85results about How to "High recognition reliability" patented technology

Various processes are disclosed for conducting database searches by telephone. A user is initially prompted to utter N characters of a search query, and/or to enter these characters on a telephone keypad. Based on the user's entry, a speech recognition grammar is either selected or is generated dynamically for processing the user's utterance of the complete search query. In one embodiment, the speech recognition grammar is selected based on a sequence of N telephone keys selected by the user, without requiring the user to uniquely specify the N characters to which these keys correspond. In another embodiment, the speech recognition grammar is selected based solely on utterances of the N characters by the user such that the correct grammar is selected even if an utterance of one character is misidentified as an utterance of a similar sounding character. Also disclosed are methods for generating speech recognition grammars from query logs.

A lane robust identification method for warning departure is that first conducts conditional edge detection on the left and right local windows, and constructs the corresponding gradient direction map. Then, according to the histogram processing method, the gradient direction range of lane edge points on every local window. Aiming at the continuity features of direction and space of lane edge points, the filtering algorithm of the candidate and effect point collection is put forward and every window is filtered. Finally, according to the established effective point collection corresponding to each window, the rapid and accurate extraction of lane symbol is realized though the Kalman filter method based on scalar processing.

A system and associated methods are disclosed for improving voice recognition accuracy when a user conducts a search by voice. One method involves prompting the user to enter a set of characters of the query (e.g., the first N letters of a query term), and then using these letters to execute a preliminary search. The results of the preliminary search are then used to generate a dynamic grammar for interpreting the full voice query. The grammar may alternatively be retrieved from a cache or other memory that stores the grammars for various combinations of letters. In one embodiment, the user enters the characters by selecting the corresponding keys on a standard telephone keypad (one depression per letter) and then saying the letters, and the keypad entries are used to reduce the number of possible interpretations of each character utterance. Another method, which is useful for search refinement, involves generating a dynamic grammar from a set of search results (e.g., when the number of hits is large), and then using this grammar to interpret utterances of additional query terms to be added to the query.

Two-dimensional image information and three-dimensional image information of a subject are acquired, facial recognition is performed using the two-dimensional image information to determine whether a recognized face is a registered user's face, an elliptical model of the user is matched to the three-dimensional image information to calculate an error if it is determined that the recognized face is the user's face, and it is determined whether the user's face is improperly used based on the error. The subject's face is determined using the two-dimensional image information and the three-dimensional image information of the subject and it is determined whether the recognized face is improperly used, thereby improving facial recognition reliability. Thus, information security is improved.

Column misalignment between a character string recognized by a MICR and a character string recognized by optical character recognition is corrected to improve character recognition reliability. A magnetic data reading unit detects the magnetism of a character string printed in magnetic ink and outputs the detected magnetic data. An image reading unit optically reads the same character string printed in magnetic ink and outputs the image data. A magnetic data recognition unit applies a character recognition process to the magnetic data output by the magnetic data reading unit and outputs magnetic data recognition result. An image data recognition unit applies character recognition to the image data output by the image reading unit and outputs image data recognition result. A recognition result comparing unit applies a column offset detection process to detect misalignment between the character columns in the magnetic data recognition result and image data recognition result, then applies a column offset correction process to correct the column offset and align corresponding character columns, and then compares corresponding character columns in the magnetic data recognition result and image data recognition result based on the character strings after correcting the column offset and outputs a final recognition result.

The invention discloses a positioning and tracking system of an underground/roadway moving object, comprising underground positioning and tracking systems distributed under a detected mine and ground monitoring systems distributed on the ground; two-way communication is conducted between the underground positioning and tracking system and the ground monitoring system; the underground positioning and tracking system comprises a plurality of wireless radio-frequency labels, a plurality of wireless signal receiving and transmitting devices and a plurality of card readers which are respectively carried by underground workers, are internally stored with the information of the identity and engaged job category information of a cardholder, can read the stored information in the wireless radio-frequency labels within respective signal coverage area in real time, and synchronously upload the read information and the internally-stored information to the card reader of the receiving and transmitting device; and the wireless radio-frequency labels are distributed on a coal mining work surface or a coal mining roadway. The positioning and tracking system has reasonable design, convenient installation and distribution, low cost, complete function and good use effect, and also improves the coverage area correspondingly while saving the cost.

The invention relates to a face recognition method based on adaptive weighting and local characteristic fusion. The method comprises the main steps of firstly performing blocking on an image, calculating information entropy of each subblock and obtaining a weighting coefficient of the subblock, then extracting a texture characteristic of the image in a local binary pattern, extracting edge information of the image by means of a histogram of gradient, performing adaptively weighted partial characteristic fusion on a sample image, and finally performing classified recognition by means of a nearest neighbor method. According to the face recognition method, robustness of the local texture characteristic and the edge characteristic of the image to image local shadow and illumination change is utilized; a finally output characteristic has relatively high interference resistance and relatively high discriminating capability; and a good recognition effect is realized on the condition of illumination and gesture change.

The invention discloses a handwritten Chinese character recognition method based on substructure learning. The handwritten Chinese character recognition method based on substructure learning comprises the following steps of taking a Chinese character segmented fragment as a substructure of a Chinese character, extracting a Chinese character substructure mode from a Chinese character segmented fragment sample, bringing the Chinese character substructure mode into training of a Chinese character classifier, and finally realizing recognition of a handwritten Chinese character string through the combination of substructure recognition information and Chinese character substructure constitution information. The handwritten Chinese character recognition method is based on the characteristic that each Chinese character is composed of one or more substructures. Due to the facts that the Chinese character substructures are extracted, and the Chinese character substructures and the individual Chinese character are simultaneously trained in the Chinese character classifier, the reliability of recognition of the Chinese character segmented fragment is effectively improved in the process of recognition of the handwritten Chinese character string, errors, caused by unreliable recognition of the Chinese character segmented fragment, of recognition of the handwritten Chinese character string are reduced, and the precision of handwritten Chinese character recognition is improved.

The invention relates to the field of classification models of artificial intelligence, and provides a domain adaptive model training method and device, an image detection method and device, equipmentand a medium. The method comprises the steps: acquiring an image sample set comprising a plurality of image samples, and inputting the image samples into a Faster-RCNN-based domain adaptive model containing the initial parameters; performing image conversion on the image sample through a preprocessing model to obtain a preprocessed image sample; obtaining a feature vector graph through a featureextraction model; obtaining a region feature vector graph through a region extraction model; obtaining a local feature alignment loss value through the local feature model; performing regularization and global feature recognition processing through a global feature model to obtain a feature regularization loss value and a global feature alignment loss value; obtaining a detection loss value through the detection model; obtaining a total loss value; and iteratively updating the initial parameters until convergence to obtain a trained domain adaptive model. According to the invention, the cross-domain image recognition is realized, and the accuracy and reliability of image recognition are improved.

The invention provides a method for identifying a mountain front deep part construction layer through utilizing gravity interface inversion and is a gravity density interface inversion method for difficulty in depicting deep part stratum structure attributes during mountain front oil gas target exploration. The method comprises steps that gravity potential field data pre-processing is carried outto form basic data before data inversion; different target area lithology and stratum physical property rules are analyzed, and physical property base of density interface inversion is made to be clear; a variable density physical property model is established, and an inversion physical property model is provided; gravity-earthquake combined step by step fine delamination is carried out, and superposition field source separation precision is enhanced; drilling well-earthquake constraints are added to the space domain, and interface constraints are provided by skeleton profile combined inversion; gravity density interface three-dimensional inversion is developed, and the geological structure attributes of the mountain front deep part construction layer is made to be clear. The method is advantaged in that distribution characteristics of the mountain front deep part construction layer is effectively depicted, geological basis is provided for mountain front deep oil gas exploration, andoil gas exploration risks are 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.

This invention relates to a method for operating a filter-tip rod machine and a filter-tip rod machine, wherein the filtering material is extracted from a storage, spread, extended and transmitted through the coating device, then provided to the forming apparatus, furthermore, a filtering material strip is formed to be a filter-tip rod. It is characterized in that at least one sensor identifies the default part and produces at least one signal after the extension of the filtering material strip and before the coating device, therefore, at least the area of the filter-tip rod with default parts will be excluded by the next process; the filter-tip rod has a filtering material output station, a spreading apparatus, a extending apparatus a coating device and a rod forming apparatus which helps making the filter-tip rod, wherein the apparatus comprises at least one sensor for identifying the default part of the filtering material strip, being characterized in that the sensor is arranged at the position of the extending apparatus or arranged at the lower part of the extending apparatus and the upper part of the coating device.

Methods and devices for insulation monitoring of an ungrounded IT power supply system having at least two phase conductors includes determining an insulation resistance separately for each phase conductor using a separate response value. In one embodiment of the invention, relevant current and voltage distributions are calculated. In another embodiment, a change time window is set within which a second response value is activated. In yet another embodiment, an option of shutting down/continuing operation of the IT power supply system is offered.

According to the invention, the electric motor is electrically commutated with the aid of circuitry, in which the phase current experiences a zero crossing at certain time points per motor phase. Owing to the inductive load portion, the time of said zero crossing of a phase current occurs at different times to the time of the zero crossing that would arise with purely ohmic loads. Without a faulty load condition, the time of said zero crossing is within an expected value range (e.g., expected time window) which can be determined by the circuitry, the ambient conditions and by diverse motor parameters. According to the method, during occurrence of the high-side and/or low-side phase connection predetermined in the circuitry, it is determined whether and when the current through the switched-on high-side or low-side switch becomes greater or smaller than a predeterminable threshold, in a particular case in the vicinity of the zero crossing, wherein said time measurement can extend in the case of a PWM control, if required, over one or more PWM cycles.

The invention provides a bar code recognition method and device. Information identification is carried out on the bar code; if the result information corresponding to the bar code is not successfullyacquired, the bar code is deleted; the method can be based on the image recognition result of the bar code. According to the method, the character area corresponding to the bar code is determined according to the identification result of the bar code, and the character information contained in the character area is extracted, so that the result information corresponding to the bar code is obtainedthrough the character information, the problem that the required result information cannot be directly obtained through the bar code under the condition that the bar code is partially damaged or cannot be identified is solved, and the bar code identification reliability is improved.

The invention discloses an embedded automatic ruler reading device and method based on visual sensing. The embedded automatic ruler reading device comprises an image collection unit, a result displayunit and an image processing unit, the image collection unit comprises an LED illumination light source and a digital camera. The result display unit comprises an LCD display screen. The image processing unit comprises an embedded main controller and an upper computer; wherein the embedded main controller receives information acquired by the camera and displays an image of a ruler to be identifiedand a reading result of the ruler to be identified on the LCD display screen, and the embedded main controller communicates with the upper computer to realize remote transmission of the reading result of the ruler to be identified. The ruler digital character can be segmented, positioned and recognized, the scale of the ruler in the center of the field of view can be accurately read, and the current ruler image and the reading in the center of the field of view are displayed on the embedded liquid crystal screen, so that human eyes are replaced for automatic measurement, gross errors generated by manual reading are eliminated, Meanwhile, the reading efficiency and precision are improved.

The invention provides a tumor recognition system based on artificial intelligence. The tumor recognition system comprises an image acquisition module, an image library, an image preprocessing module,an image feature extraction module and a tumor recognition module. The tumor recognition method comprises the steps that S1, noise reduction processing is conducted on a CT image of a human organ tobe detected, and then an image feature extraction module extracts image texture features; S2, a focus recognition BP neural network recognizes whether the human organ CT image contains a suspected focus area according to the image texture features; S3, an image preprocessing module performs image segmentation on the human organ CT image containing the suspected lesion area to obtain an area wherea suspected lesion is located; S4, after the region where the suspected focus is located is enhanced, texture features and shape features are further extracted; and S5, the focus type is identified bythe tumor identification BP neural network, and finally benign and malignant identification of the tumor is realized. Compared with the prior art, the method has the advantages of high recognition rate and high recognition accuracy.

The invention, which belongs to the wireless communication field, discloses a continuous-detection-mechanism-based anti-collision algorithm for an EPC Gen2 protocol. The algorithm is characterized in that according to the number of response tags, a reader determines whether a conflict occurs between channels, channels are idle, or tags are read successfully and different processing measures are taken for the three kinds of situations; when a conflict occurs between channels or channels are idle for the first time, a parameter Q of a protocol is not changed and a next communication cycle is carried out continuously; if conflicts occur continuously or channels are idle, the Q is updated respectively by using different updated step lengths; and otherwise, the Q remains unchanged. An instruction sending condition is set; before the reader sends an instruction, whether the Q value is changed is determined; if the Q value changes, the reader sends a query adjust instruction; and otherwise, a QueryRep instruction is sent. According to the algorithm, the performance of the algorithm is not affected by the tag number; and the tag identification rate can be increased obviously.

The invention discloses a method for rapidly reading express bar code information, comprising: reading a bar code or a two-dimension code on a package by a code reader, through the read bar code or the read two-dimension code, searching the goods receiving address and name corresponding to the bar code or the two-dimension code in a central database for storing the bar code or the two-dimension code, converting the identified address and name into digital codes according to the codes corresponding to the goods receiving address edited in advance, displaying the digitals on a display screen, simultaneously broadcasting the digitals by an audio system. The invention is featured by rapid identification speed, good identification reliability, reducing artificial workload, improving the working efficiency and effectively solving the deficiency in the prior art.