140 results about "Digit recognition" patented technology

An AI-powered augmented reality (AR) system for optimizing product placement and customer loyalty. The system includes: • a user interaction data collection unit configured to monitor consumer behavior in real time in any retail scenario (point-of-sale POS, online e-commerce and shopping areas within a public virtual technology space), with the unit analyzing consumer movement patterns, engagement metrics, eye-tracking and purchase intent; • an intelligent spatial mapping module to build a contextual understanding of high-traffic zones, underutilized areas and high-quality product placements using heatmap analysis, other spatial computer algorithms and the live movement of consumers; • A model-based product placement recommender based on machine learning, which uses one or more DL-based predictive models trained by importing current sales data, historical text interaction data, and consumer preference data to suggest optimal product placement so that customers see the advertisement and are likely to click through it to ensure a conversion. • a customer-centric, real-time AR visualization interface configured to allow customers to experience digitally enhanced product presentations directly via mobile AR applications, smart glasses, or in-store digital kiosks, enabling customers to preview product placements, rotate and resize objects, and visualize interactive advertising content before purchasing; • a customized advertising and sales promotion engine configured to generate augmented reality-based marketing content and provide users with tailored promotions and interactive ads based at least in part on an analysis of consumer sentiment by the system, as well as information on location data or consumer interaction history, wherein the customized advertising and sales promotion engine personalizes the configuration instructions for advertising and sales promotion to encourage consumer interaction and achieve higher sales revenues; • A secure, cloud-based data processing and analytics unit that uses an automated engagement model to link online and in-store sales information with established activity models for product placement decisions. The unit leverages AI-powered consumer insights to continuously recommend product placement changes based on a variety of programmatic retail trends; • a privacy-friendly AI module based on federated learning and on-device AI processing, providing consumer behavior analytics while complying with global data protection regulations and preventing the use of open data; • a processing framework characterized by a low-power mode suitable for AR rendering and AI calculations, and implemented via dynamic voltage scaling and energy management techniques for seamless and sustainable operation; The system according to claim 1, wherein the user interaction data capture unit further comprises an eye-tracking and sentiment analysis sub-module, wherein the sub-module detects facial expressions, emotional reactions, and attention levels of consumers in real time to determine the effectiveness of product placements and advertising strategies.

The invention discloses a speed limit sign detection and identification method based on YOLO and PPOCR. The method comprises the following steps: capturing a traffic scene image; adjusting the size of the original image, and scaling the image to a set resolution by adopting a nearest neighbor interpolation algorithm; the adjusted image with the set resolution is input into an improved YOLO model for speed limit sign detection, and detection frame position information of the speed limit sign is obtained; cutting the original image according to the detection frame information; carrying out image preprocessing on the cut image; adjusting the size of the preprocessed image, and scaling the image to a set resolution by adopting a nearest neighbor interpolation algorithm; and carrying out digital identification on the image with the adjusted size by using the improved PPOCR model. According to the method, the precision and adaptability of speed limit sign detection and recognition are remarkably improved, efficient and accurate detection and recognition of the speed limit sign and the speed limit value of the speed limit sign are achieved, and the method has wide application prospects.

The invention relates to a method for testing a participant's ability to accurately detect speech under background noise, comprising: (a) audibly presenting a series of 2-bit numbers, 3-bit numbers, 4-bit numbers and / or 5-bit numbers in a random or fixed order to a participant's first ear at a fixed or varying signal-to-noise ratio (SNR) in a quiet or noisy environment, (b) audibly presenting a series of 2-bit numbers, 3-bit numbers, 4-bit numbers and / or 5-bit numbers in a random or fixed order to the second ear of the participant at a fixed or random signal-to-noise ratio (SNR) in a quiet or noisy environment, where the numbers are presented in the native language of the participant or the fluent language of the participant, (c) recording the numbers or sequence of numbers correctly recognized by the participant, and (c) storing the recorded numbers or sequence of numbers at the second ear of the participant. And (d) calculating a score as a function of the percentage of the number or sequence of numbers correctly replicated by the subject, or as a function of the sound intensity difference between the speech and noise required to achieve the correct recognition of 50% sequence of numbers. The invention also relates to a device and a method for measuring the digital recognition capability related to hearing loss, and to a computer-implemented method for evaluating the speech recognition capability of a participant.

The invention provides an intelligent submission method and device for manual measurement and reading data. Synchronously collecting data of dam monitoring points; performing target detection, number identification, pointer identification and scale identification on an instrument area in the image to generate structured measurement and reading data; intelligent matching and automatic filling of the structured measurement and reading data and report fields are realized through a natural language processing technology; performing anomaly detection and threshold judgment on the filled report data, identifying a potential risk level, and selecting a submission path through a preset grading submission mechanism according to the risk level; and synchronously uploading the hash value of the original image data and the structured data to a block chain system, executing double-chain storage, and automatically starting an intelligent contract of an emergency process when a preset triggering condition is met. According to the method, the dam monitoring data submission efficiency and accuracy can be effectively improved, intelligent identification and real-time hierarchical transmission of multi-source data under complex working conditions are realized, and the credibility and traceability of the data are guaranteed through a block chain technology.

The invention discloses a method and system for improving image classification performance by using a spiking neural network, and the method comprises the steps: focusing on a neuron activation mode and synaptic plasticity in an unsupervised handwritten numeral recognition research process based on a triple pulse time-dependent plasticity learning rule, and after the fluctuation of accuracy is observed, carrying out the recognition of an unsupervised handwritten numeral recognition model, and carrying out the recognition of the unsupervised handwritten numeral recognition model. According to the method, key influence factors such as weight initialization, STDP randomness and hyper-parameter sensitivity are deeply analyzed; according to the invention, a common SNN structure is optimized, a pulse Inception module is realized, and the SNN is trained through a triple STDP; according to the method, the simulated annealing algorithm and the STDP are innovatively combined to be used for neuron parameter optimization, an optimal hyper-parameter combination is established through SA, after STDP training, the performance is remarkably improved under the condition of a limited data set, and particularly, the high efficiency of one-time perceptual classification of hundreds of digits is shown.

The invention discloses a multi-class mechanical water treatment metering instrument reading method based on machine vision, and the method comprises the steps: collecting and preprocessing a video, building an instrument image database, building an instrument type recognition algorithm, and carrying out the recognition of the type of an instrument which is recognized as a pointer type instrument. The position of a pointer tip relative to two nearest scale numbers is obtained through instrument scale number recognition and pointer position recognition of the pointer type instrument, and then instrument reading is obtained through calculation after the distance proportion between the point of the pointer and the two nearest numbers is calculated. For the instrument identified as the floater type instrument, the position of the upper edge of the floater and two scale readings closest to the position are obtained through instrument scale number identification and floater position identification, and then the numerical value corresponding to the coordinate where the intersection point of the center connecting line of the two scale numbers and the upper edge of the floater is located is calculated. According to the method, the aim of automatically identifying various common mechanical instruments of the water treatment device can be fulfilled by fusing machine vision with machine learning, and the method has the advantages of high real-time performance, high universality and the like.

The invention discloses a license plate recognition method and system based on deep learning, and the method comprises the following steps: collecting license plate image data, and generating an image data set; performing semantic decomposition operation, dividing into region code units, letter units and number units, and constructing a semantic structure chart; constructing a multi-path character recognition model, and taking an improved PCNN model as a trunk feature structure; respectively inputting the image data set into a multi-path character recognition model, outputting an area code recognition result, a letter recognition result and a number recognition result, and generating a candidate license plate recognition result; if illegal character combinations or structural unit missing information exists, path screening processing is executed, and a candidate path set is generated; reasoning and filling processing is executed, and a complemented path set is generated; and combining and outputting a license plate recognition result and a confidence alternative path set. According to the method, the character recognition accuracy is improved, meanwhile, the robustness to structure deficiency is enhanced, and the method has the advantages of being high in recognition precision and high in fault-tolerant capability.

A method and computer-implemented process for identifying behavioral anomalies and risk factors is disclosed. The method includes (A) obtaining, by a risk assessment module, first data associated with activity information of a digital identity with respect to a digital development tool; (B) associating, by the risk assessment module, second data with a first risk tag based on a risk-anomaly matrix; (C) generating, by the risk assessment module, a first predicted anomaly using a machine learning module trained using a first training set, wherein the first training set comprises the second data and the first risk tag; (D) associating, by the risk assessment module, the first predicted anomaly with a second risk tag based on the risk-anomaly matrix; and, (E) transmitting, by the risk assessment module, a notification associated with at least one of (i) the anomaly and (ii) the predicted anomaly, and the digital identify.

The invention provides a contract data identification and extraction method and system. A contract image data set is divided into a training set and a test set after the contract image data set is constructed; performing text line processing on the contract image by adopting rearranged CRNN feature serialization; constructing a character recognition model and a digital recognition model based on an FPN network; features extracted by the character recognition model and the digital recognition model are combined into a feature vector through weighted fusion, and after a CNN-RECR model is constructed, a training set is input into the model for training; and respectively inputting the training set and the test set into the trained model, and performing contract data identification and extraction. According to the method, the feature vector is converted into a multi-dimensional form by applying the residual structure unit so as to obtain the distribution condition of the statement structure, the character recognition model and the digital recognition model are fused and then introduced into the Hash layer to be combined so as to obtain the CNN-RECR model, and the high-dimensional feature is mapped into the low-dimensional binary code, so that the storage requirement can be reduced.

The invention provides a general meter reading automatic identification method, and relates to the technical field of computer vision and deep learning, and the method comprises the steps: carrying out the image collection of a meter to obtain an original image containing an instrument panel region of the meter, and carrying out the image recognition of the original image, obtaining an integer area and a decimal area in the instrument panel area and corresponding coordinates thereof, and determining physical layout positions of the integer area and the decimal area in the instrument panel area according to the coordinates; segmenting a plurality of numeric character regions from the integer region; and performing number identification on each number character area and decimal area, and sequentially splicing a plurality of numbers obtained by identification according to corresponding physical layout positions to obtain the meter reading. The method has the beneficial effects that a multi-stage cascade detection method is adopted, interference is filtered layer by layer (positioning-segmentation-classification digital recognition) through cooperative work of multiple processes, and the problem that cross-specification universal recognition is difficult to realize by a traditional single detection model is solved.

The invention provides a meter reading detection method based on a cascaded detection model, and relates to the technical field of intelligent metering of the Internet of Things, and the method comprises the steps: constructing a cascaded detection model which comprises a dial detection network, a character segmentation network and a digital classifier which are cascaded in sequence; the method comprises the following steps: acquiring a front image of a meter, inputting the front image into a cascade detection model, dynamically adjusting a feature perception range by a dial detection network according to a dial diameter of the meter, generating a dial region-of-interest frame, inputting the dial region-of-interest frame into a character segmentation network, and obtaining a character segmentation image of each character contained in the dial region-of-interest frame; and inputting the character cut images into a number classifier to obtain numbers contained in the character cut images, and taking a number sequence formed by combining the numbers according to the positions of the numbers in the region-of-interest frame of the dial plate as a reading detection result. The method has the beneficial effects that the problems of large dial positioning error, inaccurate character cutting, unstable number recognition and the like are effectively solved, the model volume is reduced by about 95% compared with a traditional single-stage model, and the calculation power demand is reduced.

The invention provides an implementation method based on light-induced linear reconfigurable artificial synapses, and belongs to the technical field of neuromorphic electronic devices and intelligent sensing signal preprocessing. The method comprises the following steps: providing an artificial synapse transistor capable of regulating and controlling conductance through voltage pulses, irradiating visible light in a channel or a conductance modulation area of the artificial synapse transistor, applying continuous voltage pulses under an illumination condition, reading the conductance in a preset time window after the last pulse is ended, and taking the conductance as a weight value of a corresponding pulse sequence; by adjusting illumination parameters, conductance is converted from nonlinearity to linearity along with the pulse number relation, the nonlinear factor is reduced compared with the dark state, the linear interval is widened, and linearity reconstruction in the weight updating process is achieved. According to the method, reconfigurable linear weight modulation can be obtained only by means of a light field and simple pulse driving, the method is suitable for high-fidelity compression coding of 5-bit and above time sequence pulses, feature separability and recognition accuracy can be remarkably improved in handwritten number recognition, handwriting identification and other scenes, and training time and energy consumption are reduced.

To provide an artificial intelligence-based collection server and its operation method that enable the collection of meter readings of a meter more conveniently and efficiently, without requiring a meter reader to directly obtain meter readings from the meter.SOLUTION: A collection server 110 includes: a collection event generation unit that generates a collection event for collecting meter readings at preset collection cycle intervals; a capture request unit that transmits a capture request command via the collection event generation unit, instructing a camera to capture and transfer an image of the area on the meter where the meter reading is displayed; an identification unit that, upon the transfer of the captured image of the area on the meter where the meter reading is displayed from the camera in response to the capture request command, identifies a first meter reading composed of numerical values from the captured image based on a preset number identification model; and a history storage processing unit that, after generating meter reading collection history information at the time the collection event occurred, stores the meter reading collection history information in a history database in which multiple pieces of meter reading collection history information are stored.SELECTED DRAWING: Figure 1

The invention provides a multivariate data regulation and control processing method based on three-dimensional visualization, belongs to the technical field of image data processing, and aims at solving the problem that an existing AI intelligent inspection unit mainly depends on a two-dimensional image to carry out instrument detection and digital recognition. In order to overcome the defects of inaccurate state analysis, insufficient multi-view data fusion and the like caused by target occlusion misjudgment and spatial information loss in the prior art, a multivariate data system is constructed by fusing a three-dimensional point cloud, a depth map and a multi-view two-dimensional image, and three-dimensional reconstruction, spatial feature extraction and a multi-modal data fusion algorithm are combined, so that the multi-view data fusion is realized. Accurate detection of a power distribution station house equipment target, three-dimensional visual analysis of state parameters and dynamic regulation and control processing are realized. According to the method, the anti-shielding capability of target detection, the accuracy of state analysis and the real-time performance of data regulation and control are effectively improved, the intelligent level of operation and maintenance management of the power distribution station house is remarkably optimized, and the labor cost and the fault risk are reduced.

The present invention discloses a fully analog circuit system for handwritten digit recognition based on a storage-computing integrated architecture, comprising a first analog circuit and a second analog circuit. The first analog circuit is composed of the following circuit modules: an input drive circuit, a first resistor array with a scale of m columns and 2n rows, a first current-voltage conversion circuit, a first subtraction circuit, and a first activation function circuit; the second analog circuit is composed of the following circuit modules: a second resistor array with a scale of n columns and 20 rows, a second current-voltage conversion circuit, a second subtraction circuit, a second activation function circuit, and a voltage comparison circuit. This system can eliminate the data handling problems brought about by the traditional von Neumann architecture and does not require the assistance of any digital modules, achieving ultra-low latency and accurate recognition. This system has very broad application prospects in the field of new neural network hardware system research, and is expected to provide a feasible non-von Neumann hardware solution for deep neural networks and edge computing.

The invention relates to the technical field of ecological product accounting, in particular to an EOD project ecological product value accounting method. Comprising the steps of obtaining and constructing an ecological product data table; and obtaining a character verification set and determining a target numeric character, thereby determining a supplementary verification character, and verifying and correcting the supplementary verification character. The method comprises the following steps: constructing a character verification set through repeated characters and associated characters to determine whether abnormity exists in an ecological product data table, if yes, determining a target numeric character, and forming analysis of a digit recognition feature, a sequence recognition feature and a character recognition feature according to the target numeric character, thereby obtaining a comprehensive recognition feature; and re-determining supplementary verification characters in combination with the comprehensive identification features, and constructing a supplementary verification data column to verify and correct the supplementary verification characters in the supplementary verification data column so as to re-generate an ecological product data table and execute ecological product value accounting while improving the accuracy of the accounting.

The invention discloses a flexible touch sensing structure and a handwritten number recognition method thereof, solves the problem that in an existing flexible touch human-computer interface, most of the number input modes are simple pulse signal detection, and encryption protection of information cannot be achieved, and belongs to the field of biological recognition identity authenticators. The flexible touch sensing structure comprises a feature region and a substrate region, the characteristic region is made of an ion conductor material, and the substrate region is made of a flexible insulating polymer; carrying out gray scale superposition on the plurality of handwritten digital images to obtain a gray scale distribution diagram; the areas with the highest and lowest handwriting probability distribution in the gray distribution map are substrate areas, and the areas with the highest and lowest handwriting probability distribution are feature areas. In order to classify and identify handwritten digits, a convolutional neural network signal classification algorithm is used. A single-dimensional time sequence voltage signal of a flexible touch sensing structure is collected to serve as input data of a convolutional neural network, and after the input data pass through the convolutional neural network, a digital classification result is output.

This invention discloses a license plate recognition method and system based on deep learning, comprising the following steps: acquiring license plate image data to generate an image dataset; performing semantic decomposition to divide the data into region code units, letter units, and number units, and constructing a semantic structure graph; constructing a multi-path character recognition model with an improved PCNN model as the backbone feature structure; inputting the image dataset into the multi-path character recognition model respectively, and outputting region code recognition results, letter recognition results, and number recognition results to generate candidate license plate recognition results; if there are illegal character combinations or missing structural unit information, performing path filtering processing to generate a candidate path set; performing inference imputation processing to generate a complete path set; and merging and outputting the license plate recognition results and the set of confident candidate paths. This invention improves the accuracy of character recognition while enhancing robustness to structural defects, possessing the advantages of high recognition accuracy and strong fault tolerance.

The present application relates to a kind of mixed STDP training method and system based on threshold increment cycle counting, construct including input layer, hidden layer and classification output layer STDP network architecture, and the threshold voltage increment of neuron is counted and controlled to the unsupervised learning process of STDP network architecture, the supervised learning process of combination classification output layer adjusts the weight value corresponding to neuron, by applying threshold increment cycle counting and neuron responsivity screening, effectively constrain weight update, reduce the weight update frequency, to greatly reduce the overall amount of computation of training, help to improve training learning efficiency.The present application method is applicable to handwritten numeral recognition, compared with traditional STDP training algorithm, can reduce 2.5 to 3.5 times amount of computation and improve about 3% classification accuracy.

The invention relates to a continuous learning method and device for a handwritten number recognition model, and the method comprises the steps: generating a main recognition model according to the training of an annotated data set, and selecting image samples which are wrongly recognized by the main recognition model from the annotated data set; training according to the error sample training set to generate an auxiliary recognition model, and cascading the main recognition model and the auxiliary recognition model which can be complementary with the main recognition model in performance to obtain a teacher model; based on the pseudo label set and the real label set, using the teacher model to guide the student model to train to obtain a distilled student model; and under the condition that the recognition performance of the student model after distillation reaches the recognition performance of the teacher model, replacing the main recognition model with the student model after distillation, returning to the model training cascade step and waiting for a new round of unlabeled data set for continuous learning, thereby not only endowing the intelligent model with continuous self-learning ability, but also improving the learning efficiency of the intelligent model. And the purpose of overcoming the disastrous forgetting phenomenon is also achieved.

The present application relates to the technical field of digital recognition, and particularly provides a nixie tube content recognition method and device, electronic equipment and storage medium, wherein the method comprises: acquiring a continuous video image sequence of a nixie tube, acquiring position information of a nixie tube character display segment according to a preset frame number by using a template matching algorithm; recognizing bright and dark states of the nixie tube character display segment by using a support vector machine (SVM); acquiring a brightness sequence unit of the nixie tube character display segment under the preset frame number according to a preset nixie tube arrangement mode; classifying the brightness sequence unit by using a deep learning method and outputting a recognition result; and the device comprises: a nixie tube and a PC terminal connected with the nixie tube, the nixie tube and the PC terminal are configured in an asynchronous state, and the PC terminal comprises: an acquisition module, a position recognition unit, a brightness recognition unit and a training module. The purpose is to solve the misrecognition problem caused by uneven brightness and defects in the nixie tube content recognition process.

The present invention discloses a meter image digital recognition system and method, the system comprises: an image acquisition terminal, a cloud recognition platform and a client; wherein the image acquisition terminal and the client are respectively connected to the cloud recognition platform in communication; the image acquisition terminal is arranged on the meter to be recognized, and is used to acquire the meter image with the meter reading, and transmit the acquired meter image to the cloud recognition platform through the network; the cloud recognition platform is used to process the meter image using a preset image processing algorithm, recognize the digital information in the meter image, and transmit the recognition result to the client through the network; the client is used to display the digital recognition result output by the cloud recognition platform. The present invention reduces the labor cost of meter reading, improves the accuracy, real-time and reliability of meter reading tasks, and is convenient for users to query.

The invention discloses a medical B-ultrasonic image automatic analysis system and method based on deep learning. The system comprises an image preprocessing module, a follicle detection module, a digital recognition module, an OCR text extraction module, an intelligent matching module and a report generation module. The method comprises the following steps: performing tilt correction, region extraction and perspective transformation on a B ultrasonic image, detecting follicle positions and morphological parameters by using a YOLO-11 model based on a rotating bounding box, and identifying digital labels; extracting an image bottom size text by adopting a CnOCR engine, and correcting to obtain structured data; intelligently matching follicles, numbers and sizes by calculating the Euclidean distance, and establishing triple association; and automatically generating a structured JSON report in combination with an anomaly detection mechanism. According to the method, full-process automation of follicle monitoring is achieved, the analysis efficiency, accuracy and consistency are remarkably improved, manual dependence is reduced, and reliable support is provided for clinical decision-making of assisted reproduction.

The invention relates to the technical field of convolutional neural networks, and discloses a handwritten number recognition method and accelerator based on an FPGA and a Winograd algorithm, an image sensor is used for collecting a handwritten number image, the handwritten number image is preprocessed to obtain the preprocessed handwritten number image, an MNIST database is used as a training sample library, and the training sample library is used for training the handwritten number image. The method comprises the following steps: building a handwritten numeral recognition network by taking a LeNet-5 network model as a basic model, inputting a preprocessed handwritten numeral image as a feature map into the handwritten numeral recognition network improved based on a Winograd algorithm, carrying out convolution processing on an FPGA (Field Programmable Gate Array) by using the Winograd algorithm, entering a full connection layer after pooling operation, realizing the full connection layer on the FPGA, and obtaining a handwritten numeral recognition result. Multiplication and accumulation operations of the weight and the input feature map are completed, and an identification result of the handwritten numerals is obtained through an output layer; according to the method, the parallel processing capacity of the FPGA and optimization of the Winograd algorithm on convolution operation are utilized, the single convolution speed is increased, the recognition speed and accuracy are improved, and the method can meet application scenes with high real-time requirements.

The invention discloses a label information identification method, system and device and a computer storage medium, and relates to the technical field of image detection, and the method comprises the steps: obtaining a to-be-identified target image; inputting the target image into a trained label detection model to obtain label information output by the label detection model; according to the label information, a digital image is extracted from the target image, and the digital image comprises an image carrying digital information in the label; and inputting the digital image into the trained label identification model to obtain digital identification information output by the label identification model. According to the method, the label detection model is used for performing label detection on the target image, so that the label information in the target image can be extracted, then the digital information in the label can be exposed, and a subsequent label identification model can conveniently identify the digital image to obtain the digital identification information in the label; the label digital information in the image is identified by means of the label detection model and the label identification model, the implementation is convenient, and the accuracy is high.

The invention discloses a new building electrical installation acceptance method and device, and relates to the technical field of electrical installation acceptance, the method comprises the following steps: obtaining house numbers of different rooms and code numbers and resistance values of resistor devices preset in different rooms to obtain a data file, the code numbers and resistance values of the resistor devices having uniqueness; measuring and recording actually measured current values shown by the corresponding electricity meters after the electric brake master control room opens the electric brake of each room; and extracting the code number of the resistor device in the data file and the house number of the room where the resistor device is located by using a digital identification model, calculating to obtain a theoretical current value, and comparing the theoretical current value with the actually measured current value to judge whether the ammeter is correctly installed. The method is low in cost and easy and convenient to operate, manual errors are reduced, and the efficiency and accuracy of electrical installation acceptance are improved.

The application provides an image recognition simulation model and system, the image recognition simulation model comprising: an optical diffraction neural network for a handwritten digit recognition task, the optical diffraction neural network comprising: an input layer, a diffraction layer and an output layer; the input layer modulates incident light into a wavefront array carrying image information; the diffraction layer performs multiple diffractions on the light output by the input layer to simulate a convolution weight matrix, and adjusts a phase distribution in a manner of error back propagation; and the output layer generates a focused light spot corresponding to the number of categories, the focused light spot focusing on a number corresponding to a plurality of detection points, and a detection surface light spot intensity distribution of the focused light spot is used to represent a classification probability. Through nano structure design, high degree of freedom regulation of light field amplitude, phase and polarization is realized in a two-dimensional plane; meanwhile, through multi-layer cascading extension network depth, feature abstraction capability is improved, nonlinear mapping is realized through multi-layer cascading diffraction, and a classification light intensity distribution is directly output, and the light intensity distribution represents a classification result.

Provided in the embodiments of the present disclosure are a reading method for a reading of a pointer instrument, and a device, a system and a storage medium. In the present disclosure, scale number recognition is performed on a pointer instrument image, such that a scale number recognition result is obtained; by taking as prompt information image position information of a detection box for a pointer of a pointer instrument, pointer segmentation is performed on the pointer instrument image; and by taking as prompt information target image position information determined on the basis of a pointer segmentation result and the scale number recognition result, scale segmentation is performed on the pointer instrument image. Therefore, the pointer segmentation and the scale segmentation can be completed in the case of zero-shot learning, and semantic segmentation of the pointer instrument image can also be implemented without studying the pointer instrument corresponding to the pointer instrument image. Since there is no need to study a pointer instrument of a specific type, pointer segmentation can be performed on an image of a pointer instrument of an unknown type, thereby improving the universality of semantic segmentation of a pointer instrument image, and further improving the universality of the reading of a reading of the pointer instrument.

The present invention discloses a method and system for recognizing the readings of pointer instruments in an inspection scenario. The method includes: acquiring an image of a pointer instrument; locating the pointer and the instrument in the pointer instrument image based on a positioning detection model, and locating the scale readings of the instrument in the image; fitting a dial image in the pointer instrument image by using an OCR direction detection model based on the above-positioned data; performing correction operations on the dial image based on plane rotation and spatial flipping to obtain a corrected dial image; performing pointer straight line fitting and digital recognition of the scale readings on the corrected dial image, and the digital recognition uses an OCR recognition model; calibrating the digital recognition result based on the prior information of the instrument; and obtaining the instrument reading based on the calibration result and the pointer straight line fitting result. The present invention can accurately perform instrument positioning and correction, ensuring the success rate and accuracy of the final reading of the pointer instrument.