38results about How to "Reduce callouts" patented technology

The invention discloses a traveling vehicle vision detection method combining laser point cloud data, belongs to the field of unmanned driving, and solves the problems in vehicle detection with a laser radar as a core in the prior art. The method comprises the following steps: firstly, completing combined calibration of a laser radar and a camera, and then performing time alignment; calculating anoptical flow grey-scale map between two adjacent frames in the calibrated video data, and performing motion segmentation based on the optical flow grey-scale map to obtain a motion region, namely a candidate region; searching point cloud data corresponding to the vehicle in a conical space corresponding to the candidate area based on the point cloud data after time alignment corresponding to eachframe of image to obtain a three-dimensional bounding box of the moving object; based on the candidate region, extracting a direction gradient histogram feature from each frame of image; extracting features of the point cloud data in the three-dimensional bounding box; and based on a genetic algorithm, carrying out feature level fusion on the obtained features, and classifying the motion areas after fusion to obtain a final driving vehicle detection result. The method is used for visual inspection of the driving vehicle.

The invention discloses a Tibetan speech corpus labeling method and a Tibetan speech corpus labeling system based on cooperative batch active learning. The system comprises a sample selection module,a manual labeling module, a labeling decision-making module, a labeling person evaluation module and a training set generation module. According to the method and the system, the construction of a sample evaluation function and the proving of the submodular function property of the sample evaluation function are solved through the adjacent optimal batch sample selection method, and the construction of a labeling decision function and the modeling of a labeling person evaluation model and a labeling person auxiliary learning model are solved through the labeling committee collaborative labelingmethod. In addition, the system disclosed by the invention can realize the functions inducing the optimal selection of a sample, the labeling and evaluation of a user, the sharing of labeling information and Tibetan speech knowledge, auxiliary learning of the labeling person and the like, so that the labeling quality of the Tibetan speech data is improved, and the construction of the speech corpus is accelerated.

The invention relates to a handing method and device of a locomotive replacing operation based on a CTC3.0 system. The handing method comprises the steps that a locomotive detaching-hanging function module is increased in the CTC3.0 system, the locomotive detaching-hanging function is realized by using an original plan, operation flow and route state information in the CTC3.0 system; according tothe interlocking route information, the operation arrangement conditions of a lead locomotive entering and leaving the warehouse are designed as the basis for planning, train planning train number information and the line information and signals or track occupancy conditions of the locomotive entering and leaving the warehouse are utilized, the internal related information of the lead locomotive and the train number are discovered, the calculation of planned orders for the lead locomotive detaching-hanging operation is realized, and the handling of detaching-hanging and route of locomotives through the detaching-hanging operation plan is completed. Compared with the prior art, according to the handling method and device of locomotive replacing operation based on the CTC 3.0 system, the problem of unmanned operation of the locomotive operation plan at an existing general speed hub station is solved, and the advantage that the safety jaw control is carried out when the locomotive detaching-hanging operation is handled is realized.

The invention discloses a manufacturing method for a rack of a marine low-speed diesel engine. The manufacturing method comprises the following steps that firstly, the rack is divided into a rack head segment module, a rack middle segment module and a rack tail segment module, wherein the rack middle segment module comprises a plurality of rack middle segments which are the same in structure and size; secondly, according to the design size of the rack, all part plates of all the rack segments are cut for blanking, then grooves are processed and welded, and finally the plates are stored in a classified mode; thirdly, a CO2 welding method is adopted for welding the rack head segment module, the rack middle segment module and the rack tail segment module step by step respectively; fourthly, the rack head segment module, the rack middle segment module and the rack tail segment module are jointed and folded; and fifthly, annealing treatment is finally conducted on connection weld joints to reduce residual stress produced by welding of connection areas, and after overall heat treatment on the rack is completed, aging treatment and flaw detection are conducted to make the rack reach all design standards. The manufacturing method has the advantages of being high in production efficiency and good in interchangeability and universality.

The invention discloses a behavior recognition method based on a multi-instance markov model. The behavior recognition method includes the following steps that local features of each video are abstracted and a feature histogram of a local video block is used for expressing certain local movement of behavior; a random sampling manner is used for obtaining multiple local video blocks, the local video blocks form multiple markov chains and the markov chains are expressed as continuous movement of some local movement in time; under a frame of multi-instance study, the model selects the markov chain with the most recognition performance to express behavior. In the time of testing, multiple markov chains are composed in the same manner to express the videos and the scores of the markov chains are calculated. If the scores of the markov chains are larger than a certain threshold value, the markov chains belong to the behavior, and otherwise, the markov chains do not belong to the behavior. According to the multi-instance markov model, the aim of behavior recognition under complicated scenes is achieved and labels of the videos are reduced.

The invention relates to a training method of a pulmonary nodule detection model, and belongs to the technical field of medical image processing. The training method comprises the steps that an original CT image sample is acquired, the original CT image sample comprises a labeled sample and an unlabeled sample, the labeled sample is cut, and a pulmonary nodule block sample is obtained; training apositioning model according to the labeled samples; pre-training the classification model according to the pulmonary nodule block sample; sequentially inputting unlabeled samples into the trained positioning model and the pre-trained classification model, and classifying the pulmonary nodule blocks; and combining the classified pulmonary nodule blocks and pulmonary nodule block samples into training set data, and training the classification model again to obtain a final classification model. Unlabeled pulmonary nodule blocks are classified through the pre-trained classification model, more training samples are obtained, the classification model is trained through the more training samples, the labeling process is reduced, and the training efficiency of the pulmonary nodule detection modelis improved.

The invention discloses a model training method and device, equipment and a storage medium. The model training method comprises the following steps: performing model training based on a training data set to obtain an intermediate model; based on a preset data acquisition type, acquiring data from the to-be-labeled data set as to-be-tested data, and generating a model reasoning result of the to-be-tested data based on the current intermediate model; based on the dynamic test set and the fixed test set, testing and evaluating the model reasoning result; and if the evaluation result of the fixed test set does not meet the standard condition, adjusting a preset data acquisition type based on the evaluation result of the dynamic test set, performing iterative training on the intermediate model based on the to-be-tested data corresponding to the adjusted preset data acquisition type until the evaluation result of the fixed test set meets the standard condition, and obtaining a target model. Under the driving of small sample annotation data, dynamic model evaluation is carried out, model iteration is rapidly carried out, and then the effect of rapidly generating a model conforming to annotation is achieved.

The invention provides a key point detection model training method, a key point detection method and a device thereof, and the training method comprises the steps: obtaining a training sample set; the training samples of the first type comprise images, key point positions and key point availability, and the training samples of the second type comprise images; a teacher network is trained by using the training sample of the first type and the first total loss function; the first total loss function comprises loss for constraining the position and availability of the key point; when the training sample is selected as a second type, the training sample is input into the trained teacher network and student network to obtain a key point position and availability prediction result; the prediction result is input into a second total loss function so as to train a student network through knowledge distillation; and the second total loss function comprises a key point position for constraining the student network to learn the teacher network and the loss of the availability prediction result. Therefore, the marking time consumption can be reduced, the low-quality image detection effect is improved, and the detection real-time performance is improved.

The invention discloses a 3D point cloud semantic segmentation and migration method based on meta-learning, and relates to the technical field of robot navigation. The method comprises the following steps: constructing a Point Net network model; selecting a training data set; for each training data set, forming a training task set by utilizing different types of data; constructing a meta-learningframework; according to the meta-learning framework, training the Point Net network model through each training task set; selecting a test task set; and inputting the test task set into the trained Point Net network model for testing until the gradient update value of the model converges. According to the method, a trained model is loaded in a new environment task, optimal similar task parametersare used, and the training efficiency of the indoor scene semantic segmentation method of the new task is high; the semantic segmentation capability of different tasks is learned through a meta-learning framework, it is ensured that learning features can be suitable for different migration environments, and the generalization performance of the model is improved.

The invention relates to an elevator fault early-warning method based on the Internet of Things technology and a coupled graph neural network. The elevator fault early-warning method based on the Internet of Things technology and the coupled graph neural network comprises the following steps: a, acquiring data in an elevator car in real time, and transmitting the collected data to a cloud; b, preprocessing the data, and carrying out feature extraction; and c, utilizing a pre-trained coupled graph neural network model for analyzing the data acquired in real time at the cloud, and carrying out early-warning on elevator faults in advance. According to the invention, early-warning for the elevator faults is realized in advance through the coupled neural network model, and therefore the elevator fault early-warning method has higher accuracy and precision compared with a traditional mode under the conditions that data are unbalanced and fault data are few.

The invention relates to the field of image processing in artificial intelligence, and discloses a target tracking method and device, computer equipment and a storage medium. The method comprises thesteps that a real-time mechanical arm state and an image sequence containing a target object are acquired according to a preset sampling period; the real-time mechanical arm state and the image sequence containing the target object are input into a target object tracking model, and motion parameters output by the target object tracking model are obtained, wherein the target object tracking model is a prediction model constructed based on a DDPG algorithm; and the motion state of the mechanical arm is controlled according to the motion parameters to enable the mechanical arm to move following the target object. The target tracking method and device can reduce the development cost of the target tracking, improve the training efficiency of the target tracking model, and can be applied to theconstruction of the smart city. At the same time, the target tracking method and device also relates to a blockchain technology.

The invention provides an identity verification model training method and device, an electronic device and a storage medium, and relates to the technical field of artificial intelligence. The identityverification model training method comprises: acquiring sample identity feature data and an identity label corresponding to the sample identity feature data; performing first learning processing on an identity branch network in a pre-established identity verification model according to the sample identity feature data so as to train the identity branch network; performing second learning processing on a domain branch network in the identity verification model according to the sample identity feature data so as to train the domain branch network; and performing third learning processing on theidentity branch network through the identity label and the trained domain branch network so as to construct the identity verification model according to the trained identity branch network. Accordingto the invention, domain differences in the identity feature data of the same person can be eliminated during identity verification.

The invention provides an integrated meter tail structure which includes a wire fixing portion, a lead, a first meter tail portion and a second meter tail portion. The wire fixing portion is of a plate-shaped structure provided with a wire fixing round hole, a cantilever rod and a buckle. The lead passes through the wire fixing round hole and is integrally fixed in the wire fixing portion to forma wire concentration area. The wire fixing portion is fixedly connected with the first meter tail portion through the cantilever rod and the buckle, and the first meter tail portion is disposed abovethe second meter tail portion. In the structure, the lead is integrally fixed on the wire fixing portion and forms an integrated structure together with the meter tail portions, to ensure that the position of a component is fixed, so that the lead can be soldered in one time by a wave soldering technology in the same manner as other components, the repeated soldering is avoided, and the productionefficiency of an electric meter is improved. The invention further provides the electric meter with the integrated meter tail structure. The electric meter is simple in structure and convenient to assemble, internal elements are connected firmly, and it is ensured that the electric meter has good overall usability.

The embodiment of the invention provides an RPA and AI combined question and answer knowledge base updating method and device, equipment and a storage medium. The method comprises the steps of obtaining to-be-audited questions; calculating an uncertain index of each to-be-audited question in a preset question and answer knowledge base; determining a target question in the to-be-audited questions according to the uncertain indexes; and obtaining knowledge points matched with the target question, and adding the target question into a preset question and answer knowledge base according to the matched knowledge points so as to update the preset question and answer knowledge base. Because a small number of to-be-audited questions with relatively high uncertainty are screened out from a large number of to-be-audited questions as target questions according to the uncertainty indexes of the to-be-audited questions, and the knowledge points of the small number of target questions are labeled and matched by a data trainer, the processes of manually screening and labeling the knowledge points of the to-be-audited questions can be reduced. Therefore, the matching efficiency of the knowledge points corresponding to the to-be-audited questions is improved, the updating efficiency of the question-answer knowledge base is improved, and the performance of the question-answer knowledge base canbe rapidly improved.

The embodiment of the invention discloses a training method for enhancing an image generation model, and an image processing method and device. The training method of the enhanced image generation model comprises the following steps: acquiring a to-be-processed sample image, and determining the category of the to-be-processed sample image, wherein the category comprises a plain scanning image and an enhanced image; identifying a region of interest of the to-be-processed sample image to obtain a sample image including the region of interest; and training an initial model based on the sample image including the region of interest and the category of the sample image to obtain an enhanced image generation model. Through the technical scheme disclosed by the embodiment of the invention, the contrast of the enhanced image and the application flexibility of the plain-scan image are improved, the injury to a patient due to multiple times of injection of the enhanced contrast agent is avoided, and the diagnosis efficiency of a doctor is improved.

The invention relates to the technical field of computers, in particular to a data annotation method and device of a target element, terminal equipment and a computer readable storage medium, and the method comprises the following steps: determining a to-be-annotated target element based on a game; the method comprises the following steps: determining to-be-labeled data after screen recording in a game process, labeling a target element in the to-be-labeled data to form labeled data, labeling at least four point locations when labeling the target element, forming a rectangular frame when the four point locations are connected end to end, and enabling the target element to be in the rectangular frame; preprocessing the annotation data through the script to form training data, wherein the training data at least comprises a plurality of to-be-annotated pictures; establishing a training model according to the training data, and training the training model through a script; and training a new model according to the training data, the script and the trained training model. The invention provides a method capable of reducing manual labeling of target elements.

The invention relates to the artificial intelligence technology, and discloses a text brief report generation method comprising the following steps: using a first brief report set and a first categorylabel set to train a pre-constructed original label classification network so as to obtain a standard label classification network, using the standard label classification network to execute label classification operation on a second brief report set, obtaining a second category label set, summarizing the first category label set and the second category label set to obtain a category label set, dividing the first brief report set and the second brief report set by using the category label set to obtain a category brief report set, generating a brief report template by using the category briefreport set, receiving a brief report text to be generated, and executing named entity identification on the brief report text to be generated to obtain an entity set, and executing a combination operation on the entity set and the brief report template to obtain a brief report. The invention further provides a text brief report generation device, electronic equipment and a computer readable storage medium. According to the invention, the calculation resource can be saved, and the brief report generation accuracy is improved.

The present invention relates to a method and device for locomotive change and hang operation based on the CTC3.0 system. The track state information realizes the function of locomotive detachment; according to the interlocking route information, design the operation arrangement conditions of the main machine entering and exiting the warehouse as the basis for planning, and use the planned train number information, the line information of the locomotive entering and exiting the warehouse, and the signal or track occupancy , excavate the internal correlation information between the lead locomotive and the train number, realize the calculation of the lead locomotive detachment operation plan, and complete the handling of each locomotive detachment route through the detachment operation plan. Compared with the prior art, the present invention solves the problem of unmanned preparation of the locomotive operation plan in the existing general-speed hub station, and realizes the advantages of safety card control when the locomotive unhooking operation route is handled.

The invention discloses a machining tool state monitoring method and system. The method includes: collecting a small amount of current data in real time after the parameters of the machining system change and storing the collected data in a target domain data set; calculating the target domain data set The distribution difference of each feature parameter between the source domain data set and the target domain data set; determine whether the distribution difference of each feature parameter between the target domain data set and the source domain data set is smaller than the distribution difference threshold of the feature parameter; Learn to obtain the target domain migration model, and use the target domain migration model as the current monitoring model; otherwise, continue to collect the current data and store the collected data in the target domain data set, and after the data volume of the target domain data set reaches the set requirements , and use the data in the target domain data set to train to obtain the current monitoring model; use the current monitoring model to monitor the tool status online. By using the present invention, the adaptability of the monitoring system can be improved quickly and efficiently.

The invention relates to a transferable image recognition method and device, and relates to the technical field of image recognition. The method includes: determining an image type of an input image recognition model; The source domain image is passed through the feature extractor and class predictor, and the cross entropy loss is determined; when the input image is an unlabeled target domain image, the target domain image is passed through the feature extractor and the domain discriminator, while the feature extractor and Category predictor; determine the adversarial loss according to the output of the domain discriminator and the similarity between the target domain image and the center point of each source domain image; determine the information maximization loss according to the output of the category predictor; according to the cross entropy loss, confrontation Loss and Information Maximization Loss to Optimize Image Recognition Models. Through the technical solution, the performance of target image recognition can be effectively improved, the annotation for target image recognition can be effectively reduced, and manpower and material resources can be greatly reduced.

The invention discloses an illegal behavior recognition method based on convolution and graph convolution, and the method comprises the following steps: S1, collecting behavior object images, and building an image database; s2, extracting the posture of each behavior object from the image database by adopting a human body posture extraction scheme HighHRNet; s3, partial optimization is carried out on the HighHRNet; s4, constructing an undirected graph based on the key point pose information identified by the HighHRNet, and performing preprocessing enhancement on the undirected graph; and S5, determining the abnormal type of the violation. According to the invention, the key point detection module is improved, so that the key point information can be extracted more quickly while the precision is maintained; the diversity of data is improved by applying various undirected graph data enhancement modes, so that the training model has stronger generalization ability; a video sequence is converted into an undirected graph sequence to construct a 3D feature graph space, and feature information is extracted based on a 3D feature graph, so that the accuracy of an illegal behavior recognition algorithm is improved.

The invention discloses a breast ultrasound image style conversion method. The method comprises the steps of acquiring an original image and a style image; loading a pre-trained convolutional neural network, and extracting content features of the original image and style features of the style image through the pre-trained convolutional neural network; respectively defining a content loss functionbetween the generated image and the original image and a style loss function between the generated image and the style image according to the content features and the style features; defining a totalloss function of the generated image according to the content loss function and the style loss function; and performing iterative operation on the generated image through a gradient descent algorithmuntil the total loss function is converged. According to the breast ultrasonic image style conversion method, breast ultrasonic images of other styles can be converted into the target style, the converted breast ultrasonic images can serve as a data set for deep learning of intelligent diagnosis, and therefore the breast ultrasonic images of various styles do not need to be collected, and the collection cost of the breast ultrasonic images is reduced.

The invention discloses a cross-supervised model training method, an image segmentation method and related equipment. The cross-supervised model training method comprises the following steps: acquiring a sample image; the sample image is input into a basic twin network, the twin network comprises a first network and a second network which are the same in network structure, and basic parameters of the first network are different from basic parameters of the second network; calculating the loss of the second network by taking the prediction result of the first network as the label of the second network, and calculating the loss of the first network by taking the prediction result of the second network as the label of the first network so as to obtain the loss of the twin network; iteratively updating the parameters of the first network and the second network based on the loss of the twin network until a training cut-off condition is met; and taking the trained and updated first network or second network as a target model. By means of the mode, annotation of the sample images can be reduced, and the model operation accuracy is improved.

The invention discloses a manufacturing method for a rack of a marine low-speed diesel engine. The manufacturing method comprises the following steps that firstly, the rack is divided into a rack head segment module, a rack middle segment module and a rack tail segment module, wherein the rack middle segment module comprises a plurality of rack middle segments which are the same in structure and size; secondly, according to the design size of the rack, all part plates of all the rack segments are cut for blanking, then grooves are processed and welded, and finally the plates are stored in a classified mode; thirdly, a CO2 welding method is adopted for welding the rack head segment module, the rack middle segment module and the rack tail segment module step by step respectively; fourthly, the rack head segment module, the rack middle segment module and the rack tail segment module are jointed and folded; and fifthly, annealing treatment is finally conducted on connection weld joints to reduce residual stress produced by welding of connection areas, and after overall heat treatment on the rack is completed, aging treatment and flaw detection are conducted to make the rack reach all design standards. The manufacturing method has the advantages of being high in production efficiency and good in interchangeability and universality.

The embodiment of the invention discloses an image semantic understanding method and device, equipment and a storage medium. The method comprises the steps that target image information is acquired; the target image information is input into a pre-trained semantic understanding model, a semantic classification result output by the semantic understanding model is obtained, and the semantic understanding model is obtained through training based on associated image information and text information; and determining a semantic understanding result according to the semantic classification result. According to the method provided by the embodiment of the invention, the target image information is classified through the semantic understanding model obtained by training directly based on the obtained associated image information and text information, so that the annotation amount during model training is simplified, and the semantic understanding model with accurate classification can be obtained by training with less annotation.

The invention discloses a model training method for cross-domain sentiment analysis based on a convolutional neural network, belongs to the field of cross-domain sentiment classification, and aims tosolve the problem of cross-domain sentiment analysis. The method comprises: S1, preprocessing a text; S2, training a word vector model; S3, performing cross-domain model migration; wherein the step S3is carried out in sequence; a neural network model is trained through a source domain; migrating the trained model. Firstly, a weight value of a convolution kernel in a model is shared, a convolutionkernel weight trained in a source domain is used for extracting corresponding features in a target domain, a small part of data in the target domain is trained again, parameters of a full connectionlayer weight of the previously trained model are adjusted, and the effect is that model migration is conducted on a cross-domain emotion text.