32results about How to "Labeling is simple" patented technology

The invention relates to an aided driving system based on a collision early-warning algorithm, and belongs to the technical field of computer vision and intelligent aided driving. The system comprises a detection and distance measurement module which collects road condition information in the driving process of an automobile through a camera, and carries out the detection, recognition and distance measurement of an obstacle through a YOLOv3 model; a collision early-warning module which is used for carrying out the collision prediction classification, calculating the time required by collision, giving early-warning judgment in time and carrying out early-warning broadcast on a driver; a positioning module which is used for acquiring driving position information of the vehicle by utilizing GPS / IMU integrated navigation, automatically switching the system to an IMU for positioning when a GPS signal is lost, and switching the system to GPS positioning again when the GPS signal is normal; and a GUI display and cloud video backup module which is used for displaying the identification video stream, the driving state and the map software annotation information in real time and carrying out cloud backup. According to the invention, the prediction precision and real-time performance of the auxiliary driving system can be improved.

The present invention relates to a radio signal identification method based on an end-to-end convolutional neural network. The method is characterized in that: an original I / Q sampling data of an observation window is subjected to execution of preprocessing and identification through a convolutional neural network in order. The preprocessing step is that: the original I / Q sampling data of the observation window is taken as input, and a frequency spectrum waterfall plot is output after discrete Fourier transform and data format alignment processing; the step of identification through the convolutional neural network is that: the frequency spectrum waterfall plot obtained by preprocessing is taken as input, and a one-dimensional boolean vector configured to show whether all the signals to beidentified are existed or not is output after the input passes through a CNN feature extraction layer, an MLP feature mapping layer and a BR multi-tag classification layer. Compared with the mode offeature extraction and classification identification, the radio signal identification method employs the end-to-end technical solution thinking to avoid complex and low-efficient feature engineering,improve the signal identification accuracy, robustness and intelligence level, and has important meaning of radio monitoring of important areas and important activity scenes.

The invention provides an electronic map caption tagging method and device; the tagging method comprises the following steps: obtaining attribute information of linear figures and map captions surrounding the linear figures in the electronic map, wherein the attribute information comprises map caption tagging positions and map caption outer wrapping frame parameters; converting the linear figure into one or more point symbols according to preset pixel values; according to the map caption attribute information and point symbol pixel values, obtaining tapping positions of the map caption when the map caption and the point symbols are not overlapped in the electronic map, and tagging the map captions according to the obtained tagging positions. The method converts the linear figure into point symbols, and the map caption and point symbols are not overlapped, thus fast avoiding the map caption from the linear figure. The method uses point symbol avoiding strategy to uniformly deploy the map captions on two sides of the linear figure; the method and device are simple, high in efficiency, thus improving mapping efficiency, and the map can be clearer and tidy.

Provided is a two-dimensional multi-feature fusion underwater target rapid detection method, relating to the underwater target detection. By adopting a transverse multi-domain feature fusion module and a longitudinal multi-source feature fusion module, the method includes performing the Fourier transform domain, fractional Fourier transform domain, wavelet transform domain or other multi-domain transforms for each submarine sound signal; after the sparse decomposition, extracting the main features corresponding to different transform domains, and by means of the multi-domain feature fusion, realizing the transverse multi-domain feature fusion; obtaining a more easily marked single target signal, and improving the detection accuracy; and then, for the complex multi-source captured from different sensors, progressively performing data-level fusion, feature-level fusion, and decision-level fusion to realize the longitudinal feature fusion, and reducing the calculating complexity of a multi-target detection algorithm. A lot of signals are subjected to sparse decomposition compression, the unnecessary computational complexity can be reduced, the detection efficiency is improved, and the energy consumption of underwater target detection equipment can be reduced.

The esophageal cancer pathological image labeling method comprises the following steps: a) performing dyeing correction processing on an esophageal pathological image subjected to H & E dyeing; b) labeling expert canceration areas; c) mapping the canceration area outline marked by the expert into a pathological image of 40X; d) constructing an epithelial tissue contour detection model; d-1) marking whether pixel points belong to an epithelial region, an interstitial tissue or an irrelevant blank region; (d-2) constructing an end-to-end convolutional neural network model; and e) fusing the labeling areas. According to the method, the epithelial tissue contour is automatically drawn in the labeling process according to the characteristic that the esophageal cancer morbidity area occurs in the epithelial tissue basal layer area, so that the time cost of expert labeling is greatly saved. The method only aims at esophageal pathological section image modeling; only the edge of the epithelialtissue is detected, the model is relatively simple, operation is rapid, epithelial boundary detection has obvious advantages in detection precision, meanwhile, the method automatically learns effective features and expressions, the complex manual feature selection process is avoided, and the actual application requirements can be met.

The embodiment of the invention discloses an audio processing method and device, a terminal and a medium, and relates to the technical field of audios. The audio processing method comprises the following steps: acquiring audio file data to be labeled; outputting a corresponding audio waveform on a display interface according to the audio file data; segmenting the audio waveform according to user operation to obtain at least two waveform areas; and determining a labeling result of the audio file data according to the label information corresponding to each waveform area. According to the embodiment of the invention, audio processing is more intuitive, simpler and more efficient.

The invention discloses an automatic drawing method and an automatic drawing system for open pores in a hull drawing, equipment and a storage medium. The automatic drawing method comprises the steps of: presetting open pore information of target open pores in the hull drawing; and automatically drawing the target open pores in the ship drawing according to the open pore information. According to the automatic drawing method, the open pore information of the target open pores in the hull drawing is preset, and the target open pores are automatically drawn in the hull drawing according to the open pore information; the target open pores are grouped and sequenced according to the open pore types and the open pore sizes, the open pores are numbered, and then open pore numbers are automaticallymarked at the centroid of the target open pores; meanwhile, an open pore detail table corresponding to the hull drawing is automatically generated according to the open pore types, the open pore sizes, the open pore codes and the open pore numbers, manual intervention in the existing open pore drawing process is reduced, the drawing efficiency is improved, the open pore drawing effect and numbermarking are more standardized, and the overall drawing quality of the hull drawing is improved.

A method for quickly structuring voice databank used in key word detection task includes recording voice databank of isolated word, confirming key word table and nonkey word table according to requirement of key word detection system, confirming parameters such as total sentence number and time length as well as key word occurrence frequency for key word detection test, connecting key word and nonkey word in accordance with requirement of key word detection system to be sentence by pasting - up means for generating out databank used in key word detection task.

The invention discloses a pulse laser ranging echo moment resolving method and system as well as a terminal. The pulse laser ranging echo moment resolving method comprises the following steps: performing batch simulating calculation according to various parameters such as the target feature of a ranging scene, the parameters of a laser rangefinder, a target distance, a ranging angle, digitalization accuracy and noise level to generate theoretical pulse laser ranging full-waveform data; intercepting echo signal waveform data according to a specified width to serve as training and test sample data, and converting the resolving resolution into the classification tag value of the training and test sample data; and subtracting a main wave moment from an echo moment to obtain the ranging flighttime, and calculating the ranging value according to the flight time. Through adoption of the pulse laser ranging echo moment resolving method and system, the features of an echo waveform do not needto be extracted, the features can be automatically identified through deep learning to construct complex features, and echo moment calculation can be finished by using the features, so that the workload is very low compared with practical measuring notification.

The invention describes a device and a method for labelling containers through label cases. The label cases implement relative movements directed from sections at the bottom sides of the containers towards sections at the opening sides of the containers so as to occupy the related labelling positions on the containers to be labelled, thereby enabling the containers to be held in the air during the labelling process and thus realizing the specifically compact device for labelling.

The invention discloses an event recognition model optimization method, device and equipment and a readable storage medium, and the method comprises the steps: inputting training texts corresponding to various types of events into a character event information extraction module, and extracting and obtaining the associated information of each character in the training texts and the character eventsof the various types of events; inputting the character event associated information and the training text into an event classification module for classification to obtain an event classification result corresponding to the training text; inputting the character event associated information and the keyword label corresponding to the training text into a keyword identification module for identification to obtain keyword identification loss; and calculating classification loss based on the event classification result and the event type label corresponding to the training text, and optimizing keyword identification loss and classification loss to optimize an event identification model. According to the method, the event keyword information is added as event classification guidance, so that the problem of event recognition errors caused by insufficient understanding or directivity errors of the model to the events is avoided, and the event classification accuracy is improved.

The invention discloses a webpage tagging and community sharing system. The system is installed and executed in an intelligent computer device and is provided for a user to perform text description tagging on a webpage. The system comprises a webpage browsing component, a webpage segmentation component and a tagging component, wherein the webpage browsing component displays the webpage; the webpage segmentation component segments the webpage into two webpage panes separated in an up-down manner according to a tagging point provided by the user, and forms a tagging space between the two webpage panes; and the tagging component is used for tagging a text description in the tagging space. According to the system, the tagging space is formed in an up-down separation manner, so that the tagging is easy; and the system can be used in the intelligent computer device with a small-sized screen, thereby meeting demands in use.

The invention relates to an intelligent visual early warning system for early gas leakage. The intelligent visual early warning system comprises a detected area, a data collecting and processing module and a central control room, wherein the data collecting and processing module is responsible for real-time storage of monitoring videos and leakage detection and positioning, and the central controlroom is responsible for organic gas leakage display and early warning. The organic gas leakage detection positioning module adopted by the invention belongs to a deep learning hybrid model, and the trained module is embedded into an embedded information processor to realize automatic detection of organic gas leakage in a monitoring video. According to the invention, the models adopted by the system comprise an unsupervised self-encoding model and a supervised target recognition model, the unsupervised self-encoding model avoids the problem that a data set is difficult to collect and label insupervised model training, a training set required by the supervised model training only aims at a data set at a leakage source, the data set is easy to collect and label, the high detection rate andthe low false alarm rate of organic gas detection and positioning are guaranteed, and safety guarantee is provided for the field of petrochemical engineering.

The invention provides a data processing method and a device thereof, a service providing method and a device, equipment and a storage medium, and the method comprises the steps: obtaining a first image sample and an initial adjustment parameter, wherein the first image sample comprises a target object; and clustering the initial adjustment parameters, and determining a target adjustment parameter according to the degree of removing interference information in the first image sample by a clustering result; and eliminating interference information in the first image sample by using the target adjustment parameter to obtain a second image sample. In the process of determining the target adjustment parameters, the elimination degree of the clustering result to the interference information is considered, so that the determined target adjustment parameters have relatively high interference information elimination capability, and the target object can be clearly displayed in the second image sample. Due to the clear display of the target object, the difficulty of image sample labeling is reduced, the accuracy of image sample labeling is improved, and the training effect of the model is further ensured.

The embodiment of the invention relates to the technical field of deep learning, and discloses a model training method and device, electronic equipment and a computer readable storage medium. The model training method comprises the following steps: acquiring a face attitude angle coarse label of a training sample according to coordinates of key points pre-marked by the training sample, wherein the training sample is a two-dimensional face image, the key points comprise a left eye center, a right eye center, a nose tip, a left mouth corner and a right mouth corner, and the face attitude angle coarse label comprises a pitch angle and a yaw angle; obtaining joint probability distribution of face attitude angles of the training sample according to the face attitude angle coarse label and a preset candidate attitude angle set; and training a preset deep learning network according to the training sample, the joint probability distribution and a preset loss function to obtain a face attitude angle estimation model. According to the model training method provided by the invention, the calculated amount in the training process is very small, the time required for training is relatively short, and a stable and reliable face attitude angle estimation model can be quickly obtained.

The invention aims to provide a simple and practical geographic position labeling method. The method has the following advantages that: 1, global codes are unified; 2, the precision is high enough (the current GPS positioning precision can reach 4m level); 3, the number of bits is small (10-bit main encoding + 0-4-bit auxiliary encoding); and 4, the actual position is easy to determine (a simple formula can be directly converted into longitude and latitude). By using the coding system, a geographic position becomes simple to label, an actual position can be accurately positioned through an electronic map and GPS positioning only by providing a dozens of codes, and a simple and practical method is provided for determining the positions of manual express delivery and unmanned aerial vehicledelivery and labeling and positioning a certain area (such as fields, water areas, forests and building lands).

The invention discloses a muscle positioning garment made by utilizing a systematic anatomical map and a muscle positioning method, the special muscle positioning garment is arranged, a muscle pattern and a skeleton pattern are printed on the surface of the positioning garment and are matched with muscles and skeletons of a human body one by one in position and form, and the positioning garment is made of a special muscle positioning garment. And each muscle pattern and each skeleton pattern on the surface of the positioning clothes are marked with names. The muscle naming and position can be intuitively and stereoscopically popularized for the public, the accuracy and interestingness of muscle anatomy knowledge teaching are improved, various people with related learning requirements of muscle anatomy are helped to understand and memorize, and the accuracy of clinical diagnosis, household rehabilitation and health-care treatment muscle positioning is improved.

The invention relates to an early gas leakage intelligent visual early warning system, including a detected area, a data collection and processing module, and a central control room; the data collection and processing module is responsible for monitoring video real-time storage and leakage detection and positioning, and the central control room is responsible for organic gas leakage display and Early warning; the organic gas leakage detection and positioning module used in the present invention belongs to the deep learning hybrid model, and the trained module is embedded into the embedded information processor to realize the automatic detection of organic gas leakage in the monitoring video. The model used in the present invention includes an unsupervised self-encoder model and a supervised target recognition model. The former avoids the problem that supervised model training is difficult to collect and label data sets, and the training set required by the latter is only for the data set at the source of leakage. , this type of data set is easy to collect and label, which ensures a high detection rate and low false alarm rate for the detection and positioning of organic gases, and provides security for the petrochemical field.

The invention discloses a heart rate fusion labeling method and system based on Bayesian prior probability; reading, preprocessing and signal segmentation are performed on initial electrocardiosignals to obtain electrocardiograph data samples, and feature extraction is performed on the electrocardiograph data samples; according to different types of initial electrocardiosignal sets, a single-lead database or a multi-lead database, a proper heart rate labeling fusion process based on Bayesian prior probability is developed. In addition, probability estimation and iterative solution are carried out on the heart rate tags marked by multiple leads or multiple algorithms through a Bayesian criterion and an expectation maximization algorithm, and a heart rate tag value with higher precision is obtained through fusion. The fusion model automatically knows potential differences among different labeled samples; due to the fact that a label generated by a single algorithm or a lead is possibly unreliable, a label value with higher dependency is obtained through a fusion model, the accuracy of long-term dynamic heart rate estimation is improved, and more accurate information is provided for diagnosis of clinical cardiovascular diseases.

The invention provides an easily-applied ball surface point positioning method. The method includes the steps of firstly, determining a positioning point on a curved surface part; secondly, enabling the ball surface of a ball point locator to be tangent to the curved surface of the curved surface part to ensure that a tangent point P is the positioning point, wherein a ball handle of the ball point locator is located at the position A1 at the moment, and the axis of the ball handle of the ball point locator, the ball center of the ball surface and the tangent point P are collinear; thirdly, rotating the ball point locator under the conditions that the tangent point P on the curved surface part is unchanged in position and the ball surface of the ball point locator and the curved surface of the curved surface part still keep tangent so that the axis of the ball handle of the ball point locator can be rotated to the required direction, wherein the ball handle of the ball point locator can be located at the position A2 at the moment; fourthly, completing the positioning work. After the positioning method is adopted, the axis of the ball handle of the ball point locator can be rotated to the horizontal reference direction or the vertical reference direction as required, the positioning effect can not be affected, and difficulty of subsequent machining can be lowered.

The invention provides an easily-applied ball surface point positioning method. The method includes the steps of firstly, determining a positioning point on a curved surface part; secondly, enabling the ball surface of a ball point locator to be tangent to the curved surface of the curved surface part to ensure that a tangent point P is the positioning point, wherein a ball handle of the ball point locator is located at the position A1 at the moment, and the axis of the ball handle of the ball point locator, the ball center of the ball surface and the tangent point P are collinear; thirdly, rotating the ball point locator under the conditions that the tangent point P on the curved surface part is unchanged in position and the ball surface of the ball point locator and the curved surface of the curved surface part still keep tangent so that the axis of the ball handle of the ball point locator can be rotated to the required direction, wherein the ball handle of the ball point locator can be located at the position A2 at the moment; fourthly, completing the positioning work. After the positioning method is adopted, the axis of the ball handle of the ball point locator can be rotated to the horizontal reference direction or the vertical reference direction as required, the positioning effect can not be affected, and difficulty of subsequent machining can be lowered.

The embodiment of the present application discloses a data processing method, device, device, and storage medium in the field of artificial intelligence, wherein the method includes: obtaining the target text of the target bullet chat; The first recognition result of the curtain; the first bullet chatting recognition model is obtained based on the first training sample training including the first training text and its corresponding weak labeling result. It is determined whether the barrage playback function is turned off; through the second barrage recognition model, the second recognition result of the target barrage is determined according to the target text; the second barrage recognition model is based on the second training text and its corresponding strong annotation The results are obtained by training the second training sample; according to the first recognition result and the second recognition result, the target recognition result of the target barrage is determined. This method can achieve better bad barrage recognition effect and reduce the cost of model training.

The invention relates to a lane topology updating method and device, equipment and a storage medium. According to the invention, the production publisher publishes the lane topology update data, so that the application end can determine the target area corresponding to the lane topology change area in the existing electronic map, and the lane set composed of the lanes driving into the target area and the lanes driving out of the target area is obtained. And selecting at least two target lanes meeting a preset constraint condition from the lane set. And further, according to the lane topology change type, updating the topological relation of the at least two target lanes. As the lane topology update data is simple in structure, small in quantity, easy to mark and low in manufacturing cost, a production publisher can dynamically publish the lane topology update data in real time, so that an application end can immediately and quickly update the lane topology according to the lane topology update data, and the lane topology update efficiency is improved. Therefore, the new lane topology can be instantly and quickly applied by the application end.

The invention discloses a mobile phone sensor data annotation method based on weak supervised learning, which comprises the following steps of: 1, establishing an annotation task, generating a fuzzy query problem corresponding to the annotation task, and judging whether the answer of the fuzzy query problem is yes or no; 2, sending a request for answering the fuzzy query question to a collected person providing data, and collecting a fuzzy label obtained by answering the fuzzy query question by the collected person and mobile phone sensor data associated with the fuzzy label; 3, training a dichotomy depth model by using the fuzzy label acquired in the step 2 and the associated mobile phone sensor data; and 4, processing subsequently collected to-be-labeled mobile phone sensor data through the trained dichotomy depth model to deduce an accurate label of the to-be-labeled mobile phone sensor data, and labeling the to-be-labeled mobile phone sensor data by using the obtained accurate label. According to the method, fuzzy question query of the mobile phone sensor becomes easier, the manual process in the process is simplified, and accurate label data is obtained only through algorithm post-processing of fuzzy answers of a collector.

The embodiment of the invention discloses a data processing method and device, equipment and a storage medium in the field of artificial intelligence. The method comprises the steps: acquiring a target text of a target bullet screen; determining a first recognition result of the target bullet screen according to the target text through a first bullet screen recognition model, wherein the first bullet screen recognition model is obtained by training based on a first training sample comprising a first training text and a corresponding weak labeling result, and the weak labeling result is determined according to whether a bullet screen playing function is closed after a bullet screen to which the first training text belongs is played; determining a second recognition result of the target bullet screen according to the target text through a second bullet screen recognition model, wherein the second bullet screen recognition model is obtained by training based on a second training sample comprising a second training text and a corresponding strong labeling result; and determining a target identification result of the target bullet screen according to the first identification result and the second identification result. The method can achieve a good bad bullet screen identification effect, and reduce the model training cost.

A method for quickly structuring voice databank used in key word detection task includes recording voice databank of isolated word, confirming key word table and nonkey word table according to requirement of key word detection system, confirming parameters such as total sentence number and time length as well as key word occurrence frequency for key word detection test, connecting key word and nonkey word in accordance with requirement of key word detection system to be sentence by pasting - up means for generating out databank used in key word detection task.

The invention belongs to the technical field of video content processing, and particularly relates to a video image information labeling method. The method comprises the following steps: 1, establishing a picture information labeling database; 2, transmitting the real-time videos recorded on site to a labeling server in a short video form one by one, and identifying images in the short videos by utilizing an identification model to obtain a plurality of pieces of independent picture information to be labeled; and 3, comparing the independent picture information to be labeled with the picture information labeling database. The defects in the prior art are overcome; firstly, a picture information annotation database is established and perfected; then, in the process that the video is recorded on site; the video is firstly transmitted to the labeling server; and the labeling server is used for labeling the video and then transmitting the labeled video to the displayer. The adopted labeling method is simple and practical, labeling is simple, the processing speed is high, and when the processing speed is high to a certain degree, a displayed video only has a very small delay with the actual recording time.