63 results about "Cost matrix" patented technology

The invention discloses a method for applying linear programming to CDN (Content Delivery Network) scheduling, which relates to the technical field of content delivery networks and comprises the steps of data preparation, strategy layer version smooth configuration, macroscopic layer and microscopic layer linear solution and online execution. Basic data are collected, cleaned and repaired, and a version change rule is set; the macroscopic layer constructs a linear programming model, and the cross-provincial bearing quota is solved with the aim of minimizing the cross-provincial cost; the micro layer takes the quota as a boundary and generates domain name class-node weight vectors in parallel; and adapting a routing request online through weighted rendezvous hashing and request features. According to the method, a dynamic cost matrix and a weight granularity control technology are integrated, the engineering problem of linear programming is solved, second-level response, approximate global optimal scheduling and accurate execution of floating-point-level weight are realized, memory overhead is reduced, smooth updating of a strategy and system stability are guaranteed, and CDN service quality and operation efficiency are improved.

The invention relates to a deep learning task resource allocation method and device, equipment and a medium. The method comprises the following steps: respectively analyzing a computational graph structure and a historical resource monitoring log corresponding to a deep learning task, and generating a tensor dependency graph and a resource use time sequence matrix; performing time-varying demand prediction on the basis of the matrix, generating a time-phased resource constraint table, performing memory allocation processing on the basis of a tensor dependency graph, and generating a tensor memory partitioning scheme and an inter-partition communication cost matrix; and performing static resource pre-allocation based on the time-phased resource constraint table and the tensor memory partitioning scheme, generating pre-allocated resource configuration, and performing resource scheduling and outputting real-time resource configuration through a deep reinforcement learning model according to the time-phased resource constraint table, the inter-partition communication cost matrix and the pre-allocated resource configuration. According to the method, by means of dynamic resource allocation, cross-partition communication cost optimization, reinforcement learning optimization and the like, the resource utilization rate and task execution efficiency of a deep learning task are remarkably improved.

The invention discloses a PC component BIM twinborn collaborative exchange method based on deep learning, and aims to solve the problems that unstructured multi-modal data of a site or a factory is difficult to automatically identify the identity, the state and the quality of a PC component and accurately map the unstructured multi-modal data with BIM twinborn objects one by one, and increment updating capable of realizing multi-party collaborative exchange is difficult to form. According to the method, time synchronization and coordinate calibration preprocessing is carried out on image data, video data and point cloud data, component instance recognition and feature extraction are carried out by using a deep learning model, and component state information and component quality information are generated; analyzing a BIM twinborn model to obtain priori features such as a unique identifier of a component, a component type, a size parameter, a spatial position and component geometry, performing cross-modal feature fusion under priori guidance, generating a candidate matching relationship, and constructing an optimal transmission cost matrix containing feature difference and constraint penalty to obtain a soft matching matrix; one-to-one mapping is obtained by adopting matching confidence rejection and assignment solution, an identification result and an evidence data index are written into a twinborn object, component-level difference is carried out on the twinborn object and a previous version to generate evidence increment exchange data, and the technical effects of high-reliability automatic updating and traceable cooperative exchange of the component-level twinborn model are achieved.

The invention relates to the technical field of computers. The target interruption trajectory association method based on the fusion strategy comprises the following steps: carrying out average distance calculation on a predicted position sequence and a position sequence of a new trajectory set to obtain an average distance value, and constructing a distance loss matrix based on the average distance value; performing feature parameter extraction processing on the historical trajectory and the new trajectory in the coarse correlation trajectory pair set to obtain a speed feature, an acceleration feature and an angular velocity feature; performing characteristic parameter loss calculation on the speed characteristic, the acceleration characteristic and the angular speed characteristic to generate a characteristic parameter loss matrix; performing weighted fusion processing on the distance loss matrix and the characteristic parameter loss matrix to generate an overall association cost matrix; and carrying out Hungary algorithm optimization processing on the overall association cost matrix, and outputting an association trajectory pair so as to achieve the technical effects of improving association robustness in a long-interruption scene, reducing the probability of misassociation of a high-maneuvering target and enhancing scene generalization ability.

This invention discloses a multilingual large-scale model dialogue optimization method and system integrating knowledge graphs, belonging to the fields of natural language processing and artificial intelligence dialogue technology. It addresses the problems of named entity recognition, time-varying attribute processing, and dynamic knowledge graph updates in cross-language dialogues. A multilingual named mention extractor scans each round of dialogue text word-by-word, forming a traceable mention index. Cross-language retrieval and candidate entity localization in the knowledge graph are performed using a multilingual pre-trained semantic model and alias matching paths. For time-varying attributes and title changes, priority graph adjustment is performed based on time slice windows and temporal evidence to ensure a smooth transition between old and new titles. The accuracy and stability of the final dialogue response are ensured by constructing a sparse cost matrix and an uncertainty re-discrimination process. This invention effectively solves the title switching problem in multi-context dialogue systems and improves the application effect of knowledge graphs in dialogue.

The invention provides a self-adaptive distribution system and method for GPU server load prediction, and the method comprises the steps: obtaining a monitoring data stream of a GPU server in real time, and constructing a monitoring sequence matrix according to the monitoring data stream; performing load mutation extraction on each sensitive mutation index in the monitoring sequence matrix to obtain a load pulse factor of each sensitive mutation index, constructing a multi-scale feature set, and generating a context representation vector sequence according to the multi-scale feature set; carrying out load prediction by using the context representation vector sequence to obtain a GPU load prediction sequence, and mapping the GPU load prediction sequence into a resource demand profile; and generating a resource migration cost matrix according to the current resource state information and the resource demand profile, and generating a resource allocation optimization strategy based on the resource migration cost matrix. According to the technical scheme provided by the invention, the abrupt change mode of the load can be accurately captured and represented, and the allocation decision is optimized by coupling the resource migration cost, so that the overall stability of self-adaptive resource allocation is realized.

The invention discloses an intelligent work order scheduling method and system based on multi-factor cost prediction, and belongs to the technical field of work order management. According to the method, a multi-factor dynamic cost model comprehensively considering in-transit time, skill matching degree, work order priority and service time limit SLA risk is constructed by acquiring work order and personnel states in real time, and a cost index is calculated for each potential scheduling scheme. According to the system, a variable neighborhood search VNS optimization algorithm is adopted, and global solution is carried out on the basis of a cost matrix so as to find an optimal task allocation scheme with the lowest total cost. According to the method, the weight can be dynamically adjusted and optimized according to operation states such as real-time traffic and personnel load, self-adaptive intelligent decision making is realized, and finally, an optimization scheme is automatically distributed to a personnel terminal, so that the work order scheduling efficiency, the resource utilization rate and the SLA fulfillment rate are comprehensively improved.

The application is specifically an AI computing power efficiency increasing method and system based on heterogeneous resource pooling and dynamic scheduling, relates to the technical field of distributed AI computing power scheduling, and comprises the following steps: collecting bandwidth, time delay and packet loss rate data of a wide area network link between each regional node in real time, calculating a transmission cost value between any two nodes, and generating a cost matrix.In the application, a standard virtual computing power unit index is introduced, floating point operation capability, video memory bandwidth and operator compatibility coefficient are uniformly included in calculation, different architecture heterogeneous chips obtain unified dimension logical resource representation, thereby providing a quantitative basis for cross-brand task allocation, and the problem that the existing scheme cannot perform equivalent comparison on heterogeneous chips is solved.

The invention relates to the technical field of atlas modeling, in particular to an industrial chain atlas modeling method and system based on deep learning. Comprising the steps of performing time sequence decomposition on a downstream enterprise sales volume and a price sequence, extracting a demand fluctuation component and calculating a demand influence factor to obtain upstream supply-demand relationship strength; determining a supply and demand intensity index of the upstream enterprise; judging whether the connection weight exceeds a threshold value or not, if so, extracting a cost structure parameter and generating a dynamic adjustment factor, updating the simulation cost matrix, and obtaining a classified cost parameter set through a clustering algorithm; iteratively optimizing the set, outputting upstream cost structure details when a deviation from downstream market performance is less than a threshold value, and calculating a bargaining ability score based on a conduction effect path; and generating an adjustment vector, updating the edge weight of the industrial chain atlas, and outputting resource configuration scheme parameters when the trend is stable. According to the invention, intelligence and precision of industrial chain supply-demand relationship modeling, cost dynamic optimization and resource configuration are realized.

The invention discloses a multi-target tracking method and system based on dynamic weight and multilevel feature fusion, and the method comprises the steps: carrying out the target detection of a video frame, and obtaining a detection box, and the confidence and appearance features of the detection box; predicting the current position based on the existing trajectory, and obtaining a multi-level appearance feature library which is stored in a hierarchical manner according to confidence; calculating position cost and appearance cost between the detection frame and the trajectory, dynamically fusing weights of the detection frame and the trajectory according to a continuous tracking state of the trajectory, and introducing a penalty term based on confidence of the detection frame to generate a comprehensive association cost matrix; matching is carried out, and the state of the successfully matched track and the feature library are updated; and managing the life cycle of the unmatched track and the detection frame. The method adapts to different tracking scenes through dynamic weight adjustment, the appearance information of each confidence detection frame is fully utilized through the multi-level feature library, and the high and low confidence detection frames are associated at the same time through the global optimization strategy, so that the tracking accuracy, robustness and continuity in a complex scene are remarkably improved.

This invention relates to a self-supervised 3D particle tracking and velocity measurement method, comprising the following steps: S1, acquiring a continuous temporal sequence of 3D source particle sets and 3D target particle sets, and performing feature encoding to obtain the feature matrix of the corresponding particle sets; S2, inputting the feature matrices of the source particle sets and target particle sets into a DFCT, and outputting the aligned cross-frame features; S3, constructing a transmission cost matrix between particles, establishing a dense soft correspondence between particles in two frames, and outputting the initial flow field estimation result and matching confidence; S4, constructing a composite self-supervised loss function, and iteratively optimizing the model parameters by minimizing the composite self-supervised loss function; S5, refining the initial flow field estimation result, and outputting the final 3D fluid velocity field. The beneficial effects of this invention are: solving the problems of algorithm dependence on large-scale high-quality labeled data, low efficiency in extracting semantic features from complex flow field point clouds, and matching ambiguity in high-displacement, high-density scenarios.

This invention discloses a training method for a reward model, an optimization method for a large language model, and related methods. The training method for the reward model includes: obtaining preference training sample pairs and a reward model to be trained, wherein the preference training sample pairs include preferred response samples and non-preferred response samples; calculating the reward score difference between the preferred response samples and non-preferred response samples based on the reward model to be trained; constructing a cost matrix based on the reward score difference and the semantic correlation between the preferred response samples and non-preferred response samples; calculating the loss margin based on the cost matrix; calculating the pairwise preference loss value with the margin based on the loss margin, and updating the parameters of the reward model to be trained with the goal of minimizing the loss value with the margin, thereby obtaining a trained reward model. This improves the model's learning ability on difficult samples and its overall generalization performance, avoids over-reliance on simple samples, and thus improves the generation quality of large language models in complex tasks.

The invention discloses a multi-target tracking method based on a self-adaptive iterative expansion intersection-to-union ratio and a related device, and belongs to the technical field of computer vision and graphics. The method follows the normal form of tracking after detection, firstly, independent target detection is carried out on each frame of image, and then a motion prediction module based on training during testing is used for predicting the position coordinates of a target in the next frame. And meanwhile, calculating a predicted position offset mean value, if the predicted position offset mean value is greater than a threshold value, constructing a cost matrix by adopting a self-adaptive iteration expansion intersection-union ratio method and utilizing an expansion intersection-union ratio between the predicted position and the detection frame and appearance characteristics, and solving the cost matrix by utilizing a Hungary algorithm so as to realize association of nonlinear motion targets between two adjacent frames. Experiments prove that the method has the best performance on two large-scale common nonlinear data sets, and irregular nonlinear motion of athletes in the tracking process can be effectively processed.

The invention discloses a method and a system for automatically generating a state sequence script based on an improved algorithm, and the method and the system are used for automatically generating the state sequence script in an XML (Extensible Markup Language) format as a test case for automatic testing. The matching priority of the state nodes is evaluated by defining a heuristic function, so that the matching process of the state nodes is optimized; a state transition path optimization step: adopting a dynamic programming algorithm to calculate an optimal solution of a state transition path by constructing a state transition cost matrix and applying a recursion formula so as to optimize the selection of the state transition path; and a step of generating a state sequence script in an XML format, namely automatically generating the state sequence script conforming to the XML format according to the optimized state conversion path matching result. The test case generated by the method is more accurate and comprehensive, and the test quality can be improved.

The invention provides a pipeline internal detection defect data alignment method, and relates to the technical field of pipeline data, and the method comprises the steps: obtaining a defect feature list; calculating a matching cost based on the axial coordinate and the angle parameter; combining defect features in the first defect feature list and the second defect feature list, and generating a cost matrix; solving the cost matrix to generate an alignment result of the defect features; dividing a rectangular bounding box based on defect features; if the rectangular bounding box crosses the boundary, segmenting the rectangular bounding box according to the boundary to form rectangular sub-frames; calculating the intersection-to-union ratio; if the intersection-to-union ratio is greater than a preset intersection-to-union ratio threshold, adding the first rectangular sub-frame and the second rectangular sub-frame into a target candidate set; constructing a bipartite graph based on the target matching set; and based on the bipartite graph, matching the first rectangular sub-frame with the second rectangular sub-frame, and generating a target alignment result, so as to solve the problems that the alignment rate of the detection defect data in the current pipeline is low and the cross-pipe-joint matching capability is lacked.

The invention relates to the technical field of centrifugal machine feeding control, in particular to an automatic feeding control method of a horizontal scraper discharging centrifugal machine. The method comprises the steps of firstly extracting a torque event queue and a vibration event queue; further constructing an initial cost matrix based on time sequence and morphological similarity, storing the time delay and energy gain of the optimal matching pair into a sample pool, and further judging whether intervention is started and obtaining reference time delay and reference energy gain according to discreteness of data in the sample pool; further analyzing the response of the vibration event to the torque event in combination with the reference time delay and the reference energy gain after the intervention is started, and constructing a matching cost matrix; further comparing the optimal cost and the random expected cost in the matching cost matrix to obtain a causal coupling index; finally, feeding is adjusted based on the causal coupling index, real-time diagnosis of the filter cake structure state and feeding self-adaptive regulation and control are achieved, and production safety is guaranteed.

The application provides a pairing method and device for non-fixed additional pseudo codes, wherein the method comprises the following steps: setting pairing indexes, the first index being the aligned cross-correlation value absolute value, and the second index being the absolute value of the difference between the leading and lagging correlation values; determining a target function and a cost matrix based on the first index and the second index; establishing an additional pseudo code allocation model based on the cost matrix and an allocation matrix; dividing the additional pseudo codes into two parts, determining an initial population based on the division result, and solving the model by using a JVC algorithm to determine the initial pairing scheme corresponding to the individuals in the initial population; performing iterative updating on the population based on a meta-heuristic algorithm or an improved meta-heuristic algorithm, and performing iterative updating on the pairing scheme corresponding to the individuals based on the JVC algorithm until a preset maximum number of iterations is reached, and determining the optimal pairing scheme obtained in the last iteration process as the optimal pairing scheme of the non-fixed additional pseudo codes; wherein the optimal pairing scheme obtained in the iteration process is determined based on the target function.

The application discloses a task dynamic scheduling method and device based on a time-sensitive rule engine, a storage medium and computer equipment, and the method comprises the following steps: determining time priority data of a plurality of to-be-processed tasks through a time-sensitive rule engine; determining a plurality of candidate processors, constructing a cost matrix containing task allocation costs of allocating each to-be-processed task to each candidate processor, and constructing a priority matrix corresponding to the cost matrix according to each task allocation cost in the cost matrix; reducing the cost matrix to update the cost matrix, performing initial task matching on the to-be-processed tasks based on the cost matrix; finding an augmented path for an unmatched task according to the priority matrix, calculating a task allocation cost adjustment amount based on the augmented path, adjusting each task allocation cost in the cost matrix according to the allocation cost adjustment amount, continuing to find an augmented path in the next round, and obtaining a feasible path; and determining the processor of the unmatched task according to the feasible path.

A system filters near-duplicate images to generate data for training or validation of a machine learning model. The system receives a set of images and generates feature vectors from the images. The system clusters the feature vectors. For each cluster of feature vectors, the system determines near-duplicate pairs of images. The system may generate a cost matrix representing a linear assignment problem and find near-duplicate pairs of images by solving the linear assignment problem. The system filters images from the set of images based on the near-duplicate pairs of images. The system uses the filtered set of images for training or validation of the machine learning model.

The application relates to a multi-to-multi task allocation method based on minimum zero-control miss distance. The method comprises the following steps: establishing a three-dimensional missile-target relative motion equation, deducing an analytical expression of the zero-control miss distance and a dynamic equation; calculating the zero-control miss distance value between each interceptor missile and a target according to the initial condition of the terminal guidance and the analytical expression of the zero-control miss distance; constructing an intermediate matrix according to a pre-set interception task, filling a cost matrix of a Hungarian algorithm into the intermediate matrix as matrix elements to obtain a unified cost matrix; substituting the zero-control miss distance value into the unified cost matrix to generate a multi-to-multi task allocation scheme based on the Hungarian algorithm; and designing an improved real true proportional guidance law for controlling the zero-control miss distance, and adopting the improved real true proportional guidance law to realize the multi-to-multi task allocation scheme. The method can effectively eliminate the zero-control miss distance and increase the interception success rate.

The invention provides a pipeline circumferential weld data alignment method, and relates to the technical field of pipeline data, and the method comprises the steps: obtaining target circumferential weld data of a to-be-detected pipeline; determining a target short pipe cluster based on the target circumferential weld data; extracting target short tube cluster features; constructing a first comprehensive cost matrix based on the target short tube cluster features; processing the first comprehensive cost matrix to obtain an initial short tube cluster pair; verifying the initial short tube cluster pair, and determining a primary anchor point; determining an initial secondary anchor point based on the primary anchor point; extracting initial secondary anchor point features and determining a target secondary anchor point; dividing the to-be-measured pipeline into a second-level interval based on the first-level anchor point and the second-level anchor point; first pipe joint features in the second-level interval are extracted; and on the basis of the first pipe joint characteristics, a target matching result is determined, and alignment processing is carried out on the circumferential weld data, so that the problem that the accumulative error of the circumferential weld position is large due to the influence of uncertain factors such as equipment performance, environmental conditions and measurement errors at present is solved.

The invention discloses an intelligent control method and system for a switch integrated test, and belongs to the technical field of switch equipment test control. Comprising the steps of monitoring a switch state, performing classification test according to a matching degree of a state required by a project and a current state, and forming a preliminary sequence by combining time consumption, logic relevance and switching direction sorting; constructing a state-test directed graph, calculating an edge weight, and then obtaining a first optimization path which focuses on time consumption and resources through iteration and tracing; defining a state transition rule and a cost function, and iteratively updating a cost matrix to obtain a second optimization path considering switch wear; comparing the two paths, extracting a consistent section, calculating a difference value between a weight value and a cost value of a difference section, and performing optimal selection and splicing to form a final path; the intelligent and scientific test classification and sorting are realized, the multi-dimensional targets of efficiency, energy consumption and equipment protection are balanced, the path adaptability is enhanced, and a scientific decision basis is provided for test planning.

The invention discloses a method and system for extracting target data in a database, and the method comprises the following steps: collecting information in the database, and constructing a database mode graph structure; constructing an enhanced GW-OT model, and initializing a basic cost matrix and a marginal distribution set; in the structure consistency propagation unit, performing consistency propagation calculation; in the cost matrix construction unit, an enhanced cost matrix is constructed; in the optimal transmission solving unit, optimal transmission optimization calculation is executed; in the candidate data generation unit, cross-table and cross-mode matching alignment is carried out; and in the target data extraction unit, performing condition filtering and discrimination, and outputting target data. According to the method, the accuracy and consistency of database target data extraction are improved.

The embodiment of the present discloses a method for training a policy model based on reinforcement learning and apparatus, and an electronic device. The method includes: determining an observation data sequence generated by an agent during a time period of executing a target task, a preset time window, and an expert data sequence corresponding to the target task; determining a cost matrix and a transport matrix based on the preset time window, the observation data sequence, and the expert data sequence; determining a reward value for each observation data in the observation data sequence based on the cost matrix and the transport matrix; and training the policy model based on the observation data and the reward value of the observation data.

The application relates to the medical field and can be applied to doctor-patient intelligent question answering, and discloses a text similarity discrimination method, wherein the method comprises the following steps: constructing a cost matrix according to token embedding of two to-be-discriminated sentences; constructing a first mathematical relationship corresponding to a total cost, a cost matrix, an optimal transmission matrix and an optimal transmission distance in a transmission process of the two to-be-discriminated sentences; defining state vectors of the two to-be-discriminated sentences as all-1 vectors normalized by lengths of the own sentences respectively; the all-1 vectors comprise a first constraint condition and a second constraint condition; removing any one of the first constraint condition and the second constraint condition in the all-1 vectors to obtain a relaxed equation; combining the relaxed equation with the first mathematical relationship to derive a relaxed optimal transmission distance, and utilizing the relaxed optimal transmission distance to discriminate the similarity of the two to-be-discriminated sentences, so that the text similarity can be discriminated simply and accurately.

The application provides a pipeline internal detection defect data alignment method, relates to the technical field of pipeline data, and comprises the following steps: acquiring a defect feature list; calculating a matching cost cost based on an axial coordinate and an angle parameter; combining defect features in the first defect feature list and the second defect feature list and generating a cost matrix; solving the cost matrix to generate an alignment result of the defect features; dividing a rectangular bounding box based on the defect features; if the rectangular bounding box crosses a boundary, the rectangular bounding box is cut by the boundary to form a rectangular sub-box; calculating an intersection-over-union ratio; if the intersection-over-union ratio is greater than a preset intersection-over-union ratio threshold, the first rectangular sub-box and the second rectangular sub-box are added to a target candidate set; constructing a bipartite graph based on the target matching set; and matching the first rectangular sub-box and the second rectangular sub-box based on the bipartite graph to generate a target alignment result, so as to solve the problems of low alignment rate of the current pipeline internal detection defect data and missing cross-pipeline section matching capability.

The invention discloses a knowledge distillation-based large model lightweight deployment method, which comprises the following steps of constructing a teacher model and a student model according to training data and target hardware resources, and determining a student model structure capacity vector and a distillation layer set; extracting teacher features and combining the structure capacity vector to construct a joint cost matrix considering semantic difference and resource cost at the same time; establishing an entropy regular optimal transmission model by taking teacher characteristics and structural capacity as edge constraints and taking the joint cost matrix as transmission cost, and solving an approximate transmission matrix by adopting a Greenkhorn algorithm under the constraints of calculation power and storage budget; and generating a student model structure reconstruction scheme by using the approximate transmission matrix and reconstructing a student model structure, and obtaining a lightweight student model meeting resource constraints based on combined training of distillation loss and task loss. According to the method, the large model deployment efficiency under limited hardware is improved.

The application discloses a kind of press bury space multi-agent multi-task planning and distribution method and system, it is related to emergency rescue task planning and distribution technical field, solve the technical problem that press bury space rescue cost and actual cost adaptation degree is lower and rescue scheme acquisition efficiency is lower, its technical scheme key point is by matrix verification function to the task-agent cost matrix Intelligent cutting, automatically delete the row and column that can not be executed, precluded a large number of invalid search space, substantially reduce the computational complexity, meet the real-time requirement of emergency response.Simultaneously in solving process, by dynamically deleting the row and column corresponding to the task and agent that has been allocated, it is not necessary to artificially adjust second time, can be directly used as execution basis.Simultaneously by the mechanism that the size of effective cost matrix scale dynamically selects solving path, the optimal balance of solving effect and efficiency is realized, overcome the defect that single algorithm is difficult to adapt to different emergency scene scale.

The invention discloses a method and system for realizing multi-target long and short term tracking based on frequency domain information of Fourier transform, and the method comprises the steps: extracting a detection frame of an effective target and a tracking trajectory corresponding to the effective target based on an input video sequence, constructing a time position matrix, obtaining a prediction position of the effective target through frequency domain analysis, and obtaining a prediction result. After a historical appearance frequency domain feature sequence of an effective target is obtained, a historical feature library of the effective target is constructed, a mean value feature is obtained based on the historical feature library, similarity identification is carried out by using appearance frequency domain features of a candidate detection frame of the effective target and the mean value feature, and a prediction detection frame of the effective target is obtained. And constructing a cost matrix of a prediction position and a prediction detection frame, finding an optimal prediction detection frame, and carrying out weighted fusion on the position of the optimal prediction detection frame of the effective target and the prediction position to obtain a correction position of the effective target. According to the method, the motion and appearance characteristics of multiple targets are processed at the same time under the unified frequency domain representation framework, and the robustness and precision of multi-target tracking are improved.

The invention discloses a loss function determination method, a defect identification method and related equipment, and relates to the field of industrial defect detection.The method comprises the steps that a transfer cost matrix is constructed according to the attention of a target user on different defect categories; obtaining a cross entropy loss function; and determining a target loss function based on the transfer cost matrix and the cross entropy loss function.