190results about How to "Reduce recognition errors" patented technology

The invention discloses a reading method of a pointer type meter based on depth learning, belonging to the field of depth learning and computer vision. The method of the invention utilizes Mask- RCNNobject detection and instance segmentation algorithm to divide the dial and pointer images firstly, then correcting the dial by perspective transformation, Then using PCA (Principal Component Analysis) algorithm to fit the segmented instrument pointer, Then judging the direction of the pointer according to the center coordinate of the smallest oblique circumscribed rectangle of the pointer and thecenter coordinate of the dial; Finally, calculating the pointer reading according to the slope and direction of the pointer by using the angle method. As that method of the invention can accurately classify the type of the instrument, high precision pixel level segmentation of the pointer and dial, under non-uniform illumination conditions, different scales still have good robustness, to solve the traditional pointer instrument recognition field of the dial and pointer positioning difficulties, uneven light, mirror reflection, blurred pictures caused by low recognition accuracy.

The invention discloses a method for real-time monitoring of the face image quality of a customer collection terminal, comprising the following steps: Step 1, face image collection; Automatically detect and segment the exact position of the face in the face image; step 3 comprehensively evaluates the face quality, and performs intelligent prompts and real-time intelligent monitoring of the image quality through the face image quality detection system for the acquisition process of the client's face image. The invention can effectively improve the accuracy of subsequent image comparison and reduce the false alarm rate (False Alarm Rate), thereby ensuring the reliability of identification information and making the intelligent monitoring system based on face recognition truly practical.

The invention discloses a noise separation-based power transformer noise assessment method. The steps include: 1) building a voiceprint feature database in advance according to voiceprint feature information corresponding to noise signals generated by a fan component, an iron core component and a winding component when a standard power transformer is in different running states; 2) collecting in real time sound wave signals sent when a transformer to be tested runs, so as to obtain noise signals to be assessed; 3) separating noise signals generated by the fan component, the iron core component and the winding component from the noise signals to be assessed; and 4) extracting voiceprint feature information of each noise signal, identifying the running state of the transformer to be tested, and assessing the noise level of each component according to the running state. The noise separation-based power transformer noise assessment method provided by the invention can realize assessment of the noise level of each component in the power transformer utilizing acoustic characteristics, and has the advantages of easy operation implementation and high assessment precision and efficiency.

The invention relates to an amount recognition system for code scanning payment cash register of merchants, and an implementation method of the amount recognition system. The amount recognition system is characterized in that: original cash register software of merchants does not need to be modified, and code scanning payment cash register software is additionally installed to achieve support for code scanning payment; when the code scanning payment cash register software is installed for the first time, a screenshot region in the cash register software is set by using a button (1) marked with ''selecting amount appearing region'', the code scanning payment cash register software automatically performs screenshot on a settlement amount region of the cash register software according to a preset range when carrying out code scanning payment settlement each time, an opencv computer vision library is used for performing preprocessing on a picture to generate a binary image and save it, then a specific amount is recognized and amount input is automatically achieved through OCR (optical character recognition) according to a language package which is specially trained and corrected.

A speech recognition apparatus predicts, based on the occurrence cycle and duration time of impulse noise that occurs periodically, a segment in which impulse noise occurs, and executes speech recognition processing based on the feature components of the remaining frames excluding a feature component of a frame corresponding to the predicted segment, or the feature components extracted from frames created from sound data excluding a part corresponding to the predicted segment.

The invention relates to a TOF depth camera three-dimensional coordinate calibration device and method based on a virtual multi-cube standard target. The device comprises a three-dimensional motion horizontally-moving platform, a TOF depth camera, a cubical target and a background board. The calibration method includes the steps that three orthometric one-dimensional motion horizontally-moving platforms move repeatedly in the three-dimensional direction, the virtual multi-cube standard target of a complex shape and with multiple feature points is formed, the space position and the three-dimensional measured coordinates of a target angular point can be accurately obtained, and high-precision three-dimensional coordinate calibration of the TOF depth camera is achieved. By means of the TOF depth camera three-dimensional coordinate calibration device and method, target angular point feature recognition difficulty of the TOF depth camera and measurement errors each time are greatly reduced, three-dimensional measurement precision of the TOF depth camera is improved, the angular point position and the number of feature points of the virtual multi-cube standard target can be set flexibly, and full-process high-precision automatic calibration is achieved easily.

The invention discloses a hand gesture recognition method based on a switching Kalman filtering model. The hand gesture recognition method based on a switching Kalman filtering model comprises the steps that a hand gesture video database is established, and the hand gesture video database is pre-processed; image backgrounds of video frames are removed, and two hand regions and a face region are separated out based on a skin color model; morphological operation is conducted on the three areas, mass centers are calculated respectively, and the position vectors of the face and the two hands and the position vector between the two hands are obtained; an optical flow field is calculated, and the optical flow vectors of the mass centers of the two hands are obtained; a coding rule is defined, the two optical flow vectors and the three position vectors of each frame of image are coded, so that a hand gesture characteristic chain code library is obtained; an S-KFM graph model is established, wherein a characteristic chain code sequence serves as an observation signal of the S-KFM graph model, and a hand gesture posture meaning sequence serves as an output signal of the S-KFM graph model; optimal parameters are obtained by conducting learning with the characteristic chain code library as a training sample of the S-KFM; relevant steps are executed again for a hand gesture video to be recognized, so that a corresponding characteristic chain code is obtained, reasoning is conducted with the corresponding characteristic chain code serving as input of the S-KFM, and finally a hand gesture recognition result is obtained.

The invention discloses a gesture identification system based on multiple forearm bioelectric sensors. A local server of the gesture identification system stores gesture data which is completed in characteristic extraction and fusion and uploads the gesture data to a cloud server when the system is networked. The local server establishes a local gesture models according to the gesture data. The cloud server receives the gesture data uploaded by one or more local servers, establishes cloud gesture models and updates the gesture models in each local server with the cloud gesture models, so that the local end and the cloud end of the gesture identification system are respectively provided with a data set module, a classifier model module and an identification module, a user is enabled to carry out gesture identification even under an offline condition, and the gesture identification system is suitable for a moving scene whose network environment changes in real time; in addition, the local gesture models in the local server are updated by the cloud gesture models, so that the gesture identification of the gesture identification system is more precise.

The invention discloses a method for identifying human faces based on an HMM-SVM hybrid model, which comprises the following steps: firstly, sampling human face images from top to bottom by sampling windows; extracting characteristic parameters of each sampling window image by respectively adopting discrete cosine transform (DCT) and singular value decomposition (SVD), and serially connecting the characteristic parameters into one-dimensional observation vectors; then, using the observation vectors of the training images of each human body to train the HMM model of each human body; adopting the Viterbi algorithm to calculate the output probability of the observation vectors of all images corresponding to each HMM model; and using the output probability to support the classified training and the identification test of a vector machine. Because each HMM model has good time sequence modeling ability, the numerical characteristics of each organ of a human face can be effectively combined by a state transfer model to more integrally describe the human face to support the excellent performance of the vector machine in the aspect of classification of limited samples.

The invention discloses an iris classification method based on texture primitive statistical feature analysis. The method comprises: S1, preprocessing a clear iris image in a training set, obtaining an interested region ROI, extracting features of the ROI region, training the extracted texture features, building a model and obtaining an iris rough classification model; and S2, preprocessing the clear iris image arbitrarily input, obtaining the ROI region, extracting the features, inputting the extracted iris texture features into the model obtained through the training in the step S1 and obtaining the class information of the input iris image. The method has the advantages of shortening average time for completing iris comparison once, achieving real-time effects and effectively speeding up feature-template comparison in large-scale databases through an iris recognition technique.

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 provides a lightweight license plate detection and recognition method based on a multi-scale attention mechanism. Construction of a license plate detection and recognition network comprises the following steps: S1, obtaining pictures as an original data set; S2, processing the original data set to obtain a data set A used for training a model for detecting the license plate and a data set B used for training a model for recognizing the license plate; S3, constructing a deep neural network for detecting the license plate; S4, inputting the original image P1 of the data set A intothe network constructed in the S3 to obtain a license plate detection area P2 and four corners of the license plate; and S5, performing perspective transformation on the P2 according to the angular points of the license plate to obtain a corrected image P3. S6, constructing a deep neural network for recognizing the license plate; S7, inputting P3 into the network constructed in the S6 to obtain alicense plate number corresponding to the detected license plate. According to the invention, a relatively low network parameter quantity and a relatively low calculated quantity can be obtained simultaneously under the condition of ensuring the network accuracy.

The invention discloses a weighted local feature comparison based vehicle fake plate identification method and apparatus. The method comprises: based on a snapshot picture of a to-be-identified vehicle, identifying vehicle type information and a license plate of the to-be-identified vehicle, and extracting a local feature graph in the snapshot picture; then, extracting reference local feature graphs corresponding to vehicles with the same license plates in a reference comparison library, performing comparison one by one, and calculating the similarity between each local feature graph and the reference local feature graphs; and at last, according to a set local feature weight corresponding to the to-be-identified vehicle, calculating the total similarity between the to-be-identified vehicle and the vehicles with the same license plates in the reference comparison library by summarization, and according to the total similarity of the to-be-identified vehicle, judging whether the to-be-identified vehicle is a vehicle with a fake plate or not. The apparatus comprises a picture identification module, a local feature graph extraction module, a license plate search module, a comparison module, a total similarity calculation module and a pre-warning module. According to the method and the apparatus, the judgment and identification of suspected vehicles with fake plates are accelerated and the identification is more accurate.

The invention discloses a convolutional neural network-based music signal multi-instrument identification method. The method comprises the following steps of S1, extracting two features from an inputaudio, wherein the two features comprise a pitch feature matrix and a constant Q transformation matrix based on tone; S2, carrying out classification according to musical instrument groups, includingtubes, strings and percussion music, inputting the constant Q transformation matrix into a primary convolutional neural network to obtain a classification matrix, and inputting the classification matrix into a classifier to obtain a coarse classification result, namely the musical instrument group type; and S3, on the basis of the classification matrix, inputting the classification matrix into a secondary convolutional neural network with an attention network in combination with a pitch matrix to obtain a subdivision result, namely a specific musical instrument, wherein the attention network allocates weights to different harmonic waves. The method is suitable for musical instrument identification tasks in music information retrieval and can be used for the musical instrument identification method in music automatic transcription.

The invention relates to a position identifying system and a detection method of a multi-layer linear array laser spot. The position identifying system comprises a linear laser transmitter and a detection device, wherein the detection device comprises laser detectors, subsequent circuits, a single-chip machine, a display, an output interface and a power supply. According to the requirements of precision and length, a certain quantity of laser detectors are uniformly arranged and installed at certain intervals to form a linear array as one layer, multiple layers are made in parallel and then are used as one block, and multiple blocks are spliced to form the detection device. The linear laser transmitter forming a required certain position relation with the detection device transmits fan-shaped laser beams, the laser detectors induce the fan-shaped laser beams and generate electric signals which are processed and calculated by the subsequent circuits and the single-chip machine so that the laser detector in the center of the spot can be identified, and the position of the spot in the arrangement direction of the laser detector is obtained. The position information data can be directly displayed on the display and also be outputted by an output interface for other kinds of equipment.

The invention relates to a mechanical fault identification method for subspace embedding feature distribution alignment under different working conditions. The method comprises the following steps: firstly, aligning a source domain feature with a target domain feature in a target domain subspace by utilizing a related alignment method so as to prevent domain offset; then, directly predicting a pseudo label for a target domain in the spatial training base classifier, and quantitatively estimating respective weights of edge distribution and conditional distribution of two domains so as to adaptto the distribution difference between a source domain and the target domain; and finally, transmitting the learning rules of the two steps through a structural risk minimization framework, constructing a kernel function to establish a classifier, and performing iterative updating to obtain a coefficient matrix of a final framework to complete fault diagnosis. Quantitative estimation of respectiveweights of two-domain edge distribution and conditional distribution is of great significance in cross-domain mechanical fault diagnosis, and feasibility and effectiveness of the method are proved through multi-class composite fault diagnosis examples. The method is suitable for the fields of state monitoring, fault diagnosis and the like of mechanical equipment.

The invention discloses a fingerprint and iris integration identification device. The fingerprint and iris integration identification device mainly comprises a micro controller, an infrared range finding unit, an iris acquisition unit, a fingerprint acquisition unit, a light feedback unit, an infrared illumination unit, a live detection unit, an image preprocessing unit and an image identification unit. Characteristic matrixes of a fingerprint and an iris are integrated and then identified, no independent identification is carried out, and identification reliability is improved; according to live detection, a method which is easy and reliable and has no algorithm dependence is provided, so live detection can be carried out fast and effectively; according to a problem of inaccurate identification caused by easily-generated head rotation in an iris identification process, a solution scheme for head rotation prevention is provided; according to influence of natural light, a method for removing the natural light is developed. The device can realize integration identification employing the iris and the fingerprint, has higher reliability compared with independent application of the iris and the fingerprint, and solves problems of live detection, horizontal head rotation and natural light employing a human engineering principle, a human body physiologically principle and an optical principle, reduces identification errors and improves reliability.

The invention discloses a cloud processing based gesture identification method. The gesture identification method comprises the steps that: local servers store gesture data subjected to feature extraction and fusion, and upload the gesture data to a cloud server when a device is connected with a network; the local servers establish local gesture models according to the gesture data; and the cloud server receives one or more gesture data uploaded by the local servers, establishes cloud gesture models and updates the gesture model in each local server by utilizing the cloud gesture model. According to the gesture identification method, data set modules, classifier model modules and gesture identification modules are arranged at a local end and a cloud end respectively, so that a user still can perform gesture identification in a network disconnection state, and the method is suitable for a mobile scene in that a network environment changes in real time; and at the same time, the local gesture models in the local servers are updated with the cloud gesture models, so that the gesture identification method provided by the invention is more accurate.

The invention provides an image segmentation method and device, electronic equipment and a storage medium. The method comprises the following steps: carrying out down-sampling coding on a to-be-segmented image comprising a foreground and a background, and carrying out up-sampling decoding on an obtained down-sampling feature map to obtain an up-sampling feature map; carrying out classification processing on the up-sampling feature map to obtain classification probabilities of foregrounds and backgrounds corresponding to all pixels in the up-sampling feature map; performing compensation processing on errors of classification probabilities of foregrounds and backgrounds corresponding to the pixels to obtain a compensation feature map of the up-sampling feature map; performing fusion processing on the compensation feature map, the up-sampling feature map and the down-sampling feature map to obtain a fusion feature map; and identifying the foreground and the background from the to-be-segmented image based on the classification probability of the foreground and the background corresponding to each pixel in the fusion feature map. According to the invention, image segmentation can be accurately carried out.

The invention belongs to the field of machine vision and detection, and discloses a vision-based identification and locating method for outer beam holes in a multilayer beam. The method comprises the steps of (a) calculating a conversion coefficient for converting an actual size into a pixel size; (b) selecting a generation template in a to-be-identified outer beam hole of the multilayer beam, performing template matching on a to-be-identified image, obtaining a size type of the hole, intercepting a new local image by taking an optimal matching point as a center, and performing sub-pixel precision-level edge detection on the new image; and (c) fitting central coordinates of the hole by using edge point data, and performing multiplication by the conversion coefficient to obtain the central coordinates of the to-be-identified outer beam hole of the multilayer beam, thereby realizing the identification and locating of the hole. Through the method, automatic blind hole identification and locating of the outer beam holes in the multilayer beam are realized, the labor intensity of workers is reduced, the machining precision of the outer beam holes and the riveting quality of the multilayer beam are improved, and the production efficiency is improved by multiple times.

The invention discloses a particle classification statistic method for a scatter diagram, a particle classification statistic system for scatter diagram and a blood cell analyzer. The particle classification statistic method for the scatter diagram comprises the following steps of acquiring two-dimensional signals representing different features of particles, and generating a two-dimensional scatter diagram; pre-classifying the scatter diagram, and performing transformation, noise elimination and smoothing on the generated two-dimensional scatter diagram; classifying scatter diagram particles, and endowing each particle in the scatter diagram with a corresponding particle type; performing scatter diagram particle parameter counting, counting the number of the particles with the same particle type, calculating the percentage of each type of the particles in all the particles, and culling the particles with the interference type to realize classification statistics. According to the particle classification statistic method, features of normal samples can be met, and features of special samples can also be met; therefore, the identification error on the special samples is reduced.

The invention discloses a farmland disease monitoring system based on machine vision, and the system comprises a target image collection module and a monitoring terminal, and the interior of the monitoring terminal is provided with a disease and pest recognition module which achieves the recognition of holes, spots, pests and pest tracks in a picture based on a neural network model; the neural network model which adopts an ssd target detection algorithm, and an inception v2 deep neural network is trained by using a coco data set; and a disease and insect pest statistical analysis module whichis used for communicating the component external graphic template and the measuring scale to measure holes and spots on the picture, and realizing statistical analysis of disease and insect pests according to an recognition result of the disease and insect pest reognition module. According to the method, holes, spots and injurious insects on leaf surfaces and rhizomes of the crops are quickly recognized by adopting an inception v2 deep neural network, so that the current pest and disease damage conditions of the crops can be accurately obtained.

The invention provides an automatic foreign name identification and control method based on context semantics in a natural language processing system by researching foreign name characteristics and combining a statistic probability model. The method is characterized by comprising the following steps of: a. analyzing a text to be identified and acquiring a candidate foreign name string set; b. correcting and screening the candidate foreign name string set by utilizing a foreign name rule set to acquire a first middle foreign name string set; c. further screening the first middle foreign name string set by utilizing probability statistics and the probability model to further screen the acquired identified foreign name set; and d. determining the unidentified foreign names according to the identified foreign name set. According to the system, the context characteristics of the names and the word characteristics of the foreign names are fully utilized, the identification error caused by word segmentation is greatly reduced, the condition that the other named entities are identified into names is well avoided, and the identification effect is improved.

The invention relates to an odor recognition method and device based on an electronic nose and an electronic nose system. The odor recognition method based on the electronic nose comprises the following steps: obtaining detection data of a plurality of odor samples collected by the electronic nose system; decomposing the detection data of each odor sample, so as to obtain an adsorption mass matrixof each odor sample, an adsorption amount scaling matrix of odor molecules of each odor sample on each sensor in the electronic nose system and an odor molecule characteristic matrix; constructing anabstract odor factor diagram of each odor sample separately according to the adsorption mass matrix, adsorption amount scaling matrix and odor molecule characteristic matrix of each odor sample; calculating similarity between abstract odor factor diagrams of two odor samples; and recognizing whether two odor samples are the same odor sample or not according to the similarity between abstract odorfactor diagrams of two odor samples. The odor recognition method based on the electronic nose has the advantages of simple pretreatment, short detection period and low detection cost, and can realizehigh-accuracy identification of different odor samples with high similarity and complexity.

The invention belongs to the technical field of computer software, and discloses an intelligent voice axis cutting method, an information data processing terminal and a computer program. The intelligent voice axis cutting method comprises the following steps: pre-training, namely, carrying out initialization on model parameters through a non-supervision learning algorithm by using abundant data not labelled; and model fine-tuning, namely, learning the model parameters by adopting the traditional neural network learning algorithm and using a small amount of labelled data. According to the technical scheme, effective voice segments are obtained through the windowing framing technology, continuous and stable voice signals are obtained, and the identification errors are reduced; voice signalsare effectively enhanced, the capacity of distinguishing the useless signals is enhanced, the noise interference is eliminated, the errors are reduced, and the voice recognition accuracy rate can be improved by 50%; the problem of background noises can be effectively solved, and the voice recognition accuracy rate is improved to 93%. For acoustic feature extraction, voice feature vector sequencescan be extracted according to voice features similar to that of human beings, background noises and channel distortion are eliminated, and the voice recognition accuracy rate is improved to 94.7%.

The invention requests to protect a visual attention detection method based on an improved mixed increment dynamic Bayesian network. According to the method, a head, a sight line and a prediction sub-model are fused to carry out comprehensive estimation on the visual attention detection method; the sight line detection sub-model is improved on the basis of a traditional human eye model, so that the recognition rate is increased, and the detection robustness of different testers is improved; in order to solve the problem of data loss caused by extreme postures and dynamic scenes, a prediction sub-model is provided, and the correlation of sampling pictures at two moments is measured by utilizing a Gaussian covariance, so that the false recognition at the current moment is effectively improved, and the recognition error is reduced. Secondly, describing the related sub-models, and respectively establishing Bayesian regression models by utilizing conditional probabilities; And parameters ofthe model are dynamically updated by using an incremental learning method, so that the adaptability of the whole model to new input data is improved.

Embodiments of the present application provide a crop pest identification method and device. The method includes: acquiring an image to be identified; a preset recognition model is adopted to respectively determine the credibility of the crops to be recognized and the pests to be recognized in the images to be recognized; calculating an initial probability value of the image to be identified basedon the credibility of the crop to be identified and the pest to be identified; adjusting the initial probability value to obtain a target probability value of the image to be recognized; according tothe target probability value, identifying the crop to be identified and the pest to be identified, Thus, the pesticides can be sprayed on crops pertinently to remove pests on the crops, the efficiency and accuracy of determining crops and pests through the identification image are improved, the identification error caused by the manual identification of the image is reduced, and the follow-up operation is ensured to be carried out smoothly.

The invention relates to a pig face recognition method, which belongs to the technical field of artificial intelligence. The method comprises the following steps: S1, multiple pigsties are arranged side by side, wherein each pigsty comprises a pigsty entrance arranged oppositely, and the pigsty is internally provided with a camera facing the pigsty entrance; S2, the cameras carry out shooting on pig faces entering the pigsties, and pig face video source data are acquired synchronously; S3, the video source data are uploaded to a server and processed to pig face images; S4, the pig face imagesare screened and labeled, effective pig face images are kept, and the coordinate information of the labeled pig face images is used as image data sources; S5, a model for pig face recognition is trained according to the image data sources; and S6, according to the well-trained pig face recognition model, the acquired pig face video source data are processed automatically. Through adopting a Tensorflow system for pig face recognition, the problem of lacking a technology for pig face recognition currently is solved.

The invention discloses an automatic identification device and method for bar codes of express items. The device comprises a mechanical arm, a first identification unit, a judgement unit and a secondidentification unit. The automatic identification method applied to the device comprises the steps that the mechanical arm grabs the express items and exposes first surfaces of the express items; thefirst surfaces are subjected to first bar code identification; with the first surfaces as the bottom surfaces of the express items, the express items are placed on a conveying belt through the mechanical arm; whether or not first bar code identification is completed is judged, and if not, the non-bottom surfaces of the express items are subjected to second bar code identification through multi-surface bar code scanning equipment arranged above the conveying belt. According to the automatic identification device, the bottom surfaces and the non-bottom surfaces of the express items are distinguished, based on a two-step identification process, the problem is solved that the bottom surfaces of the express items on the conveying belt cannot be quickly and conveniently identified, the device iseasily applied to existing automatic identification equipment and is low in transformation cost, high in identification speed and high in automatic degree, and the automatic identification process for the bar codes of the express items is optimized.

An intelligent dry burning preventive method comprises the steps that a state image of a cooker in the cooking process is collected; the state image is identified to obtain attribute information related to the cooking process, and a database is searched for dry burning critical state information matching the attribute information according to the identified attribute information; temperature information of the cooker in the cooking process is obtained; and whether the cooker is subjected to dry burning or not is determined according to the temperature information and the searched dry burning critical state information. The dry burning state of the cooker is determined by image identification and temperature detection, an artificial intelligence method is used to classify and identify the state image of the cooker in the cooking process, the corresponding dry burning critical state is used according to the identification result to monitor the cooking state in real time, the dry burningstates of different cookers can be determined more precisely, and an identification error is effectively reduced. The development cost of the whole system is low, the system can be conveniently assembled with other household cookers, and use is more convenient.