27results about How to "Solve the low recognition accuracy" patented technology

The invention relates to the technical field of classroom behavior recognition, in particular to a classroom behavior recognition method and device based on multi-scale expansion residual attention perception, and the method specifically comprises the following steps: collecting classroom behavior images to construct a data set, carrying out the labeling and data enhancement of the data set, and then dividing a training set and a verification set; constructing a # imgabs0 # backbone network, and inputting the image data in the training set # imgabs1 # into a feature # imgabs3 # in the # imgabs2 # backbone network; constructing a # imgabs4 # mixed attention module, and inputting the obtained feature # imgabs5 # into a # imgabs6 # module to obtain a feature # imgabs7 # with attention perception; inputting the feature # imgabs8 # into the # imgabs9 # part to obtain a feature pattern # imgabs10 #, a feature pattern # imgabs11 # and a feature pattern # imgabs12 #; and inputting the feature maps # imgabs13 #, # imgabs14 # and # imgabs15 # into a plurality of decoupling prediction head CBPH modules of the yolov5 for multi-scale prediction, finally positioning the position of the student in the classroom image, and identifying the class behavior category of the student. According to the method, the problem of low recognition precision caused by inconsistent target scales in a classroom scene can be solved, and discriminative features of classroom behaviors can be extracted more effectively.

The invention relates to the technical field of auxiliary paper marking, and particularly provides an intelligent paper marking method based on a large language model and optical character recognition. The method comprises the steps that test paper and answers are processed through a large language model, test question analysis texts are obtained, and the test question analysis texts comprise analysis texts of the answers; processing the test paper image needing to be marked through an optical character recognition technology to obtain a corresponding answer text; and through comparison of the test question analysis text and the answer text, score data of each question is determined, and the score data comprises scores and comments. According to the intelligent paper marking method provided by the invention, the problem of low paper marking accuracy of an intelligent paper marking method in related technologies can be solved.

The invention relates to an intelligent settlement machine, an intelligent settlement machine dinner plate recognition method and system and a medium, and the method comprises the steps: detecting a dinner plate contour and a dish contour in a dinner plate image through a convolutional neural network model, and generating a mask image; and identifying color information of the mask image, and filling the color features into the contour of the dish to generate an original dinner plate image without the dish. And performing feature extraction on the original dinner plate image through the convolutional neural network model to obtain a plurality of feature vectors of the target dinner plate. And comparing the plurality of feature vectors of the target dinner plate with dinner plate feature vectors in a database, and determining that the category of the target dinner plate is a corresponding dinner plate category in the database based on a comparison result. Accurate confirmation of the types of the dinner plates is achieved, and then the names, prices and total prices of the dishes are determined. The problem that in the prior art, intelligent settlement machine dinner plate recognition is low in recognition accuracy is solved.

The invention discloses a LightGBM-based tobacco leaf curing state identification method, which comprises the steps of S1, acquiring curing data in a tobacco leaf curing process, the curing data comprising tobacco leaf images, temperature and humidity and tobacco leaf weight change data; s2, performing feature extraction and derivation processing on the baking data to obtain heterogeneous multi-dimensional feature data; s3, preprocessing the heterogeneous multi-dimensional feature data to obtain sample data; s4, after similarity and importance processing is conducted on the sample data, the sample data are input into a corresponding LightGBM model, a result output by the model serves as input data of a second-layer clustering model, and an output result of the second-layer model serves as the baking state of the tobacco leaves; the method not only can effectively solve the problem of low recognition accuracy caused by fewer data categories of the input model, but also can solve the problem of low model generalization caused by insufficient data utilization.

The invention discloses a phase-household relation identification method and device of a low-voltage power distribution network and electronic equipment. The low-voltage distribution network topology identification method comprises the steps of collecting node current of each node and phase line current of each phase line on a target low-voltage distribution network; determining an initial weight of the node current, calculating an initial regression coefficient and an initial adjustment residual value between the node current and the phase line current, updating the initial weight, calculating the regression coefficient and the adjustment residual value based on the updated weight, and performing iterative calculation for multiple times until a difference value of the regression coefficients of two adjacent times of iterative calculation is smaller than a preset threshold value. Determining the regression coefficient of the last iteration as a target regression coefficient; and determining a target regression coefficient between the target node and each phase line, determining the phase line corresponding to the maximum target regression coefficient as the target phase line to which the target node belongs, and obtaining a user set of each phase line. According to the invention, the problem of low phase-household relation identification accuracy of the low-voltage power distribution network in the prior art is solved.

The embodiment of the invention provides a fingerprint anti-counterfeiting neural network training method, a fingerprint anti-counterfeiting method and a fingerprint anti-counterfeiting device, and the method comprises the steps: obtaining a plurality of groups of training data, each group of training data comprising first native domain data, second native domain data and third native domain data; multiple groups of training data are used to train the initial classification network to obtain a target classification network, and the initial classification network comprises a fusion sub-network and a classification sub-network; the fusion sub-network is used for generating a first fingerprint matching pair based on the feature description matrixes of the first native domain data and the second native domain data, and generating a second fingerprint matching pair based on the feature description matrixes of the first native domain data and the third native domain data; and the classification sub-network is used for performing fingerprint classification identification based on the first fingerprint matching pair and the second fingerprint matching pair. According to the fingerprint identification method and device, the problem that the identification accuracy of true and false fingerprints is low in the prior art is solved, and the effect of improving the identification accuracy of the true and false fingerprints is achieved.

The invention discloses a power equipment nameplate identification method and device and a storage medium. The method comprises the following steps: acquiring at least one original image containing a target power equipment nameplate; the original image is input into an image denoising model, a target to-be-recognized image is determined based on a model output result, and the image denoising model is a generative model obtained by training a generative adversarial network based on a sample image containing a power equipment nameplate and an expected denoising image corresponding to the sample image; and determining a target nameplate area in the target to-be-identified image based on an edge detection method and a contour search method, and extracting equipment nameplate data in the target nameplate area. And the identification efficiency and accuracy of the equipment nameplate data are improved.

The invention relates to the technical field of image character processing, in particular to a PaddleOCR-based complex background image character detection and enhancement system and method, and the method comprises the steps: reading image files in a specified directory in batches, carrying out the preliminary character detection and recognition through a PaddleOCR model, extracting a character bounding box and the corresponding character content, and carrying out the recognition of the character bounding box; and calculating the central position of each character. Based on the preliminary recognition result, the system further analyzes character features, including an average character width, an average character height, and a character pitch, to speculate character region locations that may exist but are not preliminarily recognized in the image. And carrying out image enhancement processing, contrast enhancement and histogram equalization on the predicted potential character region by the system, carrying out character recognition by applying the OCR model again, and continuously supplementing a new recognition result. The method significantly improves the recognition rate and processing effect of OCR, and is suitable for component recognition of complex backgrounds, commodity packages, food detection mechanisms and the like.

The invention relates to a low-slow small target identification method based on radar and visual data fusion, and the method comprises the steps: S1, setting a radar system for collecting radar data, and a camera system for collecting visual data; s2, synchronizing data acquisition of a radar system and a camera system, and respectively obtaining corresponding radar point cloud data and image data; s3, after radar point cloud data obtained by the radar system and image data of the camera system are preprocessed, data fusion is carried out to obtain fused data; and S4, inputting the fusion data into a pre-trained neural network model, and carrying out detection and classification of low-slow small targets.

The invention provides a tray state detection method for a tilting tray type sorting machine. The tray state detection method comprises the following steps that firstly, image information of a tray area is collected through a visual system; 2, analyzing the acquired image information by using a deep learning technology and an image processing technology, identifying a tilting-pan type trolley, a package and a side plate, obtaining an external rectangular frame, obtaining a contour mask image of each target detection object, and extracting contour points of the contour mask image; step 3, carrying out existence detection and double-piece and stacked-piece detection; 4, establishing a coordinate system of the tray trolley and a coordinate system of the running direction, and performing torsion detection on the tray trolley; step 5, performing super-region detection; sixthly, side plate blocking detection is carried out; according to the technical scheme, the tilting-pan type trolley, the parcels and the side plates are recognized, and the external rectangular frame and the contour mask image are obtained, so that detection is more comprehensive, the problem that a traditional detection device is low in recognition accuracy is solved, and the sorting accuracy and efficiency are improved.

The invention belongs to the field of computer vision, and particularly relates to a small target detection method and system based on feature fusion, and the method comprises the steps: inputting an image into a backbone network for feature extraction, and obtaining a multi-scale feature map; inputting the multi-scale feature map into CIAM for feature information integration to obtain feature information containing a target context; connecting output features of the 37th layer and the 50th layer in the backbone network to a fusion network in a cross-layer manner to obtain a fusion feature map combined with the original feature information of the target; and inputting the fused feature map into a detection network to obtain a detection result. According to the method, the CIAM is integrated into the yolov7 feature extraction network to integrate the target features and related background information in the multi-scale feature map, and meanwhile, the cross-layer feature fusion network is used to reserve the original features of the target, so that the problem of low recognition precision of the small target caused by lack of feature information and sensitive positioning error is solved.

The embodiment of the invention provides disposable underpants and a subcutaneous tissue ultrasonic recognition method and system for the underpants, and relates to the technical field of subcutaneous tissue recognition technologies. The method comprises the following steps: acquiring ultrasonic echo information; performing image conversion processing on the ultrasonic echo information to obtain an echo image; and identifying the echo image through a first model to obtain subcutaneous tissue information. According to the invention, the problem of low subcutaneous tissue recognition precision is solved, and the effect of improving the subcutaneous tissue recognition precision is achieved.

The invention discloses a dinner plate recognition method and device, equipment, a storage medium and a computer program product, and relates to the technical field of computer technologies, and the method comprises the steps: obtaining a dinner plate picture and a food picture, and the dinner plate picture comprises dinner plates in different stacked and / or hand shielding states; according to the dinner plate pictures and the food pictures, generating sample pictures of different foods placed in the dinner plates in different states; and training the sample pictures to obtain a dinner plate recognition model, and performing dinner plate recognition by using the dinner plate recognition model. The identification accuracy of the dinner plate identification model is improved by generating the difficult sample data.

This application belongs to the field of character recognition, specifically relating to character recognition methods, devices, and storage media. The method includes: acquiring image data, which includes text information to be recognized; acquiring a pre-trained character recognition model, which includes a main branch and sub-branches, the sub-branch being used to acquire semantic information from the image data; the main branch being used to recognize the text information from the image data using the semantic information; and inputting the image data into the character recognition model to obtain the recognized text information. This method can solve the problem of low accuracy in traditional character recognition methods. Because the character recognition model can fully learn the semantic information of the text during the recognition process, it can improve the model's recognition accuracy, thereby increasing the overall accuracy of character recognition.

The present invention relates to the field of AI training, and in particular to a platform training method based on AI customer service, which reads a number of customer service data as training samples and divides them into a number of sub-samples according to a preset sampling rate, uses the sub-samples to train a number of AI customer service models respectively, verifies the training results according to the sub-samples, determines the case closing rate of each AI customer service model according to the verification results, sets a case closing rate threshold, compares the case closing rate of each AI customer service model with the case closing rate threshold, and obtains the corresponding case closing comparison result, determines the average pass rate of the AI ​​customer service platform according to the case closing comparison result, compares the average pass rate with the pass rate threshold, and when the average pass rate is less than the pass rate threshold, adjusts the sampling rate of the sub-samples. By adjusting the sampling rate in real time, the problem of low recognition accuracy caused by the inability to accurately match the AI ​​customer service model with the question data is effectively solved.

The invention discloses a method and system for monitoring a cowfish through unmanned aerial vehicle cruising and real-time video, and the method comprises the steps: S1, carrying out the cruising monitoring of the unmanned aerial vehicle above a target water area, shooting a video image, and selecting a picture frame including the cowfish in the video image; s2, marking the picture frames including the cowfish to obtain a cowfish image data set; pre-processing the images in the globefish image data set; s3, training the deep learning model by using the Jiangguinea pig image data set; s4, by means of the deep learning model constructed through training, real-time detection is carried out on the real-time transmitted unmanned aerial vehicle monitoring video, the cowfishes are marked in the detected video, and the number of the cowfishes is counted; and S5, statistics is carried out by using the appearing time of the video frame image of the dolphin in the video to obtain the appearing time of the dolphin and the total time when the dolphin is exposed out of the water surface. According to the method, the problem of low accuracy of catfish identification is solved, and the method has the characteristics of high robustness, strong generalization ability and high accuracy.

The application relates to a pedestrian re-identification method and belongs to the technical field of image recognition processing. The application proposes attribute-guided attention, provides attribute-related attention guidance for each attribute, performs feature extraction on a high-level feature map by using a convolution kernel with the same number of pedestrian attributes, obtains a feature map of each attribute, and thus improves the semantic representation effect of an attribute feature vector. Then, a dependency relationship between attributes is established according to the feature vector of each attribute, the recognition accuracy of one attribute is used to assist the recognition of another attribute through the dependency relationship, and the attribute recognition accuracy is improved as a whole.

The embodiment of the invention provides a method and device for recognizing vehicle information. The method comprises the steps that a first image, collected by a first camera, of a first vehicle parked on a parking space is acquired; under the condition that the vehicle information of the first vehicle cannot be recognized through the first image, determining a target vehicle matched with the first vehicle in a second image collected by a second camera for the lane according to the position of the first vehicle in the first image; and determining the vehicle information of the target vehicle identified by the second image as the vehicle information of the first vehicle. According to the invention, the problem of low vehicle information identification accuracy is solved.

The invention belongs to the technical field of image recognition and processing, and aims to provide an intelligent homework reviewing method and system, electronic equipment and a program product. According to the method, page structure adaptive recognition is realized through template matching, high-precision image correction is completed by using anchor geometric constraints, character recognition is performed through a single character recognition model, and objective question answering information is automatically scored. According to the invention, the problem of low recognition accuracy of subjective questions and complex answering scenes caused by recognition of handwritten texts of students depending on OCR in the prior art is solved, and the problem of low recognition accuracy of OCR caused by poor handwriting of students in the prior art is effectively solved; in addition, full-process high-precision intelligent reviewing from image acquisition, area positioning, character recognition to automatic objective question scoring and subjective question scoring summarization is realized, and homework reviewing efficiency and accuracy are greatly improved.

The invention provides a three-dimensional perception intention recognition method, and relates to the technical field of intention recognition, and the method comprises the steps: obtaining the head posture data of a user in a three-dimensional space and the multi-modal information of the three-dimensional space through the three-dimensional space information, obtaining the sight direction of the user through an eye movement estimation model, and fusing the head posture data, thereby achieving the recognition of the three-dimensional perception intention. The method comprises the following steps: acquiring the sight line of a user in a three-dimensional space, fusing three-dimensional space information, acquiring candidate objects on the sight line of the user, fusing multi-modal information, acquiring an object interacted by a user intention from the candidate objects, and acquiring the user intention according to the object interacted by the user, so as to increase in an intention recognition process. The problem that existing intention recognition is low in accuracy is solved, and the method is suitable for intention recognition.

A deep learning-based method for identifying a drivable area in a rural environment comprises the following steps of S1, collecting a driving image of the rural environment by adopting a vehicle with a camera, and establishing a rural road data set; s2, building a deep learning semantic segmentation network model, wherein the deep learning semantic segmentation network model comprises a coding module, a neck module and a decoding module; s3, training a deep learning semantic segmentation model, inputting a to-be-detected target image into the network model, and obtaining a target region in the image; s4, calculating a loss function, wherein the loss function comprises a cross entropy loss function and a dice loss function; and S5, predicting and marking a drivable area. The method is used for solving the problem that the road recognition algorithm is interfered by the unclear road structure in the rural road environment, and realizing the task of stably recognizing the drivable area in the road in different light and weather. And the interference of image quality and other problems on the accuracy of identifying the drivable area is reduced.

The present invention relates to the technical field of classroom behavior recognition, and in particular to a classroom behavior recognition method and device based on multi-scale dilated residual attention perception, which is as follows: collect classroom behavior images to construct a data set, annotate and enhance the data set, and then divide the training set and the verification set; construct a backbone network, input the image data in the training set into the backbone network features; construct a hybrid attention module, input the obtained feature input module to obtain features with attention perception; input the features to the part to obtain feature maps, and; input the feature maps, and to multiple decoupled prediction heads CBPH modules of yolov5 for multi-scale prediction, and finally locate the position of students in the classroom image, and identify the students' classroom behavior categories. The present invention can solve the problem of low recognition accuracy caused by inconsistent target scales in classroom scenes, and more effectively extract the discriminative features of classroom behaviors.

This invention discloses a natural language interaction method for a construction safety AI assistant, belonging to the field of artificial intelligence interaction technology. It includes: receiving voice or text interaction requests input by construction personnel, processing the voice or text interaction requests to obtain standardized text requests; based on a pre-trained natural language processing model, performing intent recognition and entity extraction on the standardized text requests, outputting intent-entity association pairs; constructing a construction safety professional knowledge base, retrieving from the corresponding knowledge base according to the intent-entity association pairs, and generating natural language response content that conforms to the understanding of construction personnel; feeding back the generated response content to the construction personnel in voice or text form, while simultaneously recording the intent type, entity information, and response satisfaction of this interaction. This invention can accurately understand construction personnel's requests for safety procedure inquiries, operating specification inquiries, and training examinations, providing immediate and accurate intelligent responses.

The invention provides a hook lifting rod bending image recognition method based on deep learning, relates to the technical field of image processing, and aims to solve the problem that in the prior art, due to the fact that the background of a hook lifting rod bending image is disordered and the number of similar parts is large, the recognition accuracy is low. The method is advantaged in that an attention mechanism is guaranteed to acquire a more accurate characteristic graph, deformable convolution is added, characteristics of a receptive field in the network are fully utilized, characteristic extraction capability of the model for irregular targets is improved, and a problem of low identification accuracy caused by relatively disordered background and relatively many similar parts of a hook lifting lever bending image is further solved.

The embodiment of the present application provides a kind of based on deep learning's heat dissipation substrate defect detection method and system, it is related to the technical field of heat dissipation substrate defect detection technology.Its method includes: obtaining the three-dimensional point cloud data of the heat dissipation substrate to be detected;The three-dimensional point cloud data is preprocessed to generate standardized input point cloud;The geometric features of the input point cloud are extracted and fused by the pre-trained defect detection model at multiple scales to obtain defect class prediction results.Through the present application, the problem of low complex defect recognition accuracy is solved, and the effect of improving the complex defect recognition accuracy is achieved.

This application discloses a method, apparatus, computer-readable medium, and electronic device for document authenticity verification. The method includes: acquiring anti-counterfeiting verification results obtained by performing anti-counterfeiting verification processing on multiple document images of a target document using an anti-counterfeiting verification model; performing feature mining on the anti-counterfeiting verification results of the multiple document images to obtain extended features; generating at least one feature combination based on the anti-counterfeiting verification results and the extended features, and verifying each feature combination using a verification model corresponding to each feature combination to obtain a verification result corresponding to each feature combination; and determining the authenticity of the target document based on the verification results corresponding to each feature combination. The technical solution of this application can verify the authenticity of a target document from multiple dimensions, improving the accuracy of document authenticity verification.