44 results about "Discrimination function" patented technology

The invention provides a knowledge base intelligent deep auditing method and system based on an AI large model, and the method comprises the steps: obtaining an unstructured file uploaded by a user, analyzing the unstructured file, extracting a semantic auditing unit, and constructing a semantic auditing unit set; constructing a graph structure based on context dependence according to the semantic auditing unit set, marking continuous, neighbor, reference and reverse logic edge types, and fusing a compliance attention mechanism to update node representation to obtain a node semantic representation set; identifying candidate problem items through a double-layer risk discrimination function by combining the graph structure and the node semantic representation set design according to the introduced law and regulation knowledge embedding library; and executing causal chain backtracking on the candidate problem items to generate a structured auditing conclusion. The auditing accuracy and efficiency are remarkably improved, and the manual rechecking burden is reduced.

The invention provides a multi-parameter fusion organic matter type comprehensive discrimination method and system. The method comprises a multi-modal feature fusion method based on heterogeneous discrete state and continuous distribution state data; constructing a mechanism by combining a composite discriminant function of tensor product and cross convolution operation; the invention discloses a dynamic threshold controlled iterative weight optimization system. According to the method, a discrete data set of a unified coordinate system is generated through a rigid-non-rigid mixed registration algorithm for density compensation of a Vorono i graph, a continuous distribution data set is generated by adopting anisotropic kernel function interpolation, and a dual-channel feature extraction architecture is constructed to realize parallel feature extraction of a pore topology and a concentration field. And establishing a composite discrimination function containing a Kronecker product and a three-dimensional cross convolution integral, and realizing iterative convergence of the multi-modal characteristic residual error through an alternating direction multiplier method and a dynamic quantile threshold.

The invention discloses a power grid data knowledge graph completion method and system based on a steady-state model rule, and relates to the technical field of power grid data processing, and the method comprises the steps: carrying out the preprocessing and screening of power grid reference steady-state data, obtaining an effective data subset through the screening of a discrimination function, forming an initial triple set according to an entity and a relation, defining a knowledge graph structure and a storage form, and adding attribute information for nodes and edges; constructing a power flow equation set, defining an optimization problem to minimize a residual error of a power flow equation, performing iterative optimization by adopting a variational inference method, complementing missing electrical parameters, integrating the complemented parameters, updating nodes, forming a complete power grid data knowledge graph, performing verification and consistency check, and ensuring data logic consistency and accuracy. By optimizing the power flow equation, the missing electrical parameters are complemented, and the data integrity and accuracy are improved, so that more accurate information is provided for power grid operation, and the operation efficiency, reliability and stability of the power grid are improved.

The invention discloses a high-speed refresh positioning algorithm and system based on wireless signal processing, and relates to the technical field of high-speed refresh positioning, and the algorithm specifically comprises the following steps: analyzing the cycle number of a signal data unit in a current refresh cycle, judging whether there is a signal data unit with the cycle number not equal to the current cycle number, and if yes, executing the following steps; when the signal data units exist and the number ratio of the signal data units exceeds a preset threshold value, a cross-cycle data synchronization state judgment process is started; and on the basis of the time label distribution, the historical data proportion and the source period number of the signal data units in the current refresh period, extracting feature information used for representing time consistency and a periodic structure, and inputting the feature information into a classification discrimination function to determine a cross-period data synchronization state corresponding to the current refresh period. According to the method, the problem of positioning mistaken refreshing caused by cross-cycle data dislocation is solved, refreshing control based on synchronous state judgment is realized, and the positioning accuracy and the system stability are effectively improved.

The present application relates to the technical field of safety monitoring and industrial automation, and discloses a silo safety monitoring method and system, which synchronously acquires coal pile surface point cloud, lateral pressure, vertical load and environmental data through the arrangement of a laser scanning unit, an annular pressure sensing belt, a weighing sensor and a temperature and humidity node; a high-precision three-dimensional morphology is reconstructed based on an adaptive grid refinement algorithm, and a joint representation model is constructed by fusing mechanical data; a stability discrimination function is established in combination with coal body physical parameters to identify the instability critical region; and through pattern matching with a historical stability morphology library, hierarchical early warning of abnormal working conditions such as suspended arch, bonding and blockage and generation of blockage removal strategies are realized. The present application constructs a joint representation model of coal pile three-dimensional morphology and mechanical state by fusing laser point cloud, pressure distribution and load sensing data, breaks through the limitation of traditional single-point material level measurement which cannot reflect the overall coal pile structure, and realizes high-precision measurement with a storage calculation error of less than 2%.

The invention discloses a radar track target classification method based on statistical knowledge and a neural network, and the method comprises the steps: collecting original track data, and carrying out the marking and cleaning processing, and obtaining a standard track data set; extracting kinematics, bistatic radar cross section and track morphological features based on the standard track data set, and constructing an original feature set; performing feature weight learning through a neighborhood component analysis method, and screening according to a preset weight threshold to obtain an effective feature subset; performing statistical distribution analysis on the effective feature subsets according to target categories, determining significant features of each target category, and setting a threshold interval; performing preliminary classification based on a statistical knowledge classifier, and determining whether the to-be-classified track is an easily-classified sample or a difficultly-classified sample through a category discrimination function; and classifying the samples difficult to classify based on a back propagation neural network classifier, outputting the probability value of each target category through a neural network model, and taking the target category with the maximum probability value as a final classification result. According to the invention, the target classification accuracy and real-time performance can be improved.

The invention is suitable for the technical field of software engineering, and provides a GraphCodeBERT-based code quality analysis method. According to the method, the consistency of codes and annotations serves as the key dimension of code quality analysis, whether the codes and annotation semantics under the function granularity are consistent or not can be judged, the reason of a model for the consistency relation of the codes and the annotations is analyzed, and the contribution value of each division dimension of the codes and the annotations to a prediction result is quantitatively determined and visually presented. According to the method, the processed data is suitable for analysis of various pre-training code models, and model interpretability research and code quality evaluation are supported; multi-dimensional data is provided through deep contribution degree analysis to assist in deep research; developing personnel are assisted in optimizing code management by means of a visual means, and high-risk codes are identified and improved; in the aspect of code quality evaluation, automatic detection of consistency of function-level codes and annotations is realized, data support is provided for code review and optimization, and the efficiency of software development and maintenance is improved.

A system and method for denoising a grey scale image of a pattern on a substrate, including: scanning an electron beam along a pattern, detecting a scan line waveform from the scanned electron beam, obtaining a model scan line waveform from the detected scan line waveform, augmenting the model scan line waveform and adding noise to the augmented model scan line waveform, inputting the noisy augmented model scan line through a deep learning neural network (DNN) process, comparing the output of the DNN process to the augmented model scan line waveform before noise is added, and backpropagating the compared results into the DNN process to obtain a noise discrimination function.

The invention discloses a voice activity detection model training method, a voice activity detection method and a related device, and relates to the technical field of audio processing, and the training method comprises the steps: carrying out the training of a first training audio marked with a frame-level audio class, and obtaining a first voice activity detection model with a semantic integrity discrimination function, the audio category of an audio frame of the first training audio is one of voice, non-voice at a semantic incomplete position and non-voice after semantic complete voice; a second training audio marked with a frame-level audio category is utilized, the first voice activity detection model is assisted, a second voice activity detection model capable of capturing semantic information in the voice is obtained through training, and the audio category of an audio frame of the second training audio is one of voice and non-voice. The voice activity detection model trained by the training method disclosed by the invention can capture semantic information of the audio, so that a reasonable category can be given for each audio frame of the audio by referring to the semantic information.

The invention provides a test paper automatic detection method, electronic equipment and a storage medium, and belongs to the technical field of test paper instant detection. The detection method comprises the following steps: acquiring a plurality of detection points and color characteristics of the detection points in a test paper reaction region; and calculating the distance between the detection point and the center of the reaction area so as to endow the detection point with a decision weight. And combining the color features of the plurality of detection points into an observation matrix, and combining the basis matrixes of the plurality of detection points based on the decision weight into a block diagonal matrix. And based on the observation matrix and the block diagonal matrix, obtaining an optimal coefficient matrix of the color features of the detection points so as to reconstruct the color features of the detection points. And respectively establishing a background model of each detection point, and verifying the consistency of each background model so as to filter the color of the test paper. And estimating the concentration of the detection point according to the color feature of the detection point, and carrying out substance identification through a discrimination function and a decision weight. And outputting the concentration and name of the substance in the reaction area. According to the method, the accuracy of a detection result can be improved.

A method for determining the optimal gain ratio for image fusion of large dynamic range is proposed. Firstly, a dual-gain imaging model is constructed based on the low and high gain data obtained by SCMOS. Then, the model is used to simulate the high and low gain images under different gain ratios, and the HDR results of image fusion of large dynamic range are obtained based on the dual-gain images. Secondly, the problem of image quality degradation caused by the large number of overexposed pixels in the high gain image, and the quality evaluation requirements of the HDR image are considered, and the optimal gain ratio discrimination function of SCMOS detector is constructed based on the relative brightness saturation ratio, the sharpness index and the gray weighted index. Finally, the gain ratio corresponding to the maximum value of the optimal gain ratio discrimination function is selected as the optimal gain ratio value for image fusion of large dynamic range.

The application discloses a security box video monitoring system and method with intelligent discrimination function, and relates to the technical field of video monitoring. The system can dynamically extract the trend of the image frame brightness gradient change rate set by continuously collecting images and constructing the image frame average brightness sequence and the image frame time sequence. The system uses the abnormal occlusion image frame data set as the input basis, combines the spatial change trend of the image frame brightness gradient change rate set and the optical flow displacement vector set, and outputs the occlusion behavior possibility score to realize the probability determination of the occlusion intention, avoid the misjudgment of the incidental non-human interference, performs the edge structure analysis in the optical flow disturbance concentrated area, generates the optical flow disturbance concentrated area boundary contour morphological feature data set, calculates the occlusion area proportion, and then fuses the occlusion behavior possibility score to form the optical flow disturbance concentrated area feature fusion input set. Finally, the intelligent classification is realized through the interference behavior discrimination label.

The invention relates to a polyhedral structure sound wave orientation sensor module, an array and a sound wave orientation method in the technical field of sound wave detection. The sound wave directional sensor module comprises a sensor and a calculation unit. The sensor comprises a rigid backing and n piezoelectric sensing units. And the calculation unit is used for determining a primary direction surface and a secondary direction surface according to the signal amplitude of the sensor, calculating the amplitude ratio P of the primary direction surface and the secondary direction surface, inquiring the amplitude ratio-incident angle mapping database based on the P to obtain the azimuth angle theta of the sound wave on the primary direction surface, and calculating to obtain the sound wave direction based on the theta and the primary direction surface. According to the invention, through the innovative polyhedral integrated design, the sound wave direction discrimination function which can be completed by a sensor array originally is realized in a single compact device, the problem of direction fuzziness of a traditional single-unit sensor is solved from the physical principle, and in addition, through a two-step decoding algorithm which is tightly combined with the sensor structure, the sound wave direction discrimination function is realized. And an optimal closed loop from physical perception to information output is formed.

The invention relates to a rainfall data analysis method based on multi-modal data fusion and space-time correlation perception, and the method comprises the steps: collecting historical non-blocking operation and maintenance data of a rain gauge in a target region in a set time period through a system, and storing the historical non-blocking operation and maintenance data in a computer system in a partitioned manner to construct an equipment historical baseline library; the system receives real-time multi-mode monitoring data of a preset number of rain gauges in a target area to construct a real-time monitoring data set, converts historical non-blocking operation and maintenance data and time sequence data of the real-time monitoring data set into scores capable of being quantized in a unified mode according to time step column values, and determines analysis points and peripheral points in the target area. And calculating a score residual error between each time step column value and a baseline midpoint value in a historical baseline library, generating a two-dimensional baseline residual error matrix, calculating a dynamic deviation index and a state continuity parameter, constructing a blockage judgment function, and combining an operation result with a preset threshold value to complete blockage judgment. According to the method, full-process automation can be realized, and efficient analysis is realized.

The invention provides a blast furnace condition trend early warning system based on PCA (Principal Component Analysis) and LDA (Latent Dirichlet Analysis), which comprises a data preprocessing module used for cleaning and standardizing multi-source heterogeneous data of a blast furnace; the feature extraction module ensures feature quality; the working condition identification module is connected with the feature extraction module to realize intelligent identification and classification of the blast furnace operation working condition; and the early warning decision optimization module forms an optimal early warning decision. According to the method, a PCA technology is used for carrying out dimension reduction on operation parameters, then through analysis of load vectors, principal components are clearly mapped into characteristic indexes with direct process significance, the characteristic indexes are input into an LDA model, through maximization of the ratio of inter-class divergence to intra-class divergence, the optimal projection direction is found, the characteristics are projected to a low-dimension discrimination space, and finally, the optimal projection direction is obtained. And a classification rule based on a discrimination function is constructed, normal, early warning and serious abnormal working conditions are identified, and the classification accuracy of the system on complex operation modes is improved.

The invention discloses a circuit board production defect detection method and system based on image recognition, and relates to the technical field of defect detection.During operation of the system, high-resolution equipment is configured, multi-mode collection is conducted on each production link of a PCB, comprehensive signal construction is conducted through fusion, and a composite signal Sc is obtained; inputting the collected composite signals into a feature calculation model, calculating to obtain defect feature measurement Fd, forming initial defect characterization, carrying out classified calculation on extracted feature values, obtaining a discrimination function value Cdef, converting a discrimination result into a space error parameter, calculating offset delta x and delta y of a welding spot relative to a design reference, and generating compensation quantity Qw by combining the defect feature measurement Fd; and driving the chip mounter or the welding head to carry out position adjustment or secondary welding, collecting detection data again after operation, verifying the correction precision, filing the defect characteristic value Fd, the discrimination function Cdef and the compensation amount Qw, and updating the defect database and the discrimination threshold value.

The invention relates to the technical field of safety monitoring and industrial automation, and discloses a silo safety monitoring method and system.The method comprises the steps that a laser scanning unit, an annular pressure sensing belt, a weighing sensor and a temperature and humidity node are arranged, and coal pile surface point cloud, lateral pressure, vertical load and environment data are synchronously obtained; a high-precision three-dimensional form is reconstructed based on an adaptive mesh refinement algorithm, and mechanical data are fused to construct a joint representation model; establishing a stability discrimination function in combination with coal physical parameters, and identifying an instability critical region; through mode matching with a historical stable form library, graded early warning and blockage clearing strategy generation of abnormal working conditions such as suspension arching, bonding and blockage are realized. According to the method, the laser point cloud, the pressure distribution and the load sensing data are fused, a combined characterization model of the three-dimensional form and the mechanical state of the coal pile is constructed, the limitation that traditional single-point material level measurement cannot reflect the whole coal pile structure is broken through, and high-precision measurement with the reserve calculation error smaller than 2% is achieved.

The invention discloses a method for predicting the service life of a relay for an operating mechanism based on a grey theoretical model, and relates to the technical field of a relay for a high-voltage circuit breaker operating mechanism of a power system, and the method comprises the following steps: simulating a salt mist aging test parameter test system through an electromagnetic relay, and simulating the operation environment of the relay for the operating mechanism; an alternating salt spray test is carried out, and characteristic parameters of the relay are extracted; performing dimension reduction processing on the characteristic parameters of the relay by using a principal component analysis method, and extracting digital characteristics of the characteristic parameters; on the basis of the digital characteristics of the characteristic parameters, the characteristic parameter with the maximum discriminant function value is selected by utilizing Fisher discriminant criteria; and taking the characteristic parameter with the maximum discrimination function value as a prediction variable, constructing a gray theoretical model, predicting the service life of the relay, and completing the service life prediction of the relay for the operating mechanism based on the gray theoretical model. The problem that the existing relay life prediction precision is low is solved.

The application relates to the technical field of computers and discloses a transportation track generation method and device, equipment and a storage medium, wherein GPS track data is obtained by acquiring road GPS point data and carrying out noise filtering processing; a state-action value function, a strategy function and a discrimination function are established based on the GPS track data and five elements of a Markov decision process; parameters of the state-action value function, the strategy function and the discrimination function are updated to construct a generated imitation reinforcement learning model; a meta-learning algorithm is used to train the generated imitation reinforcement learning model, and a transportation track is generated according to task-specific parameters obtained by training the meta parameters; the transportation track generation method and device, the terminal and the storage medium provided by the application can construct the generated imitation reinforcement learning model, improve the accuracy of the model, generate a city transportation track, and thus adapt to a complex travel environment.

The invention relates to a dysphagia recognition system and method based on computer vision, and relates to the technical field of computer vision, and the system comprises a data processing module which is used for collecting multi-modal data of a user and carrying out the normalization preprocessing, and obtaining the preprocessed multi-modal data; the feature fusion module is used for generating time sequence sensitive fusion features; the time sequence modeling module is used for performing dynamic time sequence modeling on the fusion features based on a recursive attention mechanism, capturing time sequence dependence of swallowing actions, and obtaining an attention distribution value as a time sequence modeling result; the classification detection module is used for performing anomaly detection and classification on the time sequence modeling result according to a dynamic threshold discrimination function to obtain a discrimination score and a classification result of the dysphagia; and the result identification module is used for carrying out comprehensive evaluation on the classification result through a multi-scale optimization integration algorithm, outputting evaluation indexes of the dysphagia and obtaining a dysphagia identification result. The system can improve the accuracy of dysphagia identification.

The present invention discloses a method for dynamic melt pool size detection and forming direction discrimination, comprising: collecting dynamic melt pool images, performing image segmentation, and generating a dynamic melt pool image source data set; generating a dynamic melt pool image data set; constructing a melt pool size detection convolutional neural network with a direction discrimination function, improving the loss function, adding a melt pool forming direction loss, and performing sufficient training to obtain a melt pool size detection convolutional neural network with a direction discrimination function; collecting real-time dynamic melt pool images, and using the trained melt pool size detection convolutional neural network to perform size detection and forming direction discrimination on the real-time dynamic melt pool images, thereby obtaining the length, width, and forming direction information of the melt pool morphology. The present invention designs a melt pool size detection convolutional neural network with direction discrimination, which improves the detection method of traditional target recognition neural networks and has the characteristics of fast processing speed, high recognition accuracy, and simple and fast detection method.

The invention discloses a deep neural network black box test method based on key region analysis, comprising the following steps: firstly, proposing a partition strategy to guide initialization of an image region and dynamic adjustment in an iteration process for an original input image; secondly, performing regional importance analysis on an input image by designing a discrimination function, and automatically positioning a key region on which a model decision depends; then, dynamically adjusting the initialized region, carrying out fine-grained analysis by utilizing a discriminant function, and carrying out iteration until the region is subdivided into a predefined size; and finally, based on the regional importance score, preferentially applying semantic preserving disturbance to the key region, and generating a targeted test sample through iterative query and feedback. According to the method, the decision dependency of the deep neural network is utilized, key region positioning can be achieved through model input and output, and compared with a traditional black box test method, the query frequency is effectively reduced; and a self-adaptive disturbance strategy is adopted, so that the sample semantic rationality is maintained while the test effectiveness is ensured. Meanwhile, the generated input image can be used for finely adjusting a target model, and the robustness of the model is improved.

The present invention discloses a real-time fire detection method and system for firefighting drones, wherein the method comprises: acquiring a target fire scene image in real time, and preprocessing the target fire scene image; discriminating the fire situation according to the pixel color space characteristics of the target fire scene image, and obtaining a first discrimination result; extracting the flame depth characteristics of the target fire scene image by using a YOLOV4 image recognition model, and discriminating the fire situation again based on the flame depth characteristics, and obtaining a second discrimination result; combining the first discrimination result and the second discrimination result according to a preset weight coefficient through a fusion discrimination function, and performing fusion discrimination to obtain a final fire incidence rate; the method can effectively reduce the false alarm rate and the missed alarm rate, thereby improving the efficiency of fire rescue.

The invention relates to the technical field of three-dimensional pipeline design, and discloses a three-dimensional design pipeline support hanger logic sequence automatic adjustment method and system, and the method comprises the steps: obtaining the logic sequence information and physical coordinate information of each pipeline in the modeling process of a three-dimensional pipeline model; based on the physical coordinates of the support points and the pipe fittings in the pipeline, the physical distances between the support points and the front and back adjacent pipe fittings in the logic sequence are calculated; based on the physical distance and a preset discrimination function, whether the physical position sequence and the logic sequence of each support point and the adjacent pipe fitting are consistent or not is judged; and if not, moving the support points based on the preset flow direction of the logic sequence through an iteration method until the logic sequence and the physical position sequence of each support point and the adjacent pipe fitting are consistent. According to the method, information is automatically obtained, the consistency of logic and physical sequences is judged, and the support points are automatically moved through the iteration method for adjustment, so that the pipeline design efficiency is improved, and the design period and the design quality are guaranteed.

The invention provides a vehicle navigation method and device, terminal equipment and a storage medium, and the method comprises the steps: building a discrimination function, and determining the state of a GNSS signal of a current vehicle through the function value of the discrimination function. When a GNSS signal is normal, GNSS information can be normally utilized to carry out GNSS / IMU fusion navigation. When a GNSS signal is poor, V2X information is introduced, and the V2X information and GNSS information jointly determine target measurement input, so that an SINS error is determined, SINS output is corrected, GNSS / V2X / IMU fusion navigation is completed, final navigation information is obtained, and accurate navigation of a vehicle when the satellite signal is poor is ensured. When the GNSS is unavailable, the target measurement input can be determined only based on the V2X information, the V2X / IMU fusion navigation is completed, the final navigation information is generated, and the navigation availability can be ensured to a certain extent when no satellite signal exists.

The present application relates to a binary function similarity detection method using hypergraph twin neural network, belonging to the field of computer and information science technology. The present application firstly constructs a directed heterogeneous hypergraph containing data flow and control flow hyperedges based on the abstract syntax tree of binary functions; then uses the improved hypergraph twin network to obtain hypergraph embedding vectors, and calculates the cosine distance as the similarity measure of function pairs, effectively learns the structure and semantic features of long-distance statements through multi-layer convolution, uses the initial residual connection and identity mapping to retain the initial features of the statements and avoid the performance degradation caused by the stacking of network layers; finally, the similarity measure is compared with the threshold to determine whether the function pairs are similar. The present application aims at the problem that the existing methods do not generate sufficient information representation of function semantics and structure, and it is difficult to accurately model the mutual influence between long-distance statements in the function, and proposes a binary function similarity detection method using hypergraph twin neural network, which improves the detection accuracy.

The invention relates to a high-voltage direct-current power transmission system commutation failure fault diagnosis method and system, and belongs to the technical field of power systems. According to the method, the decision binary tree in the fault diagnosis model is improved, and the support vector machine is used as a discrimination function in the decision binary tree, so that the discrimination result of the fault diagnosis model on the commutation failure reason type is finer and more accurate; according to the method, the time sequence data for the commutation failure fault diagnosis of the high-voltage direct-current power transmission system is obtained and input into the trained improved fault diagnosis model, the commutation failure reason of the high-voltage direct-current power transmission system is obtained, and the improved fault diagnosis model is relatively accurate in commutation fault diagnosis result; maintenance personnel can be helped to quickly locate the root of the problem, measures are taken in time to prevent fault expansion, and the negative influence of the commutation failure on the stability and reliability of a power grid is reduced.

This invention proposes a method for quality evaluation and multi-class discrimination of yam based on PCA and LDA, belonging to the technical field of agricultural product quality evaluation. The method includes the following steps: A) Constructing a discrimination model, specifically including A1) Data acquisition; A2) Data preprocessing and feature selection; A3) Constructing a linear discriminant model; B) Distinguishing yam categories, B1) Measuring 12 original indicator data of the sample to be tested according to the method described in A1); B2) Calculating the 12 original indicator data x... j Standardized to Z j B3) Z j Substitute the values ​​into the LDA discriminant function to calculate the values ​​for each type of yam; compare them, and the category corresponding to the maximum value is the discrimination result. When users apply the discrimination method of this invention, they do not need to understand the principle of PCA; they only need to perform simple standardization and a single linear combination calculation to complete the classification, which greatly reduces the threshold for using the model and the computational cost.

Methods, systems, apparatuses, devices, and computer program products are described. A mocking service may log a set of real application programming interface (API) data including a set of API requests and corresponding API responses. Using a generator function of a generative adversarial network (GAN), the mocking service may generate a set of sample (e.g., mock) data that mimics the API requests and responses. The mocking service may use a discriminator function to compare the sample data to the real API data and weight parameters of the GAN (e.g., retrain a machine learning model of the GAN) until the generator function generates sample data similar enough to the real API data. When the discriminator function is unable to distinguish the sample data, the real data, the mocking service may store the trained GAN and use it to generate mock API responses to API requests from users.

The application discloses a one-to-many framework-based marine plankton image classification and two-stage detection method, which comprises the following steps: a feature extraction module extracts an embedded feature vector of an input image by using a convolution prototype network; a one-to-many classification module based on distance converts a multi-class classification problem into multiple binary classification subtasks, constructs a discrimination function for each class, and calculates a probability estimation of a sample belonging to each class; and a posterior fusion and hierarchical detection module fuses the probability estimation based on evidence theory, generates a posterior probability distribution containing unknown classes, and sequentially judges whether a sample is an out-of-distribution sample and an out-of-capability boundary sample through a hierarchical decision mechanism, so that joint processing of known class recognition, out-of-distribution sample detection and out-of-capability boundary sample recognition is realized.