64results about How to "Guaranteed recognition accuracy" patented technology

The invention relates to an in-vehicle gesture recognition method and system based on a directional loudspeaker, and the method comprises the following steps: collecting gesture data through the directional loudspeaker, and extracting an audio clip containing a gesture, thereby obtaining a gesture waveform; preprocessing the gesture waveform; sTFT transformation is carried out on the preprocessed gesture waveform, the gesture waveform is converted into a spectrogram, and the spectrogram is evenly cut and then serves as input of a gesture recognition model; and gesture recognition is carried out by using a gesture recognition model constructed based on the ResNet network. Compared with the prior art, the method has the advantages of high recognition accuracy, high recognition speed and the like.

The invention discloses a face recognition feature online updating and recognition method and device, and relates to the field of face recognition, and the method comprises the steps: matching a first feature of a face image with a first feature library corresponding to a lightweight model; similarity comparison is carried out on the second feature of the face picture and a second feature library corresponding to the high-precision model, and a first feature library corresponding to the updated lightweight model is judged through the similarity; the high-precision model is introduced on the basis of the lightweight model, and the first feature library corresponding to the lightweight model is updated through similarity judgment of the high-precision model, so that the accuracy and reliability of the features in the first feature library are improved, and online updating of the feature library in face recognition is realized. The recognition method based on the face recognition feature online updating method not only can guarantee the recognition rate, but also effectively improves the recognition precision, and achieves the high efficiency and accuracy of face recognition.

The invention discloses a network security threat perception device and method based on knowledge reasoning and algorithm service, and the device comprises a security threat perception and research and judgment subsystem which is configured to complete network security threat perception, and achieves the suspected threat research and judgment together with a knowledge extension subsystem, and generates a defense plan; the knowledge base knowledge extension and service subsystem is configured to extract knowledge from the network security threat field data for knowledge extension and provide knowledge service for the threat perception subsystem and knowledge base users; the intelligent algorithm model training and application subsystem is configured to perform intelligent enabling on a situation threat perception model uploaded by the threat perception subsystem, a knowledge reasoning model uploaded by the knowledge extension subsystem, a semantic association model uploaded by the semantic fusion subsystem and the like in the algorithm training subsystem; and the semantic knowledge fusion and application subsystem is configured to complete dynamic expansion, visual display and application service of the network security threat knowledge graph.

The invention relates to the technical field of intelligent security and protection, and discloses an intelligent security and protection system based on AI face recognition, which comprises the following modules: an image acquisition module used for acquiring image information recorded at a camera of current security and protection equipment, a face detection and recognition module used for carrying out portrait feature recognition, and a face recognition module used for carrying out face recognition. The system comprises a portrait recognition module, an image acquisition module, an accompanying movement module, a data storage and management module, a human face recognition module, an accompanying movement module, a human face recognition module and an accompanying movement management module, wherein the accompanying movement module controls a video recording device in the image acquisition module to move horizontally and vertically; the face recognition module is used for establishing, managing and storing a database of face data. According to the invention, through the built accompanying moving module, when it is detected that there is an abnormal person in the current area, the AI controls the accompanying driving module to drive the image recording device to carry out accompanying movement of the angle and the recording position along with the movement track of the abnormal person, so that the complete abnormal behavior of the abnormal person can be traced more comprehensively.

The invention provides a diversified attack behavior recognition method and system based on a fusion recognition model, and the method comprises the steps: constructing an original training data set based on an attack received by a local end, and determining the type of missing training data based on the number of all types of labels in the original training data set; on the basis of a pre-trained data generation model, supplementing the training data of the missing training data category, and updating a local training data set; performing local training on the attack recognition model based on the updated local training data set, uploading model parameters of the trained local model to a cloud, and determining a model weight by the cloud based on the number of training data in the uploaded original training data set of the local model; and carrying out weighted calculation on the uploaded model parameters of each local model based on the model weights, issuing the updated model parameters to each local end to construct an aggregation model, and carrying out attack identification on data received by the local ends based on the aggregation model.

The invention relates to the technical field of wearable intelligent equipment, and further relates to a method and device for determining the home position of a user and wearable intelligent equipment. The method comprises the following steps: when the wearable intelligent equipment is in an indoor environment, acquiring wireless signal data and multi-source sensing data; determining wireless signal feature data according to the wireless signal data, and determining a standing proportion, a charging behavior similarity and a position aggregation degree according to the multi-source sensing data; performing weighted summation on the wireless signal feature data, the standing proportion, the position aggregation degree and the charging behavior similarity; and when the weighted summation result is not less than the total threshold, determining that the target coordinate is the user home position. According to the method, the wireless signal data and the multi-source sensing data are effectively fused, the problem that a single judgment condition is prone to interference or misjudgment is avoided, the accuracy and reliability of user home position recognition are remarkably improved, a solid foundation is provided for subsequent intelligent services based on home scenes, and therefore the user experience is optimized.

This invention belongs to the field of materials testing technology and discloses a deep learning-based method and system for identifying the interface of SiC modified asphalt. It acquires multi-dimensional interface images through a layered time-series acquisition method using four devices: optical microscope, SEM, EDS-Mapping, and Raman spectrometer. Adaptive algorithms are used for targeted denoising, normalization, and registration, while attention mechanisms and weighted fusion strategies are combined to enhance interface features. Simultaneously, a multi-condition dataset containing a high proportion of minor defects is constructed to achieve deep fusion and accurate identification of multi-modal features, eliminating subjective errors from manual judgment and improving the fine-grained accuracy and detection efficiency of SiC modified asphalt interface identification. The U-Net model is improved for specific scenarios and designed to be lightweight. By optimizing the encoder-decoder structure, introducing a dual-channel attention mechanism and a condition-adaptive branch, and combining a loss function, the feature extraction capability of interface boundaries and minor defects is enhanced.

The embodiment of the present invention provides a control method, a mouse, a terminal and an Internet of Things system for an intelligent voice mouse, firstly collecting an original voice signal; then extracting its signal-to-noise ratio; finally, when the signal-to-noise ratio is greater than a preset threshold, according to a lightweight neural network, determining the target information in the original voice signal and transmitting it to a target terminal for identification; when the signal-to-noise ratio is not greater than the preset threshold, inputting the original voice signal into the lightweight neural network, obtaining the first output value of the noise processing module at the same time, inputting the first output value into the lightweight neural network, obtaining the target information and transmitting it to the target terminal for identification; wherein the input of the noise processing module is determined by the lightweight neural network and the original voice signal. Based on the original technology, the present invention adds a noise processing module to the terminal part, uses the intermediate value of the lightweight neural network to remove noise, and ensures the recognition accuracy under complex background noise.

This invention discloses a hybrid quantization method for image recognition models based on bit-flipping attacks, comprising: Step 1, performing initial quantization on the image recognition model to convert it into an initial quantized model; starting from the first layer of the initial quantized model, performing bit-flipping attacks layer by layer, and recording the sensitivity of each layer under attack; forming a sensitivity set of the initial quantized model from the sensitivities of all layers; constructing an optimization objective function and determining optimization constraints, and determining the optimal hybrid precision quantization strategy for the image recognition model based on the optimization objective function and the optimization constraints; Step 3, setting the bit width for storing the parameters of each layer of the image recognition model according to the optimal hybrid precision quantization strategy, thereby obtaining the optimal hybrid quantization model; Step 4, acquiring the image to be recognized and performing image recognition using the optimal hybrid quantization model.

This invention discloses a lightweight few-shot incremental learning method for power line defects. The method includes: first, constructing an inspection image dataset containing basic and incremental categories, and designing a lightweight feature extraction network; training the network using the basic category data to generate prototype vectors for each category, forming an initial prototype classification model; in the incremental stage, constructing incremental prototypes for a small number of samples of the new category, and introducing a prototype transfer mechanism based on inter-class tension, adaptively adjusting the new prototype based on the geometric similarity and discrimination conflict strength between the new and existing prototypes to alleviate category conflicts; simultaneously, introducing an adaptive feature fusion mechanism that perceives discrimination uncertainty, dynamically fusing the discrimination results of the main and auxiliary branches to improve recognition stability. This invention, employing the above-mentioned lightweight few-shot incremental learning method for power line defects, can achieve efficient and stable power line defect recognition in scenarios with few samples and continuously expanding categories.

The invention provides a substation drawing processing method and device based on deep learning, equipment and a medium. The method comprises the steps of receiving a transformer substation drawing and a design specification file which needs to be followed by the transformer substation drawing, dividing layout into a drawing area and a table area, processing the drawing area and the table area respectively, obtaining first structured data and second structured data, establishing semantic association of the first structured data and the second structured data, achieving association of the drawing area and the table area, obtaining structured semantic information, and displaying the structured semantic information. Analyzing the design specification file, and constructing a graph database knowledge graph and a vector database according to an analysis result; and searching and combining results by using structured semantic information, and judging whether the results accord with specifications or not. According to the invention, high-precision semantization and automatic compliance auditing of batch drawings can be realized, and the labor cost is reduced.

The present disclosure relates to a model decimation method, device, medium and electronic equipment. The method comprises: dividing a model surface into a plurality of sub-regions; rendering the model and determining state information of each sub-region under multiple perspectives, wherein the state information comprises brightness, contrast and color; determining a decimation degree of each sub-region according to the state information; and performing a decimation operation in the corresponding sub-region according to the decimation degree. The sensitivity of the human eye to different brightness, contrast and different color wavelengths is different, and the decimation degree of each sub-region determined according to the state information can ensure the recognition accuracy of the human eye. By determining the state information under different perspectives, the accuracy of the decimation degree of each sub-region determined can be improved. In this way, the complexity of the model can be reduced while ensuring the visual experience of the user.

The invention discloses a search agent training method and device and a storage medium, and the method comprises the steps: collecting a current state vector under the condition that a search agent carries out the key sign search of any sample cardiovascular three-dimensional diagram; inputting the current state vector into a search agent, and outputting an action instruction through the search agent; the observation window is controlled to execute the action instruction, an experience tuple is collected and stored in an experience playback pool, and the experience tuple comprises the current state vector, the action instruction, the instant reward value and a next moment state vector; and performing iterative training on the search agent based on an experience tuple stored in the experience playback pool, so that the trained search agent can output an action instruction sequence enabling the observation window to quickly cover a key sign area according to the state vector.

The utility model relates to the technical field of paper currency identification, and discloses an adjustable multi-currency automatic identification and classified storage mechanism, which comprises a machine body, a first conveying device is arranged in the machine body, a second conveying device is arranged on the lower side of the first conveying device, a third conveying device is arranged on the lower side of the second conveying device, and a third conveying device is arranged on the lower side of the third conveying device. An inclined plate is fixedly connected between vertical plates on the two sides of the machine body, and through cooperation of a suction fan, a second connecting pipe, a rotating flange, a conveying roller, an air suction opening, an arc-shaped plate and other structures, single paper money stripping is achieved through controllable adsorption force (dynamic adjustment can be conducted according to paper money materials). And the adsorbed banknotes are conveyed to the upper side of the second conveying device, so that the recognition accuracy of the subsequent recognition module on the banknotes is effectively improved, and the separation effect is improved more remarkably.

The utility model relates to early warning equipment technical field, and disclose an emergency lighting system based on wisdom building, including prop, the circumference of prop is fixedly connected with disc, the top of disc is fixedly connected with the ring of teeth, the circumference of prop rotatably connected with the board of turning, the bottom of board of turning is fixedly connected with the round angle board, the inner wall of round angle board rotatably connected with output shaft, the circumference of output shaft is fixedly connected with gear, the top of board of turning is fixedly connected with detector, the front of board of turning is fixedly connected with photovoltaic board, the present application can carry out the detection of larger range between building, when detecting the fire, can give the signal to the detection center and carry out the early warning and open the lighting lamp simultaneously and carry out the emergency lighting.

The application provides a lightweight identity recognition method combining voiceprint and earprint features. The method comprises the following steps: obtaining voice and ear canal echo signals of a known registered person and a person to be verified; extracting and fusing 13-dimensional mel-frequency cepstral coefficients (MFCC) of the voice and ear canal echo signals, i.e., obtaining voiceprint and earprint fusion features; inputting the fusion features into a lightweight identity recognition model to extract 128-dimensional embedding features of the known registered person and the person to be verified; calculating the similarity of the embedding features of the two types of persons by using a probabilistic linear discriminant analysis (PLDA) method; and determining whether the person to be verified is the known registered person according to the similarity. In the application, the fusion of earprint and voiceprint features improves the recognition performance of the classification model. The lightweight identity recognition model is obtained through a pre-training process, which reduces the equal error rate (EER) and greatly reduces the parameter quantity of the identity recognition model.

This invention discloses a method and system for geological hazard identification based on multimodal semi-supervised learning. The method includes the following steps: acquiring labeled and unlabeled orthophotos and elevation models of geological hazard areas; constructing teacher and student networks containing dual-branch encoders, extracting features from the orthophotos and elevation models respectively, and linearly fusing them, generating ensemble predictions using learnable dynamic weights; in semi-supervised training, calculating the uncertainty of the student network's prediction results in real time, and triggering an exponential moving average update of the teacher network using the student network parameters only when the decrease in uncertainty compared to the historical mean exceeds a preset evolution threshold; optimizing the student network parameters by combining supervised loss, confidence-weighted consistency loss, and student historical consistency loss; inputting the test data into the trained student network, and outputting fine-grained geological hazard segmentation results. This invention can achieve high-precision automatic identification of geological hazards under small sample conditions.

This invention discloses a method for detecting subsidence using simple linear iterative clustering, applicable to the road infrastructure industry. It can simply, accurately, and efficiently acquire subsidence information on asphalt pavements. Step 1: Scaling the image; Step 2: Dividing the image into multiple small regions using simple linear iterative clustering, significantly reducing the relative weight of spatial and gray-level distance to obtain a reasonable number and shape of regions; Step 3: Analyzing the hierarchical relationships between regions based on their geometric features to establish inclusion pairs and filtering out non-target regions; Step 4: Performing local planar fitting on the periphery of the remaining bottom-level sub-regions to eliminate the influence of pavement slope on subsidence depth calculation; Step 5: For the remaining bottom-level sub-regions, calculating the gray-level difference between each point on the outer edge of the sub-region and the center of the sub-region, and determining whether it is a subsidence area based on the proportion of depth differences exceeding a threshold. This invention can quickly segment images and identify potential subsidence areas.

The invention provides an automobile kicking type induction automobile door control system based on millimeter wave radar. The system comprises a millimeter wave radar sensing module, an automobile body controller, a target automobile door and a projection device. The projection device projects an induction projection area on the ground to guide a user to collect actions. The vehicle body controller is in communication connection with the millimeter-wave radar sensing module, the millimeter-wave radar sensing module can jump between a matching verification stage and a data input stage, the system judges whether a jump condition is met or not according to received current vehicle information, a current input action data set is obtained in the data input stage, and data is input through a statistical algorithm; a current verification action data set is obtained in the matching verification stage, and whether matching succeeds or not is judged through a data verification matching algorithm; if matching succeeds, a matching result command is generated and transmitted, and the vehicle body controller controls the target vehicle door to be opened according to the command.

The application relates to a kind of, belong to single-pixel imaging and target identification technical field, solve the existing single-pixel unmanned aerial vehicle recognition reconstruction image resolution low, noise artifact many, general detection model poor adaptability problem. Including the acquisition of detection data by the single-pixel imaging system of self-adaptive variable beam scattering layered detection, target image is obtained by TVAL3 algorithm reconstruction;YOLOv11 network is improved in direction, the output level of downsampling is simplified, the lightweight core module is replaced, the feature channel is compressed, the lightweight identification model is constructed, the training is completed in combination with microblock CLAHE preprocessing and multi-noise data enhancement, and the identification result is output by inputting reconstruction image. The identification precision and robustness in low signal-to-noise ratio scene are improved, the model calculation overhead is reduced, and full-range efficient unmanned aerial vehicle identification can be realized.

The invention discloses a kitchen open fire hidden danger early warning model based on YOLO lightweight, and relates to the technical field of artificial intelligence, and the model comprises the following steps: S1, obtaining an original monitoring image of a kitchen environment, employing a preprocessing algorithm fusing bilateral filtering and adaptive histogram equalization, carrying out the denoising and enhancement of the original monitoring image, and outputting a standardized image; and S2, based on the standardized image, applying a structured pruning strategy to a pre-trained YOLOv5 model, and removing redundant convolutional layers and channels in a backbone network of the pre-trained YOLOv5 model. According to the kitchen open fire hidden danger early warning model based on YOLO lightweight, through a preprocessing mode of fusing bilateral filtering and adaptive histogram equalization, image noise is removed, flame edges and key features are completely reserved at the same time, the anti-interference capability of flame recognition in a complex kitchen environment is greatly improved, and the method is suitable for popularization and application. Clear risk guidance is provided for the user, the hidden danger can be quickly handled, and the kitchen fire protection level is comprehensively improved.

This invention discloses a method for ship traffic detection and analysis based on lidar, relating to the field of lidar positioning technology. The method includes: using historical point cloud detection data sequences from lidar to predict the distribution sequence of ship states in a target water area within a preset time window; optimizing the concurrent video monitoring angle sequence based on this prediction sequence, and collecting video stream data according to the adapted monitoring angle sequence; formulating an adapted image dimensionality reduction mechanism based on the current image interference intensity and the predicted ship state distribution sequence to perform data dimensionality reduction on the video stream data and obtain a key image frame sequence; acquiring the current point cloud detection data sequence of the target water area within the preset time window through lidar monitoring, combining it with the key image frame sequence to perform ship traffic statistics and ship type identification, and outputting a ship traffic detection result sequence. This invention effectively improves the perception performance of ship traffic detection while saving computing power.

The application provides a sensitive information identification method and device, network equipment, a medium and a program product. The method comprises: performing semantic coding processing on a target text to obtain a global semantic representation of the target text and a context vector representation of a word element; performing classification identification processing on sensitive information in the target text according to the global semantic representation to obtain a category identification result; performing sequence labeling processing on the sensitive information in the target text according to the context vector representation to obtain a first identification result; performing sequence labeling processing on the sensitive information in the target text according to the context vector representation to obtain a first identification result of the sensitive information in the target text; and cross- verifying the category of the sensitive information in the first identification result and the category identification result to update the first identification result. The method provided in the embodiments of the application has high identification accuracy and efficiency for sensitive information.

The invention discloses an intelligent identification linkage protection method for 35kV and below equipment based on multi-source adaptive fusion and a deep and shallow double-path network. The method comprises the following steps: acquiring multi-source data of target equipment, splitting the multi-source data into instantaneous and steady-state feature flows, extracting features by using a deep and shallow dual-path network, and performing dynamic self-adaptive fusion according to a real-time operation scene to obtain comprehensive features; inputting a time sequence formed by continuous comprehensive features into an LSTM network subjected to equipment exclusive gating adjustment, and outputting a combined recognition result of an equipment type and a working state and an initial confidence coefficient in combination with a hierarchical time sequence matching mechanism; the confidence coefficient is calibrated according to the characteristic fluctuation degree and the sample difficulty, and finally the protection strategy is adjusted in a hierarchical linkage mode according to the calibrated confidence coefficient. According to the method, the dynamic scene adaptability and the identification accuracy of a small number of devices are improved, the risk of misjudgment and missed judgment is effectively reduced through linkage of confidence coefficient calibration and a dynamic constant value, and the reliability of a protection system is enhanced.

The application provides a human action recognition and quality evaluation method and system based on multi-modal data fusion, the application collects RGB visual data, 3D skeleton data, inertial IMU data and electromyography sEMG data of human action, and carries out corresponding pretreatment; through branch parallel feature extraction, RGB visual feature vectors, 3D skeleton feature vectors, IMU inertial feature vectors and sEMG electromyography feature vectors are obtained respectively; based on the modal reliability values of each branch, multi-modal fusion feature vectors are obtained by dynamically distributing and fusing the weights of the cross-modal attention mechanism to the multi-path features; the action recognition result is obtained by using a lightweight action recognition network; and the action recognition result is quality evaluated through the comprehensive action quality score. The application can not only accurately recognize the action category, but also quantize the action quality from three dimensions of form, dynamics and force, thereby improving the robustness and accuracy of action recognition in a complex scene.

The invention belongs to the technical field of natural language processing, and particularly relates to a multi-choice question misunderstanding reordering method based on a causal big language model, which comprises the following steps of: aiming at the problems of noise existing in semantic vector retrieval candidate misunderstanding, insufficient logic discrimination capability of a traditional model and high resource consumption of big model deployment; the method comprises the following steps: firstly, obtaining a query context containing a question and a student error option and a Top-K candidate misunderstanding, and splicing the query context and the Top-K candidate misunderstanding into a to-be-discriminated sequence; inputting a causal large language model subjected to LoRA fine tuning, calculating a correlation score through a generative or classification head scoring mechanism, and optimizing sorting robustness in combination with list-level cross entropy loss; and finally, the perception weight quantization compression model is activated through 4-bit. According to the scheme, by means of the strong logical reasoning ability of the causal big language model, interference terms with similar texts but different logics are accurately removed, and misunderstanding recognition precision is greatly improved; meanwhile, video memory occupation is reduced to about one fourth of an original model, reasoning delay is shortened, low-resource edge equipment such as a learning machine and a mobile terminal APP is successfully adapted, and an efficient and feasible landing scheme is provided for intelligent education personalized tutoring.

The application provides a security video monitoring method and system based on cloud edge collaborative computing, relates to the technical field of image communication, and solves the technical problem that in the prior art, in a resource-limited scene, high-value security video data processing quality is significantly reduced due to the lack of content perception capability. The method specifically comprises the following steps: acquiring a video data stream; performing semantic importance evaluation on the video data stream to obtain a semantic weight; generating a joint scheduling strategy based on the semantic weight and a current node resource state; the joint scheduling strategy comprises a calculation task offloading granularity and a data coding parameter; the current node resource state comprises a local available calculation resource, a cloud available calculation resource and a network transmission bandwidth; and performing calculation task offloading and data coding on the video data stream according to the joint scheduling strategy. The application is used for security video monitoring.

The invention discloses a distributed dynamic load identification method and system based on time variable separation. According to the method, distributed loads are divided into a distributed load function capable of directly separating space-time variables and a distributed load function in a general form, time-domain loads are converted into frequency-domain loads under two conditions based on orthogonal polynomial fitting one-dimensional and two-dimensional function theories, and the time-domain loads are converted into frequency-domain loads. A Chebyshev orthogonal polynomial is used to complete coefficient identification of the orthogonal polynomial in a frequency domain, and a time domain-frequency domain Fourier transform theory and a least square method are combined to further complete distribution time domain dynamic load identification of the continuous beam. According to the method, a large amount of discrete time domain data is not needed for modeling, and the identified distributed load can be directly and clearly represented; only the column vector matrix needs to be identified, so that the identification efficiency is improved and the identification precision is ensured on the basis of greatly reducing identification calculation data.

The invention discloses a multi-mode weak texture slope remote sensing image segmentation method based on SAM. The method comprises the following steps: step 1, constructing a weak texture slope segmentation model based on an SAM network; step 2, performing bilateral filtering and normalization processing on the multi-mode weak texture slope remote sensing image; 3, training a weak texture slope segmentation model by using Dice loss and cross entropy loss in a combined manner, and freezing parameters of an image encoder in the training process; and step 4, inputting the processed multi-modal weak-texture slope remote sensing image into the trained weak-texture slope segmentation model, and outputting a segmentation result of slope crack features in the multi-modal weak-texture slope remote sensing image based on the weak-texture slope segmentation model of the SAM network. According to the method, a lightweight HRSI module and a modal perception module are introduced, and a multi-scale attention mechanism is introduced into an image encoder, so that efficient application of SAM in a multi-modal weak texture slope image segmentation task is realized.

This application discloses a detection device and its control method, control device, and readable storage medium, relating to the field of equipment detection technology. The control method of the detection device includes: controlling a current generator to send a first current signal to a current transformer and a current processor respectively; controlling the current processor to receive a second current signal sent by the current transformer, the second current signal being a feedback current signal after the first current signal passes through the current transformer; when the current processor receives the first and second current signals, controlling the current processor to determine the current difference between the first and second current signals; and determining the performance state of the current transformer based on the current difference. This application improves the detection accuracy of the detection device for current transformers.