487 results about "Category recognition" patented technology

The invention provides an underwater target identification method and system based on multi-source sensor data fusion. The method comprises the following steps: firstly, synchronously acquiring an original sound wave reflection signal and original optical image data of a target underwater area, then carrying out acoustic compensation processing on the original sound wave reflection signal to generate a target sound wave reflection signal, and meanwhile, carrying out optical compensation processing on the original optical image data to generate target optical image data; performing dynamic weighted fusion on the reliability measurement of the target sound wave reflection signal and the reliability measurement of the target optical image data through an adaptive fusion module to generate an enhanced multi-modal data set; and finally identifying the specific category of the underwater target through a multi-modal feature matching and decision-making mechanism. According to the technical scheme provided by the invention, the limitation of a single sensing mode is overcome, the quality of collected information is improved, the direct confirmation efficiency of common target recognition is also improved, and high accuracy and data reliability of target category recognition in a complex scene are ensured.

The invention discloses an emotion recognition method and system based on cross-modal alignment and matching enhancement. According to the method, firstly, feature extraction is carried out on text, audio and video modalities in a data set, and then a text and audio cross-modal emotion alignment module and a text and video cross-modal emotion alignment module are constructed respectively, so that cross-modal semantic alignment is realized. Constructing an emotion label matching module based on an alignment result, generating modal pairs with similar emotions but different labels by using a difficult negative sample mining strategy, and paying attention to cross-modal emotion consistency through a dichotomy task guide model; performing modal feature fusion on the three modals through a six-layer attention crossing mechanism, finally splicing feature vectors, inputting the spliced feature vectors into a long-sequence context fusion modeling module for deep modal fusion, and capturing cross-modal interaction information; and the fused features are sent to an emotion classification module, and a final emotion category recognition result is output.

The invention discloses a seabed sonar image recognition system based on a graph neural network, and the system comprises the following modules: an image collection module which is used for collecting an original sonar image and constructing an original sonar image data set; the preprocessing module is used for generating a de-noised sonar image data set; the three-dimensional landform structure modeling module is used for constructing a three-dimensional landform structure diagram; the graph structure coding module is used for obtaining multi-scale landform semantic feature representation; the decoding and evaluation module is used for reconstructing a preliminary seabed landform semantic segmentation map through jump connection and calculating a fitness vector; the optimization scheduling module is used for obtaining an optimized submarine landform semantic segmentation map; and the post-processing and output module is used for generating a final segmentation result and obtaining a landform category identification result. According to the method, the problems of fuzziness and fracture of a traditional convolutional neural network on a sea ditch-sea mountain-hill complex nested landform boundary are effectively solved.

The invention discloses a visual large model PCB defect detection system based on a YOLOv10 deep learning network. The system comprises an image screening module, a defect area positioning module, a defect category identification module, a defect data integration module and a defect evaluation display module. According to the method, the image of the printed circuit board is acquired through an industrial camera, an image quality evaluation standard is set, a high-quality input image is screened, the influence of interference factors is suppressed, image features are captured by a YOLOv10 network trunk extraction layer, a positioning area is accurately screened by applying the intersection and combination ratio of a prediction frame and a real frame, and the offset between the center of a candidate frame and the position of an anchor frame is calculated. According to the method, the positioning precision is strictly controlled, the false detection rate is effectively reduced, defect area images are cut and classified through YOLOv10 classification branches, and detection results are uniformly mapped into standard defect type identifiers according to informatization processing, so that the uniformity and standardization of detection types are realized, and the PCB defect detection efficiency is effectively improved.

The invention discloses a snapshot type hyperspectral imaging industrial online sorting system which comprises a support assembly, an optical imaging assembly and a motion control assembly. The bracket assembly consists of a bottom plate, and an adjustable bracket and a fixed bracket which are fixed on the bottom plate; the optical imaging assembly comprises a lighting unit, an imaging objective lens and a snapshot type hyperspectral camera, and can synchronously obtain space image information of a measured object and spectral information of a continuous wave band in a single exposure process; the motion control assembly realizes automatic transmission of the measured object through the transmission platform; the data processing module is integrated with an image preprocessing unit, a spectral analysis unit and a spectral database, and is used for end-to-end closed-loop control from original hyperspectral data generation to sorting decision making; the system adopts a metasurface light splitting structure to realize real-time decoding, performs material or category identification by using a pre-trained classification model, and finally automatically generates a sorting decision signal according to an identification result to drive a motion control assembly to complete sorting operation.

The invention provides a WSI long-tail data identification method and device based on multi-modal distillation guidance and a readable storage medium thereof, and aims to solve the problem of low tail category identification accuracy caused by long-tail distribution in weak supervision multi-instance learning of a digital pathological total image (WSI). According to the method, performance improvement is realized through double-branch integrated multi-instance learning, text feature extraction and prompt generation and multi-modal distillation optimization. The double branches are trained by adopting balanced distribution and original long-tail distribution data respectively, an aggregator is shared, and prediction consistency is constrained; text features containing semantic prompts are generated to serve as supervision signals, and semantic alignment is achieved through image-text comparison loss; soft and hard label collaborative distillation loss is designed, and a total loss function is constructed in combination with consistency loss. Experiments show that the algorithm can significantly improve the tail category and overall recognition performance, is high in universality, can adapt to various multi-instance learning methods, and provides reliable technical support for intelligent analysis of pathological images.

The invention relates to a rapid three-dimensional construction and recognition method for a complex scene, belongs to the technical field of intelligent mobile robots or unmanned vehicles, and solves the technical problems of low three-dimensional construction efficiency and low semantic recognition accuracy of the complex scene. Comprising the following steps: acquiring a point cloud registration image and an RGB-D image of a complex scene; performing multi-scale feature extraction based on the RGB-D image to obtain multi-scale image features; obtaining multi-scale scene point cloud features of the point cloud registration graph; performing multi-level feature fusion on the multi-scale image features and the multi-scale scene point cloud features to obtain a multi-scale fusion feature vector, and performing voxelization representation based on the multi-scale fusion feature vector to obtain a voxelization three-dimensional scene; and reasoning the voxelized three-dimensional scene to obtain a semantic recognition result corresponding to the complex scene. And rapid three-dimensional scene construction and category identification of a complex scene are realized.

The invention discloses a communication interference signal open set identification method, system and device and a medium. The method comprises the following steps: carrying out short-time Fourier transform on a communication interference signal to obtain a time-frequency graph; performing multi-scale feature representation on the time-frequency diagram by using a Res2Net-based multi-scale feature extraction network; constructing a projection layer to carry out hyperspherical embedding on the multi-scale features, carrying out modeling by utilizing vMF distribution, and designing a compactness and dispersity loss function; multi-scale features after modeling are input into an OpenMax layer for open set recognition, output of a projection layer is used as an activation vector to improve OpenMax, Euclidean distance is replaced with cosine distance, input sample unknown is calculated based on an extreme value theory, category probability distribution is adjusted, and some abnormal samples can be classified into unknown categories; the Res2Net more flexibly captures multi-scale information than the convolutional network; the separation of inter-class features and the compactness of intra-class features are effectively enhanced through the hyperspherical vMF distribution; the OpenMax layer endows the network with unknown category identification capability; the method has better performance and wider universality; the system, the equipment and the medium are used for realizing the communication interference signal open set identification method.

The invention belongs to the technical field of information data processing, and particularly relates to a method and device for solving catastrophic memory of learning forgetting based on knowledge distillation. The method comprises the steps that an initial model is selected, a standard data set is used for training, and initial model parameters are optimized by minimizing cross entropy loss; selecting a target forgetting category, adjusting model parameters by applying a category forgetting method, and generating a forgetting model which is low in target forgetting category identification capability and high in other category identification accuracy; and selecting residual category data from the original data set as incremental learning data, adding quantitative target forgetting category data as input data, taking the generated forgetting model as a teacher model, initializing a student model, constraining output distribution by the student model on the input data through knowledge distillation loss, and updating parameters of the student model. The problem that the forgotten model cannot keep the forgetting performance in the incremental learning process, and catastrophic forgetting may occur when new information is absorbed is solved.

The invention provides a fruit and vegetable identification and identification model and optical element combined training method and device, and relates to the artificial intelligence fields of deep learning, computer vision, image processing, micro-nano optical design and the like. The fruit and vegetable recognition method comprises the steps that a to-be-processed target image is obtained, the target image is a to-be-recognized fruit and vegetable image generated by a photosensitive element according to an obtained second optical signal, and the second optical signal is an optical signal obtained after an optical element modulates an obtained first optical signal, the first optical signal is an optical signal that light emitted by the light source irradiates to-be-identified fruits and vegetables and then enters the optical element through reflection; and generating a recognition result corresponding to the target image by using the recognition model, wherein the recognition result comprises a category recognition result and a freshness recognition result of the to-be-recognized fruits and vegetables.

The invention relates to the technical field of computer vision and deep learning, in particular to a fresh cut flower automatic classification method and system based on a two-step recognition neural network. According to the method, an image is collected, the background of the image is preprocessed, and meanwhile a flower main body area is reserved; feature extraction is carried out on the processed image, and the fresh cut flower category is identified; processing the image, and cutting the image to a flower core area; and finally, performing feature optimization on the cut image, extracting fine-grained features, and performing grade classification on the fresh cut flowers according to the fine-grained features. Based on parameter sharing and a two-step identification strategy, classification precision and training efficiency are improved through a hierarchical architecture of'large class identification-grade subdivision ', so that fine-grained features are rapidly extracted, and accurate grade classification is performed on the fresh cut flowers according to the fine-grained features.

The embodiment of the invention provides an image processing method and related equipment, and the method comprises the steps: obtaining a first sample image which comprises one or more objects; calling a pseudo tag generation model to perform feature extraction processing on the first sample image to obtain a first feature map; the pseudo label generation model is obtained by training a second sample image, and the image styles of the second sample image and the first sample image are different; performing domain generalization processing on the first feature map through a pseudo label generation model to obtain a domain generalization first feature map; the style information contained in the first feature map of the domain generalization is less than the style information contained in the first feature map; and through a pseudo label generation model, performing category identification processing according to the first feature map of the domain generalization to obtain a pseudo label of the first sample image, wherein the pseudo label is used for indicating the category to which the object in the first sample image belongs. According to the embodiment of the invention, the generation quality of the false label of the image can be improved.

The invention belongs to the technical field of industrial visual inspection, and provides an obstacle detection model training method and an obstacle detection method based on comparative learning and feature fusion. The training method comprises the steps of obtaining a point cloud sample set; extracting statistical and geometric feature vectors of each point cloud sample; pre-training the multi-scale feature fusion network by using the point cloud sample set based on comparative learning; constructing an obstacle detection network, wherein the obstacle detection network comprises a pre-trained multi-scale feature fusion network and a classifier; training an obstacle detection network by using the point cloud sample set to obtain an obstacle detection model; the multi-scale feature fusion network comprises M cascaded feature extraction modules; the cross-layer splicing module is used for splicing the statistical and geometric feature vectors of the point cloud samples and M feature matrixes; and a channel attention module and a point cloud feature projection head. According to the method, the precision of 3D point cloud obstacle category identification and the generalization ability of the classifier can be improved, and the model training convergence speed is high.

The invention discloses a ship category identification method based on incremental learning, and the method comprises the steps: carrying out the improvement of a training mechanism and a learning strategy of an original YOLO11 model, inputting a training set 1 into the original YOLO11 model for training, and obtaining the optimal weight of the model; the optimal weight of the model is loaded to the improved YOLO11 model, and an incremental target detection model is constructed through an EWC algorithm and a periodic data playback strategy; training an incremental target detection model by using the training set 2, and introducing a differential regularization strategy to enable the model to obtain the ability of inhibiting disastrous forgetting; and testing the incremental target detection model by using the test set, and continuously adjusting the regularization strength to obtain an optimal detection result. Incremental learning and target detection are combined, new ship categories are continuously learned through the model, meanwhile, stable detection performance of original ship categories is kept, and the problem of knowledge forgetting in the model iteration process is effectively solved.

The invention discloses a battery pack consistency evaluation method and device and electronic equipment, and belongs to the technical field of lithium batteries, and the method comprises the steps: extracting a plurality of voltage characteristics of a charging and discharging voltage change curve segment of a single cell in a battery pack, and a plurality of temperature characteristics of a temperature change curve segment; adding category labels for the curve segments according to the performance of the single battery cells; training a neural network model according to the voltage change curve segments, the voltage features, the temperature change curve segments, the temperature features and the category labels to obtain a category recognition model; and identifying the category of each single cell in a target battery pack based on the category identification model to obtain the category of the single cell, and evaluating the consistency of the target battery pack according to the category of the single cell. According to the invention, through feature extraction and analysis of the voltage change curve and the temperature change curve, the type of each battery monomer is determined, and rapid and accurate battery pack consistency evaluation is realized.

The invention relates to an input image category identification method and device, computer equipment and a storage medium, and the method comprises the following steps: carrying out the preprocessing of an input image, and obtaining a standard input image; extracting character information from the standard input image based on a pre-trained OCR model in combination with a polygon fitting technology; performing preliminary classification according to keywords, themes and context information of the text information to obtain a preliminary classification result; performing feature fusion on the visual features, the text features and the preliminary classification result of the standard input image by adopting a hierarchical feature fusion method to obtain image fusion features; and performing category division on the standard input image according to the image fusion feature and a pre-stored image template feature to obtain an input image category. The method and the device can be applied to financial service system application scenes, and can effectively realize accurate category identification of input images in a financial service system so as to facilitate subsequent system image information input.

The invention provides a power transmission line icing detection method and system, and relates to the technical field of intelligent control, and the method comprises the steps: eliminating the transmission interference of an optimization processing signal through a Manchester encoding protocol and a local cache verification algorithm in a channel switching process, and obtaining a clean transmission signal; and inputting the clean transmission signal into a pre-trained deep convolutional neural network, extracting ice layer space attributes through a plurality of convolutional layers and pooling layers, performing feature fusion through a full connection layer, calculating ice layer thickness distribution, identifying ice layer categories through a softmax classifier, and identifying the ice layer categories based on a thickness distribution diagram and an ice layer category identification result. And generating a detection result containing the ice layer thickness value and the spatial distribution attribute. According to the method, regional accurate analysis can be realized, and local feature omission caused by traditional overall analysis is avoided.

This disclosure provides systems and methods for controlling a vehicle based on a combination of high precision detection and high recall detection. The disclosed systems and methods can efficiently generate trajectories by reducing duplicate detection or duplicate calculation of objects or obstacles of common and known object types and objects or obstacles without class identification.

The invention discloses a track fusion method for incremental updating of a Gaussian mixture model and multi-state transition filtering, and relates to the technical field of track fusion, and the method comprises the steps: carrying out the preprocessing of observation information of a radar and electronic reconnaissance equipment; obtaining a feature vector of the observation information and performing target category identification to obtain a target category label set; generating an initial track of the observation subset by using a multi-correlation gate and a cost function; identifying track pairs with potential interaction and / or conflicts from the initial track set to form a target track set without intersection and / or interference; based on a plurality of preset state transition models, using the track quality score set and the sensor weight set to obtain a track fusion result; on the premise of considering the real-time performance and the precision, the cross track and the high maneuvering target are efficiently processed based on the range and performance difference of multiple sensors, and key filtering and associated parameters are adaptively adjusted according to the change of the environment and the sensor state, so that the continuous and accurate tracking of the target track information is realized.

The invention relates to the technical field of computer vision and computer graphics, in particular to an intelligent scene toning method and device based on feature recognition. The intelligent scene toning method based on feature recognition comprises the following steps: S1, pre-defining a scene category library, and constructing a corresponding LUT resource pool; s2, constructing a mapping table of multi-dimensional feature vectors of scene categories and LUT; s3, segmenting the input video into a plurality of video clips according to scene transformation; s4, extracting a multi-dimensional feature vector of a key frame in the video clip; s5, scene categories conforming to the multi-dimensional feature vectors are matched in the mapping table, intelligent scene classification is achieved, and corresponding LUTs are dynamically loaded; according to the method, accurate and rapid scene category identification and LUT matching can be realized, and the flexibility and rapidity of a film and television post-production process can be improved.

The invention provides an electromagnetic signal automatic modulation identification method and system during testing based on time-frequency fusion. The method comprises the steps of firstly constructing an electromagnetic signal data set; secondly, optimizing a kernel function to realize time-frequency feature modeling, and constructing a double-input path; constructing a deep neural network again, inputting the preprocessed signal sample into a time-domain branch for processing, analyzing a time-frequency spectrogram through the time-frequency branch, and introducing a channel attention mechanism to realize feature fusion; secondly, freezing a time-frequency branch, only training a time-domain branch, outputting a logit value from an input signal through a neural network, and smoothing logit distribution into probability distribution through a scaling variable; carrying out sample slicing processing on an input signal, and realizing convergence by fusing a prediction result; and finally, optimizing affine parameters of the batch normalization layer to complete identification of the electromagnetic signal modulation category. According to the method, the recognition accuracy of the model in a complex wireless environment is improved, so that the electromagnetic signal can still be stably and reliably recognized under the condition of low SNR (Signal to Noise Ratio).

The invention discloses a power quality composite disturbance detection classification and time positioning method based on MS-TCN + +. According to the method, composite disturbance category codes are designed for sampling points, an electric energy quality identification model comprising a prediction generation stage and a plurality of refining stages is constructed on the basis of MS-TCN + +, and multi-time-scale disturbance signal features are extracted by using double expansion layers in the prediction generation stage. And smooth loss is added in model training to avoid sudden change of categories of sampling points in a time sequence, and finally category identification and time positioning of disturbance signals are completed through a classification result of the sampling points, so that powerful support is provided for treatment of power quality disturbance.

Generate accurate answers to user questions. [Solution] The information processing system includes: a state information acquisition unit that acquires state information indicating the state of an object in question in response to the notification of question information, which is a question from a user regarding the object in question; a category identification unit that identifies a question category, which is the category of the question; a user support control unit that selects the state information that is classified into the question category identified by the category identification unit from the state information acquired by the state information acquisition unit, generates an answer request information by attaching the selected state information to the question information, requests a language model to generate an answer to the question using the answer request information, and acquires the answer generated in response to the request.

The invention provides a multi-dimensional enhanced open vocabulary video instance segmentation method, which comprises the following steps of: 1, proposing a novel open vocabulary segmentation thought through mathematical modeling and analysis; step 2, preprocessing a video input by a user, carrying out frame sampling, and realizing category-related segmentation in a Transform architecture by respectively strengthening interaction of category texts, image features and query vectors; step 3, adopting an instance-driven TopK time sequence matching strategy to improve the stability and accuracy of cross-frame matching; and 4, improving the category recognition capability through multi-scale feature fusion, and optimizing target classification. The open vocabulary video instance segmentation method can efficiently perform open vocabulary video instance segmentation, is widely applied to the fields of video monitoring, automatic driving, video indexing and the like, and promotes the development of video understanding and reasoning technologies.

The invention discloses a pet size intelligent measurement system and method based on a multi-task segmentation network. According to the system, the three-dimensional visual perception capability of a scene is constructed by using a binocular stereoscopic vision technology, and accurate depth information is obtained; meanwhile, an advanced multi-task segmentation network is adopted to process the pet image, efficient target detection, category recognition and pixel-level instance segmentation are achieved, and the pet contour (especially the hair edge) is accurately extracted; and finally, depth data provided by three-dimensional vision and the accurate contour output by the multi-task segmentation network are fused, and the real three-dimensional size of the pet is calculated. According to the method, the problems of low precision and poor contact stress and dynamic edge processing of a traditional method are solved, and non-contact and high-precision pet size intelligent measurement is realized.

The invention discloses a capacitive screen-based finger identification method and system, and an error correction method and system, and belongs to the technical field of human-computer interaction. The method comprises the following steps: acquiring original touch data of a capacitive screen, extracting and tracking a finger contact area, and generating a single-finger capacitance image sequence; and extracting space and time sequence features of the image by using a deep neural network of an encoder-decoder architecture to complete finger category identification. And for a misrecognition result, the user can activate an off-screen or non-off-screen error correction mode, finger recognition is carried out again, a wrong result is corrected, and misoperation is recovered. The problem of insufficient accuracy of traditional finger recognition is solved, flexible correction of error results is supported, withdrawing and repeated input operations are reduced, the recognition precision and user experience are improved, the model can be iteratively optimized through the recognition result, and the subsequent recognition effect is further improved.

A generator fault identification method, device, product and medium relate to the technical field of generator fault category identification. The method comprises the following steps: acquiring a vibration signal and a noise signal; performing multi-mode decomposition processing on the vibration signal and the noise signal to obtain a first sub-mode component and a second sub-mode component; determining a vibration signal sequence based on each first sub-mode component and the vibration signal; determining a noise signal sequence based on each second sub-mode component and the noise signal; determining a vibration characteristic matrix based on the vibration signal sequence; determining a noise feature matrix based on the noise signal sequence; determining a first recognition result and a first confidence coefficient based on the vibration feature matrix; determining a second recognition result and a second confidence coefficient based on the noise feature matrix; when the identification results are inconsistent, determining a fault identification result according to the confidence coefficient; and when the identification results are consistent, determining the first identification result as a fault identification result. The method is advantaged in that generator fault identification accuracy is improved.

The invention discloses an industrial part grabbing method and device and a medium, and relates to the technical field of robot control in industrial automation, and the method comprises the steps: obtaining an RGB image and a depth image including an industrial part; inputting the RGB image and the depth image into a pre-trained category recognition model to obtain a part category output by the category recognition model; inputting the RGB image, the depth image and the part category into a pre-trained pose estimation model to obtain a 6D pose of the industrial part output by the pose estimation model; and on the basis of the part category, the part attribute corresponding to the part category and the 6D pose, the grabbing requirement is determined, and the robot is controlled to grab the industrial part on the basis of the grabbing requirement. The industrial part grabbing method and device are used for solving the problems that in the prior art, when part grabbing is carried out without considering the characteristics of industrial parts, safety is poor, and operation efficiency is low, accurate recognition and safe grabbing of the industrial parts are achieved, and operation efficiency and reliability are improved.

The invention provides a category recognition model training method for rail transit fault diagnosis and a fault category recognition method, which can be applied to the technical field of rail transit fault diagnosis. The method comprises the steps that a training sample is acquired, the training sample comprises sample fault feature data and a sample fault category label, and the sample fault feature data represents fault feature data collected in a rail transit scene; oversampling processing is carried out on the sample fault feature data to obtain sample synthesis feature data, and the number of the sample synthesis feature data is larger than a preset number threshold value; processing the sample synthesis feature data and the sample fault feature data by using a generative adversarial network to obtain sample target generation data; processing the sample target generation data by using the initial category recognition model to obtain a sample fault category recognition result; and training the initial category recognition model based on the sample fault category recognition result and the sample fault category label to obtain a trained category recognition model.

The present application discloses an object category recognition model training method and apparatus, and an object category recognition method and apparatus, which can be applied to various scenarios such as cloud technology, artificial intelligence, intelligent transportation, and the Internet of vehicles. The method comprises: acquiring a newly added sample image of a sample object; on the basis of an initial object category recognition model, constructing a teacher model and a student model; inputting the newly added sample image into the teacher model for performing first object category recognition, and obtaining a first sample category; performing fusion on the first sample category and a newly added object category, and obtaining a fused category; inputting the newly added sample image into the student model for performing second object category recognition, and obtaining a second sample category; and, on the basis of a difference between the second sample category and the fused category, training the student model, so as to obtain a final object category recognition model. According to the model training method of the present application, the training duration is greatly shortened, and the model updating efficiency is improved.