40 results about "Final Labeling" patented technology

The invention discloses a method for automatically labeling images based on community potential subject excavation, which comprises the following steps: 1) adopting a hidden Dirichlet allocation modelto excavate implicit subjects in a single community; 2) after obtaining the probability distribution of image labels and the implicit subjects by analyzing community potential subjects, deleting theimage labels of which the probability of the community image labels and the implicit subjects is smaller than a set value k to perform 'de-noising' filtration on the community image labels; 3) generating image candidate labeling labels of the images to be labeled by propagating similar image labels; 4) optimizing the image candidate labeling labels according to the relativity between the image candidate labeling labels and the implicit subjects of the images; and 5) obtaining the final labeling result of the images through the information fusion of a plurality of communities. The method makesfull use of the information on different communities in which the images are positioned and the information on the community potential subjects in a social shared network to label the images, and compared with the conventional labeling method, the generated labeling result is more accurate.

Embodiments of the invention provide a data labeling method, device and system and a storage medium. The data labeling method comprises the following steps of: selecting a first number of unlabeled data from an unlabeled data set of a data pool; pre-labeling the first number of unlabeled data by utilizing a labeling model so as to obtain pre-labelling information of the first number of unlabeled data, wherein the pre-labelling information comprises a pre-labelling result; outputting a second number of unlabeled data and a pre-labeling result of the second number of unlabeled data to a displaydevice, wherein the second number of unlabeled data is at least a part of the first number of unlabeled data; receiving feedback information aiming at the pre-labeling result of the second number of unlabeled data; and determining a final labeling result of the second number of unlabeled data. According to the method, unlabeled data is submitted to users to be checked after being pre-labelled by the data labeling system, so that the labor cost of data labeling can be greatly decreased.

The invention discloses an interactive classification labeling method for a sketch data set in a computer. The interactive classification labeling method comprises the following steps: in a learning process, carrying out multi-feature extraction on a labeled sketch data set, carrying out metric learning of feature space, and calculating a distance measurement function; in a selection process, if judging that non-labeled sketches do not exist in the sketch data set, coming to an end to obtain a final result; or, according to the result of metric learning, carrying out feature space construction on the sketch data set to be labeled, carrying out layering clustering, and selecting an optimal sample subset; carrying out online labeling, carrying out interactive confirmation on the sketches in the optimal sample subset, carrying out classification labeling on a confirmed sample, and updating the labeled sketch data set; maintaining the non-labeling state of remaining non-similar sketches, and updating the sketch data set to be labeled; then constantly circulating the process until the user completes all sketch labeling to obtain a final labeling result.

The invention discloses an automatic picture labeling method and device, relates to the technical field of picture processing, and aims to realize simultaneous labeling of multiple labels of a pictureand improve the accuracy of the labels while ensuring the labeling efficiency. The method comprises the following steps: training a multi-modal feature extraction model based on gallery data; constructing a visual semantic similarity nearest neighbor index of each picture and the label group according to the corresponding relationship between each picture and the label group and the classification; extracting features of a to-be-detected picture through a feature extraction model to obtain a feature vector, and matching a similar picture from the picture library data based on the feature vector and the visual semantic similarity nearest neighbor index; screening out an initial label of the to-be-detected picture according to the frequency and the weight of the keyword in the label group corresponding to the similar picture; and performing label filtering and weight sorting on the initial labels by adopting a pre-trained word vector model to obtain a final label group of the to-be-detected picture. The device applies the method provided by the scheme.

The invention discloses a high-efficiency labeling method and system for a high-resolution remote sensing target big data set. The method comprises the following steps: intercepting an image picture on each obtained high-resolution remote sensing image according to a preset condition; creating an image picture creation engineering vector file; labeling attributes on the vector layer on the engineering vector file; judging the target object type of the labeling position according to the labeling attribute; marking the selected target objects according to different types of the target objects; storing the marked engineering vector file; judging whether labeling information of other types of images exists or not, and if yes, corresponding the engineering vector file with the labeling box to the multispectral image and the fused image; or, if not, storing the final labeling result. A complete system for marking the target object is formed, marking can be conducted on the panchromatic imagewith the highest resolution ratio, finally, the panchromatic image corresponds to the multispectral image and the fusion image, marking is more attached to the target object, and marking is faster and more accurate.

The invention provides a target detection automatic labeling method and device based on moving object detection, and the method comprises the steps: employing a scene classifier to determine the scene category of a to-be-detected video, selecting at least one moving object detection algorithm having a corresponding relationship with the scene category, and enabling the to-be-detected video to comprise the same type of moving objects; detecting a moving object in a plurality of frames of pictures of the to-be-detected video through at least one moving object detection algorithm, and marking a target box for the detected moving object; tracking the moving object in the multi-frame picture by adopting a preset target tracking algorithm, and marking a target box for the tracked moving object; and removing the overlapped target frames on the moving object in the multi-frame picture by adopting a preset algorithm, and taking the remaining target frames as a final labeling result of the moving object. By adopting the automatic labeling method provided by the scheme of the invention, the labor cost of target detection labeling can be greatly saved, the efficiency is higher than that of a traditional pure manual labeling method, and the labor cost is effectively saved.

The invention discloses a literature screening and labeling platform for evidence-based medicine, which comprises a back-end server, and a management end, an arbitration end and at least two labelingends which are respectively connected with the back-end server, and is characterized in that the management end imports literatures, publishes labeling tasks to the labeling ends and exports final labeled literature data; Each labeling end carries out multiple rounds of labeling processing on the documents contained in the labeling task to obtain labeled document data; The back-end server obtainslabeled literature data obtained after each labeling end processes the labeling task in real time, automatically judges the consistency of labeling results of different labeling ends for each literature, and sends the labeling task of the literature to the original labeling end or the arbitration end for processing according to the judgment result; And the arbitration end receives and processes the labeling task of the literature and outputs corresponding final labeling literature data. The platform can conveniently, accurately and quickly complete the screening process meeting the evidence-based requirement, and can export standardized question data for subsequent use.

The invention discloses a fine-grained image recognition algorithm of distributed labels based on inter-class similarity. The fine-grained image recognition algorithm comprises the following steps: (1) extracting feature representation of an input image by using a backbone network; (2) calculating center loss and updating a category center by utilizing a center loss module through feature representation; (3) calculating, by a classification loss module, classification loss (such as cross entropy loss) by using the feature representation and final label distribution, wherein the final label distribution is obtained by calculating the weighted sum of one-hot label distribution and distributed label distribution generated by a category center; and (4) performing weighted summation on the center loss and the classification loss to obtain a final target loss function so as to optimize the whole model. The problem of overfitting is effectively relieved by reducing the certainty degree of model prediction, discriminative features of fine-grained data can be accurately learned, data of different fine-grained categories can be accurately and efficiently distinguished, and the invention can be widely applied to the fields of visual classification and multimedia.

The invention discloses a sample labeling method, a sample labeling device, a storage medium and equipment. The sample labeling method comprises the steps of obtaining a to-be-labeled sample set; determining a labeling executor set for labeling the to-be-labeled sample set; according to the attribute information of each to-be-annotated sample in the to-be-annotated sample set and the annotation capability attribute information of each annotation executor in the annotation executor set, dividing the annotation executor set into a plurality of annotation groups, so that the annotation efficiencyof each annotation group meets a preset annotation efficiency index; sending the to-be-labeled sample set to a plurality of labeling groups; according to the attribute information of each to-be-annotated sample and the annotation capability attribute information of each annotation executor in the annotation group, distributing the to-be-annotated samples to each annotation executor in the annotation group; and obtaining a plurality of initial labeling results of each to-be-labeled sample, and determining a final labeling result. According to the invention, the labeling efficiency can be improved on the premise of ensuring the labeling quality.

The embodiment of the invention belongs to the technical field of artificial intelligence, and relates to a data labeling method based on cross validation and related equipment. The method comprises the steps of obtaining a small sample initial standard data set; inputting the initial standard data set into a classification model for cross validation to obtain an initial standard data model; obtaining a large sample data set, inputting the large sample data set into the initial marking data training model for pre-marking, and determining a correction data set according to a pre-marking result; inputting the correction data set into an initial standard data training model for cross validation to obtain a final standard data model; and labeling the received data to be labeled through the final labeling data model. In addition, the invention also relates to a blockchain technology, and the to-be-labeled data can be stored in the blockchain. The labeling efficiency is improved to a greater extent, and false labeling data in the labeling process is detected by adopting a cross validation method; and repeated annotation of most of data is avoided.

The invention discloses an automatic express packaging device and method. The device comprises a bagging mechanism, a bag conveying mechanism, a wind force bag opening mechanism, a delivery mechanism, an adhesive tape removal mechanism, a bag sealing mechanism, a transportation-out mechanism, a drive part, a detection part and a control part. Each of the mechanisms is fixed by a fixing plate. The packaging method comprises the steps that express bags are conveyed into a conveyor belt through the bagging mechanism and then pressed tightly, and bag openings are blown open by an air blowing device; then goods are pushed by the delivery mechanism and slide into the bags, then the express bags are released, and the express bags fall in to the adhesive tape removal device under the effect of the gravity; after the express bags are fixed by a push device in position, bag opening adhesive tape is torn off by a scraper, after the adhesive tap is torn out, an overturning plate turns over, the bag openings are sealed, and the goods are transported out; and meanwhile, express labels are printed by printing equipment and transported out along with the express bags, and the final labeling task is completed manually. By the adoption of the device, the full automation of the express packaging is achieved; the structure is simple, and mass production is facilitated; and the logistics speed is increased effectively.

The embodiment of the invention relates to the technical field of information processing, in particular to a corpus labeling method and device, a server and a storage medium. The corpus annotation method comprises the steps of acquiring an even number of manual annotation results of an initial corpus and a model annotation result of the initial corpus, wherein the model annotation result of the initial corpus is obtained according to a preset annotation model, and the preset annotation model is obtained by training a plurality of manually annotated initial corpora; and obtaining a unique labeling result meeting a preset condition from all labeling results including the manual labeling result and the model labeling result, and taking the labeling result as a final labeling result of the initial corpus. According to the embodiment of the invention, the high-quality corpus annotation result of the initial corpus can be obtained, and the influence of a single annotator on the corpus annotation quality is reduced.

The embodiment of the invention provides a data labeling method and device and a storage medium. The data labeling method comprises the steps: classifying to-be-labeled data according to contents expressed in the to-be-labeled data; according to the type of the to-be-labeled data, selecting a corresponding pre-annotation model to perform primary annotation on the to-be-labeled data to obtain a first labeling result; carrying out secondary labeling on the data subjected to primary annotation to obtain a second labeling result; and taking the second labeling result as a final labeling result. Bymeans of the mode, the efficiency and accuracy of data labeling are guaranteed; the model is used for pre-labeling, so that the efficiency problem of manual labeling is solved; and meanwhile by meansof secondary labeling, the accuracy of pure model labeling is improved.

The present invention discloses a salt water black fungus production technology. The production technology includes the steps that salt water is prepared, wherein tap water is processed by a filter and a sterilizer to obtain 100 parts by weight of sterile water, 6-10 wt% of pickle salt, 3-6 wt% of sucrose and then 2-4 wt% of seasonings are added into the 100 parts by weight of sterile water, and the mixture is put into a pickle jar to conduct a sealed fermentation at a temperature of 25-40 DEG C for 5-30 days to obtain the sterile processed water; black fungus is subjected to pre-treatment, wherein the black fungus is soaked with 0.5-1 wt% of sodium alginate for 2-4 hours, then the soaked black fungus is rinsed with water, then the rinsed black fungus is soaked with 0.5-2 wt% of calcium chloride for 3-6 hours, the soaked black fungus is repeatedly washed with water, the impurities are removed, and the cleaned black fungus is subjected to air-drying treatment; the treated black fungus is mixed with the prepared salt water at a weight ratio of 1:(2-6), and the mixture is vacuum bagged or bottled; and the treated salt water black fungus is subjected to sterilization treatment through pasteurization at a temperature of 85-95 DEG C for 30-60 minutes, then the sterilized products are subjected to segmented cooling, the water on the package surface is wiped out, the products are subjected to air drying further, and the air-dried products are subjected to final labeling, code printing and box packing. The salt water black fungus can meet market demands, facilitates far transportation and foreign export, enhances economic efficiency, and is convenient and fast, and suitable for life tastes of modern people.

The invention discloses a short video label extraction method and device, a computer device and a storage medium. The method comprises the steps of extracting a prior label of a short video from a label lexicon according to description information of the short video; constructing directed edges between the short videos according to the sequence of watching the short videos by the user; accumulating and normalizing all the in-degree edges between every two short videos to obtain weights of directed edges between the short videos; constructing a priori label vector of the short video through thepriori label of the short video; calculating to obtain a final label vector of each short video through the priori label vector of the short video and the weight of the directed edge between the short videos; and determining a final label of the short video based on the final label vector. According to the method, label extraction is carried out on the relationship between the short videos, so that the label extraction efficiency is improved.

The invention discloses a short video automatic labeling method based on feature and multi-label enhanced representation, which comprises the following steps: reconstructing an original feature matrixby using a dictionary mapping matrix, a product of public low-rank representation and a sparse error matrix to form a multi-view low-rank representation item; clustering an overall data set to obtainall data sets and potential label correlation information in different clusters to form global and local label correlation learning items; taking the common low-rank representation as a prediction label, subtracting the prediction label from a real label to obtain a labeling error, and minimizing the labeling error to form a minimized labeling error term; and weighting the multi-view low-rank representation item, the global and local label correlation learning item and the minimized labeling error item to obtain a total target function, optimizing the total target function by using an alternating direction multiplier method, introducing a Lagrange multiplier, and sequentially iteratively updating each matrix variable until the value of the target function converges to obtain a final labeling result. According to the method, the accuracy in the short video multi-label labeling problem is improved.

The invention discloses a named entity category labeling method and system for a Chinese electronic medical record, and mainly solves the problem of how to more accurately identify and label named entities in the Chinese electronic medical record under the multi-classification problem based on a joint neural network model. The method mainly comprises the steps of obtaining word vector representations of Chinese words, pre-extracting structured features of entities based on a HmapCNN model, extracting context sequence features of the entities based on a bidirectional LSTM model, performing sequence labeling of the entities based on CRF, and obtaining a final labeling result. The model is further packaged, an interaction system based on a B/S framework is designed and developed, a graphical display interface friendly to a user is provided, and processing such as use of the model and display and export of a result is facilitated.

The invention relates to the technical field of data processing, in particular to a data labeling system and method based on automatic verification. The method comprises the following steps: S1, establishing a standard annotation data sample library; s2, issuing a to-be-labeled data task; s3, enabling the user to select to-be-annotated data and carry out annotation processing; s4, responding to the user labeling action, and storing a user labeling result; s5, judging whether multi-person labeling results of the same data are consistent or not; by means of the principle of large numbers, manualqualitation is converted into machine quantitative control, inconsistent labeling result verification and final labeling result generation decision are automatically completed, and the efficiency isimproved while the labeling data quality is guaranteed.

The invention discloses a method for automatically labeling entities in a medical text. The method aims to correctly label entities in a text and comprises the following steps: collecting the text; preprocessing the text; collecting entity dictionary data; preprocessing the dictionary data to generate an abbreviation dictionary, a phrase dictionary and other entity dictionaries; labeling the textdata by applying a labeling rule to obtain a labeling entity; and labeling post-processing: processing a labeling result by adopting a fuzzy recognition mode, adding entities with missed recognition,removing nested entities, and finally obtaining a final labeling result.

The invention discloses an automatic labeling method for medical text data. The method comprises the following steps: obtaining preprocessed medical text data according to original medical text data;performing initialization operation on preset parameters to obtain an initialization result; obtaining a primary population according to the initialization result and the preprocessed medical text data; carrying out fitness calculation on the primary population to obtain fitness corresponding to a first preset number of primary individuals; determining a second preset number of pairs of parents according to the primary population; obtaining a second preset number of cross individuals according to the second preset number of pairs of parents; obtaining a second preset number of variant individuals according to the second preset number of cross individuals; obtaining a second preset number of candidate individuals according to the second preset number of variant individuals; determining a new population according to the second preset number of candidate individuals; and obtaining a final labeling result according to the new population. Through the technical scheme of the invention, the obtained labeling result is high in accuracy.

The invention discloses an entity label determination method and device, a storage medium and a processor. The method comprises the steps of obtaining target data of a target entity; determining a plurality of fields of the target data; determining at least one atomic label of the target entity based on the value of each field; and performing logic processing on the at least one atom label to obtain a final label of the target entity. According to the invention, the technical problem of low efficiency of adding labels to entities is solved.

The invention discloses a crowd feature determination method and device in information pushing and a readable storage medium. The method comprises the steps of obtaining a historical pushing data setof historical pushing information with the same category as to-be-pushed information; using a preset assessment index to segment the historical push data set in each dimension to obtain a segmentationresult, and the segmentation result comprises push data corresponding to each label in each dimension; determining a feature label set according to the segmentation result; segmenting the historicalpush data set according to the feature label set, screening out historical push data corresponding to the feature label set, and continuously segmenting the historical push data corresponding to the feature label set in each dimension by adopting the assessment indexes until a final label set is obtained; and determining crowd features of crowds corresponding to the to-be-pushed information according to the final label set. According to the scheme, an operator can scientifically and effectively plan target audiences.