boundary tone in the English speech synthesis systemis realized, so that the marking time is shortened, and the cost is saved.

The invention discloses an electronic medical record named entity recognition system and method. The method comprises the following steps: performing data cleaning, performing rule-based pre-annotation on cleaned data, returning a result to an annotation algorithm for secondary annotation and generating a pre-annotation data set, and returning the result to annotation personnel for correction and annotation so as to generate a standard data set; correcting the rule and the algorithm by comparing and analyzing the difference between the pre-annotated data set and the standard data set; and acquiring online prediction data, supplementing the online prediction data into a standard data set through manual checking and verification, sending original data into a pre-labeling system to supplement a pre-labeling data set, and retraining a model iteration model after accumulating to a certain scale. According to the method, the whole industrial application process of named entity recognition is integrated and transformed, and a named entity recognition framework suitable for an industrial scene is constructed.

The embodiment of the invention discloses an annotation object control device. According to the device, after annotation boxes are adopted to annotate objects in an image, control points of the annotation boxes are acquired, and a corresponding relation between the control points and the objects is established; if a mouse clicking event is monitored, a mouse position is acquired; and if it is known that the mouse position is in at least two annotation boxes through judgment, target control points closest to the mouse position in the annotation boxes are acquired, and target objects corresponding to the target control points are obtained according to the corresponding relation. According to the embodiment, the mouse position is acquired after the mouse clicking event is monitored, the target control points closest to the mouse position are calculated, and the target objects corresponding to the target control points are obtained according to the corresponding relation between the control points and the objects. Therefore, the target objects can be determined according to the mouse clicking position, the number of interactions between an annotation tool and a user is reduced, the annotation process is simpler, and meanwhile annotation time is saved.

The invention provides an ultrasonic image segmentation system and method based on a side window attention mechanism. The system comprises an ultrasonic data acquisition module, a first data transmission module, a server module, a second data transmission module and a visualization module. The first data transmission module is respectively connected with the ultrasonic data acquisition module, the server module and the visualization module; the server module comprises an image preprocessing unit and an ultrasonic image segmentation model; the ultrasonic image segmentation model comprises a convolutional neural network, a cavity convolution module, a side window attention module and a classifier; the convolutional neural network is respectively connected with the image preprocessing unit, the cavity convolution module, the side window attention module and the classifier; the second data transmission module is connected with the visualization module; according to the method, the convolutional neural network based on the side window attention mechanism is adopted, side window convolution is fused into the convolution process, the segmentation effect is more accurate in combination with the attention mechanism, and the calculation amount is greatly reduced.

The invention discloses a semi-automatic labeling method, which belongs to the technical field of image labeling, and comprises the following steps of: initializing each pixel generation region of a picture as a candidate region, and pairing all adjacent candidate regions in pairs to generate a calculation list; similarity calculation is carried out according to the calculation list, including color similarity calculation and texture value similarity calculation, and a first score and a second score are obtained respectively; score calculation is carried out according to the calculation list to obtain a third score; and according to the first score, the second score and the third score, adding the calculated similarities, and combining every two similarities reaching a similarity combination threshold in the list until all candidate regions are combined. After the label candidate boxes are generated by adopting the semi-automatic labeling method, only the proper candidate boxes need tobe manually selected, so that the labeling time is greatly saved, and the labeling efficiency is improved. The invention further discloses a semi-automatic labeling device, electronic equipment and acomputer storage medium.

The invention discloses a rock slice microscopic image automatic labeling method, which is used for improving a manual point tracing and frame drawing link of an image, realizing automatic labeling of a complex rock slice image, classified labeling of different particle types in the same graph, and point tracing and frame drawing of each particle, and endowing a label name corresponding to the particle. According to the method, automatic point tracing and frame drawing of the rock slice image are realized, the labeling time of experts is greatly saved, the experts can be effectively assisted to carry out batch complex rock slice image labeling work, the accuracy is relatively high, and the working efficiency is greatly improved.

The invention relates to a full-automatic V-cut device for a PCB board. The device comprises an information processing terminal, a storage module for storing circuit diagram information, a collectionmodule for extracting the PCB board circuit diagram information, a recognition module for comparing the stored circuit diagram with the PCB board circuit diagram and recognizing and distinguishing thesame, a marking module for automatically generating V-cut coordinate program data, wherein the information processing module is respectively communicated with the storage module, the collection module, the recognition module and the marking module. The CAM manual annotation timeliness is saved through the V-cut device automation, and the error rate of the data annotation is greatly reduced at thesame time; the generated V-cut program improves the machining efficiency, the manual data entry is free, and the yield is improved.

The embodiment of the invention discloses a title entity recognition model training method and device and a title entity recognition method and device. A specific embodiment of the method comprises the steps of training an initial mask text recognition model based on an article information sample group to obtain a trained mask text recognition model, the initial mask text recognition model comprising a pre-trained initial text coding model, and the pre-trained initial text coding model comprising an item information sample group; the trained mask text recognition model comprises a trained text coding model; determining the trained text coding model and a preset decoding network as an initial title entity recognition model; and training the initial title entity recognition model based on the title sample group to obtain a trained title entity recognition model. According to the embodiment, the marking time is shortened, and the recognition accuracy and robustness of the title entity recognition model are improved.

The embodiment of the invention discloses a point cloud labeling method and device, a storage medium and electronic equipment. The method comprises the steps: determining a first labeling box containing point cloud data of a to-be-labeled object in a labeling main scene picture; respectively mapping the center coordinates in the first annotation box into an annotation view, and generating a second annotation box corresponding to the annotation view and associated with the first annotation box; in the second labeling box, adjusting the size of the second labeling box in a preset adjustment direction based on the point cloud data in the second labeling box; mapping the adjusted coordinate data of the second annotation box into the first annotation box to obtain a third annotation box; and adjusting the size of the third labeling box to obtain a fourth labeling box containing all the point cloud data of the to-be-labeled object. According to the embodiment of the invention, in the labeling process, the precision can be determined without repeatedly adjusting the labeling view angle, the labeling time is shortened, the labeling efficiency is improved, and the labeling precision is guaranteed.

The invention discloses an image annotation method and device, and is used for solving the technical problem that the reliability of annotation quality cannot be ensured when an existing image annotation method is used for annotating an image related to a power transmission scene. The method comprises the steps of taking to-be-labeled image data; sending the to-be-annotated image data to annotation equipment to complete an annotation process of the power transmission hidden danger targets contained in the to-be-annotated image data to obtain to-be-classified image data; classifying the to-be-classified image data through an image classification model, and determining a classification result; and resending the classification error image data to the corresponding annotation equipment, and carrying out secondary annotation to obtain classification correct image data. By means of the method, the labeling accuracy is improved, and then the reliability of labeling quality is guaranteed.

The invention relates to a deep learning SAR (Synthetic Aperture Radar) image ship identification method based on a self-supervision condition. The method comprises the following steps: firstly, preprocessing SAR data, acquiring an image pixel threshold value by utilizing accumulative inverse exponential probability distribution, carrying out rapid segmentation by utilizing the threshold value to obtain a binary image, then carrying out eight-neighborhood communication processing on the binary image, acquiring geometric information of a candidate target, constructing an SAR ship slice data set according to the geometric information of the candidate target, finally, establishing a CNN model, and training and tuning the CNN model so as to be used for self-supervision identification on the ship target. According to the CNN model based on the self-supervision thought, only a small number of training samples need to be labeled in the recognition process, the sample labeling time is greatly shortened, and the ship detection efficiency is improved; and the backbone model adopts a lightweight model Shufflenet network, model parameters are few, high training precision can be obtained with short training time, the convergence speed is high, and the precision is high.

The embodiment of the invention discloses a driving data labeling method, device and system. The method comprises the following steps: acquiring unlabeled driving data; utilizing a target detection model in a convolutional neural network constructed by labeled sample driving data to obtain obstacle information included in the driving data; utilizing a segmentation model in the convolutional neuralnetwork to obtain road surface recognition information included in the driving data; and based on driving data obtained by a plurality of sensors in the same area, performing matching filtering on the obstacle information and the road surface recognition information to obtain a marking result of the driving data. By utilizing the embodiment of the invention, a large amount of manpower screening and labeling cost can be saved, and the problem that a large amount of wrong labeling is easily generated in automatic labeling can be avoided, so that the data labeling accuracy can be improved whilethe data labeling quality is ensured.

The invention is suitable for the technical field of robots, and provides a target detection model training method and device and communication equipment, and the method comprises the steps of obtaining first images which are manually labeled images, and one first image uniquely corresponds to one piece of labeling information; processing the first image to obtain a second image different from the first image, the annotation information of the second image being the same as the annotation information of the first image; and training a to-be-trained target detection model according to the first image and the second image to obtain a trained target detection model. Through the method, the trained target detection model with relatively high detection precision can be rapidly trained.

The invention discloses a processing method of a sentence vector generation model based on artificial intelligence, which is applied to the technical field of artificial intelligence and is used for solving the technical problems of low learning efficiency of text sentence vectors and poor expression accuracy of the sentence vectors. The method provided by the invention comprises the following steps: acquiring a text sample which does not carry a sample label; inputting the text sample into a BERT module in a sentence vector generation model to be trained, and outputting an intermediate feature of the text sample through the BERT module; selecting different discarding masks for the middle features of the same text sample to execute dropout discarding operation, and obtaining middle feature vectors, corresponding to the discarding masks, of the same text sample; determining the intermediate feature vectors belonging to the same text sample as positive samples, and determining the intermediate feature vectors belonging to different text samples as negative samples; and training the sentence vector generation model according to the positive sample, the negative sample and a loss function to obtain a trained sentence vector generation model.

The invention belongs to the technical field of traffic, particularly relates to a traffic mode recognition method based on a genetic algorithm and a fuzzy neural network, and aims to recognize the traffic mode of an elevator, perform preprocessing and label construction on training data by using a k-means clustering method, and improve the traffic mode recognition accuracy. And constructing a three-layer fuzzy neural network model to output the prediction probability of each traffic mode, and initializing a weight coefficient in the constructed fuzzy neural network model by using a genetic algorithm. According to the method, the neural network is effectively prevented from falling into a local optimal solution when the target is optimized, and the program performance in the traffic mode recognition process is improved. The weight of the neural network is initialized by using a genetic algorithm, and a basis can be provided for subsequent neural network back propagation optimization. By adopting the fuzzy logic method, the situation of overfitting of the neural network can be reduced, and the process of rapid change of the passenger flow volume under a certain special condition is smoothed, so that the training of the neural network and the prediction of the traffic mode are more accurate.

The invention discloses an irregular object semantic segmentation rapid labeling method, which comprises the following steps of: 1) carrying out image data acquisition and preprocessing to obtain a data set of an image; 2) carrying out grid division on the image in the data set, and labeling a plurality of obtained small grids; 3) converting the label into a segmentation mask and carrying out edge optimization to obtain a data set for model training; and 4) selecting a segmentation model structure to carry out model training. Compared with pixel-level labeling, the irregular object semantic segmentation rapid labeling method has the advantages that the labeling time is greatly saved while the precision is slightly reduced; compared with rectangular frame labeling, the method can obviously improve the recognition precision, can greatly improve the labeling efficiency, and reduces the interference of background information on model training.

The invention relates to the technical field of behavior analysis, in particular to a method and device for labeling abnormal behaviors. The method comprises the steps of pre-training a neural networkbased on a current abnormal behavior data set to obtain a first neural network model; copying all network architectures and model parameters except the output layer in the first neural network model,and creating a second neural network model; adding an output layer of which the tensor size corresponds to the number of abnormal behavior detection categories into the second neural network model; training the second neural network model added with the output layer according to a target data set obtained by labeling in a PASCAL VOC labeling mode or a COCO labeling mode to obtain an abnormal behavior labeling model; inputting the to-be-annotated data set into the abnormal behavior annotation model, and annotating each piece of to-be-annotated data in the to-be-annotated data set to obtain anannotated data set; judging whether the labeled data set is correctly labeled or not; and inputting the data with wrong annotations into the abnormal behavior annotation model for re-annotation.