144results about How to "Reduce calculation" patented technology

The invention belongs to the technical field of model training, and particularly relates to a large model increment training method and system based on dynamic sparsification. The method comprises the following steps: S1, evaluating the parameter importance of a large model by adopting a multi-dimensional evaluation method, and reflecting the contribution of parameters to the performance of the large model; s2, performing dynamic sparsification control on the large model structure by adopting a self-adaptive structure adjustment method; s3, performing incremental training management on the large model by constructing a capacity-adjustable empirical playback buffer pool and proposing a multi-objective loss function; and S4, carrying out model compression optimization on the large model by adopting a mixed precision quantification strategy. The method has the characteristics that the network structure can be dynamically adjusted by evaluating the importance of model parameters in real time, so that the efficient training is realized while the model performance is ensured.

The invention discloses an electric power system state estimation method based on graph contrast learning. The method comprises the following steps: constructing graph sequence data of electric power system state estimation; constructing an electric power system state estimation model based on graph contrast learning, and putting the graph sequence data into the electric power system state estimation model based on graph contrast learning; inputting the node feature sequence and the masked edge features into a hybrid encoder of a dynamic edge convolutional neural network combining a self-attention mechanism and a residual block and a graph convolutional neural network with a dynamic propagation strategy of a gating mechanism to comprehensively process the features of nodes and edges; comparing an updating mechanism and a sampling symmetry training strategy of model parameters of the learning framework through a graph; and calculating a node voltage phase angle by using the branch phase angle difference, and outputting a state estimation result. According to the method, the internal structure and mode of the power system data can be effectively learned by combining graph contrast learning and a deep neural network technology, and the capability of identifying and filtering abnormal data is improved.

The invention provides a multi-degraded image restoration method based on self-adaptive prompt. The method comprises a two-stage hidden space diffusion model, and various degraded images can be flexibly processed. Specifically, the model generates a degradation cue by learning the difference between a degraded image and a clean image, and encodes the cue information in a hidden space. In the first stage, a degradation prompt is obtained through a degradation adaptive prompt learning mechanism as a key feature for guiding the diffusion model in the second stage to carry out image restoration. Finally, prompt information is integrated into a Fourier transform restoration network, and accurate restoration of different types of degraded images can be realized. The method not only improves the accuracy of image restoration, but also significantly reduces the calculation complexity, and is suitable for image restoration tasks in various real world scenes, including denoising, deblurring, defogging, rain removing, snow removing, raindrop removing and low illumination enhancement.

The invention discloses an unmanned aerial vehicle target detection method based on feature fusion and detection head collaborative optimization, and the method comprises the steps: obtaining a multi-scale feature map through employing a backbone network formed by stacking a convolution module CBS and a residual module C2f, sequentially transmitting the feature maps of various scales to a multi-level feature fusion module (ML-FFM), carrying out the feature fusion from a deep layer to a shallow layer, and carrying out the detection of a target target. Semantic information of shallow layer features is enhanced; inputting the fused feature map into a refined feature enhancement module (RFEM), eliminating position conflicts and redundant information based on a space-time attention mechanism, and optimizing details and positioning capability of deep features; and the enhanced feature map is sent to a prediction head based on Transform for detection, and target classification and positioning of the aerial image of the unmanned aerial vehicle are completed. According to the method, the problems that the unmanned aerial vehicle aerial image is large in scale change, and small targets are high in proportion and dense in distribution are solved, so that the precision of target detection in the unmanned aerial vehicle aerial image is improved. The modules proposed in the method can significantly reduce the calculation amount, and are suitable for the real-time detection scene of the unmanned aerial vehicle.

The invention relates to the technical field of image recognition, in particular to a circuit board defect detection method and device, equipment and a storage medium, and the method comprises the steps: obtaining a to-be-detected image, and carrying out the preprocessing of the to-be-detected image, and obtaining a preprocessed image; inputting the preprocessed image into a pre-constructed defect detection model, performing feature extraction processing on the preprocessed image based on a pre-confirmed convolution weight, performing fusion processing on a plurality of feature maps of different scales based on a pre-confirmed fusion weight, and performing defect detection on the fused feature map by adopting a detection head to obtain a defect detection result; the defect detection method disclosed by the invention aims to solve the problems of high calculation complexity, poor real-time performance and insensitivity to small target detection in the existing PCB defect detection method. The convolution weight is adjusted by introducing a dynamic convolution selection mechanism, and it is ensured that the model has high precision when processing small defects; and the fusion weight is adjusted by introducing a weight adaptive mechanism, so that the calculation of redundant information can be reduced, and the detection efficiency is greatly improved.

The invention belongs to the technical field of unmanned aerial vehicles, and particularly provides a transformer substation unmanned aerial vehicle autonomous inspection route generation method based on an optimization A star algorithm. According to the method, a heuristic function is improved, the Euclidean distance between a target point and a current node is comprehensively considered, obstacle avoidance processing is carried out, and a safer and more efficient inspection route is generated. And by calculating the total cost of each node, the node with the minimum total cost is selected for expansion until a target node is found. According to the improved A star algorithm, the unmanned aerial vehicle can be effectively prevented from flying into an obstacle dense area, the obstacle keeps a safe distance, and the flying safety is remarkably improved. The distance and the obstacle density are comprehensively considered, the generated inspection route is smoother and more efficient, and unnecessary flight distance and energy consumption are reduced.

The invention discloses an unmanned aerial vehicle assisted Internet of Vehicles cross-domain authentication method based on a block chain and federated learning, and the method comprises the steps: initializing an Internet of Vehicles system, and generating system initialization common parameters; the vehicle and the unmanned aerial vehicle complete identity registration with the control center, and the vehicle performs login verification; and respectively carrying out initial authentication and cross-domain authentication on the vehicles which are successfully logged in and verified, guaranteeing the communication security through key negotiation, and completing the cross-domain authentication of the unmanned aerial vehicle assisted Internet of Vehicles. According to the method, the reliability and the security of data in the cross-domain authentication process are realized, the privacy of the data is improved, relatively low calculation and transmission overhead in the cross-domain authentication process is realized, and the authentication efficiency is improved.

The invention discloses a transformer substation scene three-dimensional semantic segmentation method based on combination of convolution and attention, and the method comprises the following steps: S1, carrying out the region and category division of three-dimensional point cloud data in a transformer substation scene, setting semantic tags for different categories, dividing a training set and a test set according to the region, and forming a three-dimensional semantic segmentation data set; s2, constructing a three-dimensional semantic segmentation model based on a convolution and attention mechanism combination module, wherein the three-dimensional semantic segmentation model is used for predicting a semantic tag of point cloud data in a substation data set; s3, determining a loss function and a parameter optimizer, and training a three-dimensional semantic segmentation model in combination with the three-dimensional semantic segmentation data set; and S4, performing semantic segmentation on the point cloud data in the substation scene based on the trained three-dimensional semantic segmentation model. Compared with an existing three-dimensional semantic segmentation method, the accuracy of three-dimensional semantic segmentation can be effectively improved, and accurate semantic segmentation of the substation scene point cloud data is achieved.

The embodiment of the invention provides a training method and device of a text abstract generation model, equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring a sub-word index sequence of a text sample; processing the sub-word index sequence to obtain a coding tensor sequence; according to the coding tensor sequence, carrying out internal loop training and external loop training on a preset TTT-Learar-Transformer model, and obtaining a target loss function; according to the target loss function, weight updating is carried out on the TTT-Learar-Transform model in a dual form until the TTT-Learar-Transform model converges, a text abstract generation model is obtained, and the text abstract generation model is used for generating an abstract of a text. According to the embodiment of the invention, the performance of the text abstract generation model can be improved, and the text abstract generation quality and efficiency are improved.

The invention provides a fuzzy privacy set intersection method and system, a medium and electronic equipment, and the method comprises the steps: carrying out the encryption processing of an initial data set X through a homomorphic encryption algorithm, obtaining first data, and obtaining a processed data structure D through a defined space Hash algorithm; the receiver performs homomorphic addition or scalar multiplication operation processing on the first data by using own data Y, and performs confusion processing by using a random number R to obtain second data; after the sender carries out Augment algorithm processing on the random number, a Hash algorithm is carried out to obtain a corresponding Hash value, and the Hash value is recorded as third data to be sent to the sender; and the sender decrypts the second data, processes the second data through an Augment algorithm, executes a PSI protocol with the receiver to obtain fourth data, and compares the third data with the fourth data to obtain intersection data. According to the method, an addition homomorphic technology is adopted for a novel application scene of intersection of the fuzzy privacy set, and compared with a full homomorphic technology in the prior art, the time consumption is less, and the communication is small.

The invention discloses a rapid batch switching authentication method for predictability and aggregation promise proof, which realizes batch authentication of a plurality of user terminals by generating system parameters and secret keys through a trust mechanism and combining predictability promise and aggregation signature technologies. In the authentication process, the edge base station to be switched exchanges with the current communication edge base station through an encrypted message, verification information such as a pre-authentication certificate and a signature is generated through the user terminal, and identity authentication is completed step by step. The method comprises the following specific steps: a trust mechanism generates a private key, a public key, a pseudo identity and an expiration timestamp for each device; through multiple rounds of message exchange, the edge base station and the user terminal perform identity verification by using a shared key and a hash function; and finally carrying out efficient verification on a plurality of users through batch signature and aggregation commitment. According to the method, on the basis of ensuring the security, the speed of switching the authentication process is greatly improved, and the method is suitable for a large-scale network environment, especially in edge computing and mobile communication networks.

The invention discloses a virtual power plant optimal scheduling method and device, a medium and a product, and relates to the technical field of power plant electric energy resource scheduling and allocation, and the method comprises the steps: obtaining basic data of a virtual power plant; respectively carrying out clustering analysis on the temperature control load data and the electric vehicle data by adopting a density clustering algorithm and a K-means + + algorithm based on an HHACOA algorithm to obtain a clustering result; according to the basic data of the virtual power plant, an internal approximation constraint space integration method is adopted, and adjustable power domain data of the whole temperature control load cluster and adjustable power domain data of the whole electric vehicle cluster are determined; establishing a planning function optimization model according to the adjustable power domain data of the temperature control load cluster, the adjustable power domain data of the electric vehicle cluster and the clustering result; and solving the planning function optimization model to obtain an optimization strategy result of the virtual power plant. The method can improve the adjustment capability of the virtual power plant to the power grid, and guarantees the safety and stability of the power system.

The invention discloses a method for realizing capacitor voltage balance of a four-level energy storage converter under a low modulation ratio condition, and the method comprises the steps: converting a space vector diagram of the four-level energy storage converter under an alpha-beta coordinate system into a g-h coordinate system at a low modulation ratio, and obtaining three nearest vectors for synthesizing a reference voltage vector Vref and duty ratios of the three nearest vectors; therefore, the sector judgment process is simplified, and the calculation process of the duty ratio acted by the basic voltage vector is reduced. Then, according to the small triangle where the reference voltage vector Vref is located and the duty ratio of the action of the three vectors for synthesizing the reference voltage vector Vref, a five-segment type modulation synthesis mode and a seven-segment type modulation synthesis mode are adopted, and a switch sequence corresponding to each small sector is obtained; and finally, the switching sequence corresponding to the small sector is evaluated through the target function, and the switching sequence when the target function is the minimum value is selected as the optimal switching sequence by using comparison operation. And capacitor voltage balance of the four-level energy storage converter is realized. The technical problem that an existing additional auxiliary circuit balancing method is poor in practicability can be solved.

The invention discloses a federal recommendation privacy protection method based on ring signature and selective aggregation, which realizes anonymization privacy protection and selective aggregation in a federal learning process through an innovative double-server architecture and a hierarchical security mechanism. The method comprises the following steps: 1) deploying an aggregation server and a security server in a system initialization stage, and completing parameter initialization and key distribution; 2) the client executes local model training, and anonymization privacy protection uploading is realized by adopting a ring signature technology; 3) the security server performs selective aggregation of credibility verification, and completes grouping and parameter fusion based on user similarity; and 4) ensuring the completeness and source credibility of model updating through a hierarchical security parameter distribution mechanism. While the recommendation precision is ensured, the user identity leakage and privacy tracking are effectively prevented, the comprehensive security guarantee is provided for the federal recommendation system, and the method has the advantages of low calculation overhead and high communication efficiency.

The invention relates to the technical field of micro-grids, and discloses an AC micro-grid distributed power control method with quantitative event triggering. According to the method, due to droop control in an island DC micro-grid, distribution of reactive power and unbalanced power in a system cannot be guaranteed, and a distributed control method with quantitative event triggering is provided for solving the problem. According to the control scheme provided by the invention, communication is reduced on the basis of realizing reactive power and unbalanced power distribution among distributed power generation units by adding a designed distributed secondary controller with a quantitative event triggering method, and parts which cannot be accurately distributed can be readjusted by a negative feedback system through PI control; in addition, the proposed method does not need global power information, each distributed generator only communicates with a neighbor when meeting a triggering condition, the communication burden of a micro-grid communication network is greatly reduced, the stability of the method is proved through a Russell invariant set principle, and Zeno behaviors are also eliminated through an anti-security method.

The invention relates to the technical field of data security and privacy protection, in particular to a block chain and edge computing-based medical data sharing and access control platform, which comprises a multi-layer data storage system, a block chain network, an intelligent contract module, a plurality of data consumers, a plurality of data providers and a plurality of hospital servers. According to the invention, the calculation and communication overhead in data transmission of the medical internet of things is reduced, and the sharing efficiency and security requirements of the medical data between a provider and a consumer can be met.

The invention discloses a method for controlling running of a networked vehicle at a non-signal-lamp intersection by considering an input constraint, which solves the challenge confronted when an automatic driving vehicle safely and cooperatively passes through the non-signal-lamp intersection under the input constraint, and comprises the following steps: obtaining real-time state information and roadside environment information of the vehicle; establishing a vehicle dynamics model considering input constraints; a geometric topological structure is developed to model the passing relation of vehicles, then the communication topology of information exchange between the vehicles is defined, and the problem of vehicle coordination at the intersection is converted into the problem of consistency of conflicting vehicles through state conversion; a distributed control method is designed, and the global consistency and stability of the vehicle cooperative control system are ensured; and finally, a time-based triggering mechanism is provided, so that the vehicle passing efficiency and the driving safety are improved, and the operation load of the system is reduced.

The invention provides a large-array intelligent metasurface codebook optimization and switching method and device, and belongs to the technical field of wireless communication. Aiming at the problem of low efficiency of a traditional codebook selection method in a high-speed moving scene of a user, the invention provides an intelligent optimization method and device, which can efficiently position optimal codebook configuration in a codebook library through a quick search mechanism. According to the method, in combination with signal-to-noise ratio feedback driving, multiple groups of configurations in a codebook library are tested step by step, and a gain optimization module is combined, so that the optimal beam forming effect is achieved, and the method does not need to depend on accurate position information of a user. Besides, a clockwise adjacent codebook test strategy is introduced, the redundancy calculation and traversal time is shortened, and it is ensured that in the user moving process, codebooks can be rapidly switched, and stable communication quality is kept. The algorithm has the advantages of real-time response and complexity reduction, is suitable for 5G and other high-speed mobile communication scenes, and effectively improves the signal transmission efficiency and the communication reliability.

The embodiment of the invention provides attack detection equipment, method and device and a medium, which are used for solving the problems that a single CPU core cannot be expanded, more computing resources are occupied and the processing efficiency is low in the prior art. For each NetFlow flow, sending the NetFlow flow to a corresponding target flow statistics CPU core according to the source IP and the destination IP in the quintuple of the NetFlow flow and the stored IP address range processed by each flow statistics CPU core; each flow statistics CPU core counts the message number and the byte number of the NetFlow flow of each IP address in the IP address range, carries out DDoS detection, directly completes NetFlow flow summarization and DDoS detection of the same IP address in the flow statistics CPU core, does not need to be transmitted to a single CPU core for summarization, reduces the occupation of computing resources, improves the processing efficiency, and can also carry out core expansion.

The invention provides a dynamic look-around splicing method and system based on tire pressure monitoring, and the method comprises the steps: carrying out the quantitative processing of a tire pressure value according to the preset vehicle information of a target vehicle, so as to obtain a tire pressure data segment array of the target vehicle; the tire pressure data segment array comprises a plurality of tire pressure data segments; according to each tire pressure data segment, respectively constructing a mapping relation table of each tire pressure data segment to obtain a total mapping relation table; performing matching search according to the real-time tire pressure data of the target vehicle and the total mapping relation table to obtain a target mapping relation table of the real-time tire pressure data; and performing all-round stitching of the target vehicle according to the target mapping relation table. By quantifying the tire pressure value in the tire pressure monitoring system, the calculation amount of the Map mapping lookup table caused by frequent change of the tire pressure value is reduced, and the calculation amount of subsequently establishing the tire pressure value and the Map mapping lookup table is also reduced.

The invention relates to a space semantic occupation prediction method and device, computer equipment, a readable storage medium and a program product. The method comprises the following steps: acquiring original point cloud data in a target space scene; converting the original point cloud data into a multi-view feature plane, and performing feature extraction on the multi-view feature plane to obtain multi-view point cloud features; inputting the multi-view point cloud features into a pre-trained visual generative model to obtain a two-dimensional semantic image, and performing feature extraction on the two-dimensional semantic image to obtain multi-view image features; fusing the multi-view point cloud features and the multi-view image features to obtain multi-view fusion features; and for any query point in the target space scene, according to the multi-view fusion feature corresponding to the query point, outputting a space semantic occupation prediction result. By adopting the method, the efficiency of space semantic occupation prediction can be improved.

The invention relates to the technical field of power supply network reconstruction, in particular to a power distribution network island fault recovery reconstruction method, which comprises the following steps of: dividing an island power supply area and a main network power supply area, and judging whether a fault area is in the island power supply area; performing network reconstruction on a non-fault area through a POP NSGA-II algorithm; the switch state is coded through binary coding, so that the switch state participating in reconstruction is digitally represented; obtaining a multi-objective optimization problem model by determining a plurality of objective functions and constraint conditions; a Pareto optimization path is recorded through POP, adaptive adjustment is carried out according to conditions, and a clear record of the algorithm optimization process is obtained; and determining an optimal compromise solution from the optimal solution set through a fuzzy set theory. The method has the beneficial effects that the power supply safety and stability are improved, the calculation complexity is optimized, the calculation time is shortened, the reconstruction speed and efficiency are improved, and meanwhile, the safe and stable operation of the network is ensured.

The invention discloses a privacy protection method for data sharing in VANETs based on a block chain and a reputation mechanism, relates to a privacy protection method for data sharing in VANETs, and belongs to the field of privacy protection. The invention provides a privacy protection method for data sharing in VANETs based on a block chain and a reputation mechanism in order to solve the problem that malicious users leak user privacy or inject damaged data and the like in the vehicle data sharing process in existing VANETs. The method comprises the following steps of: 1, respectively initializing a certificate authority (CA), a block chain member, a vehicle user and the reputation of the vehicle user; 2, performing data sharing based on an initialization result in the step 1; 3, reputation feedback is carried out based on data sharing in the step 2; and 4, vehicle reputation updating is carried out based on the reputation score obtained in the step 3.

The present application discloses a trajectory path retrieval method, device, server and storage medium. The method includes: obtaining multiple path feature information of a target trajectory path; encoding each path feature information of the target trajectory path respectively to generate a hash fingerprint for the target trajectory path; obtaining a predetermined number of candidate trajectory paths for the target trajectory path from a pre-established path index library according to the hash fingerprint of the target trajectory path; obtaining the similarity between the target trajectory path and each candidate trajectory path according to the hash fingerprint of the target trajectory path and the hash fingerprint of each candidate trajectory path, and obtaining at least one similar trajectory path for the target trajectory path from the predetermined number of candidate trajectory paths according to the similarity and a preset threshold. Thus, by using indexing technology and hash fingerprints for trajectory paths to retrieve similar trajectory paths, the retrieval speed can be accelerated, the amount of calculation can be reduced, and the retrieval efficiency can be improved.

This invention provides a method and system for UAV path planning based on a fusion algorithm, relating to the field of UAV autonomous navigation and path planning technology. The method includes the following steps: acquiring the starting coordinates, ending coordinates, and obstacle information in the flight environment of a flight mission, and constructing a flight environment spatial model; defining path constraints and optimization objectives, and generating an initial set of feasible paths based on the flight environment spatial model, path constraints, and optimization objectives; performing weighted fusion and guided enhancement on the initial set of feasible paths to obtain a weighted fused path and a guided enhanced path; employing a particle swarm optimization algorithm to construct an initial particle swarm based on the initial feasible path set, the weighted fused path, and the guided enhanced path, and iteratively optimizing the initial particle swarm to obtain a globally optimal particle; and decoding the globally optimal particle to obtain the final flight path. This invention significantly improves the efficiency and quality of UAV path planning in complex static obstacle environments.

The invention discloses a tea bud detection method and system based on a lightweight FPGA (Field Programmable Gate Array), and the method comprises the steps: carrying out the lightweight processing of a network model through eight-time channel pruning, and carrying out the weight fixed-point quantification of weight distribution, thereby not only compressing a model redundancy filter according to the model parameter quantity and calculation quantity, but also improving the detection precision of a tea bud. And meanwhile, optimization can be carried out according to the calculation characteristics of the FPGA single-engine accelerator. The pruning multiple is set to be eight to regularize the pruned network structure, calculation of a redundant filter is reduced, hardware deployment of a target detection network on a resource-limited FPGA platform is completed, and the reasoning performance of a network model is improved; besides, the PE parallel computing matrix adopts a convolution computing strategy of cyclic partitioning, and all output feature sub-blocks are obtained by performing cyclic traversal in input and output size dimensions, so that the problem that all input feature maps and weight parameters cannot be completely stored by limited FPGA storage resources is solved.

The invention provides a secure communication method and system for a low-power-consumption wireless network. And a global message sending cache set mechanism is provided, and the gateway can manage the messages which are not received, so that the data cannot be lost due to node dormancy. In combination with a timed task mechanism, the gateway can retransmit messages in a moderated manner at preset wake-up time or other appropriate time of the wireless nodes, so that the situation that the network load is increased due to too frequent retransmission, and then the performance of the whole system is affected is avoided; a key management system is introduced, and a gateway is responsible for applying and distributing working keys of wireless nodes in a unified manner. And the gateway obtains the working key which needs to be used at present through communication with the key management system, and distributes the working key to each node. The problem that the calculation amount is too large when the wireless node carries out asymmetric key negotiation is solved, the key management process of the node can be simplified, and the calculation and communication burden of the node is reduced; in order to prevent communication abnormity caused by secret key mismatching between the wireless node and the gateway, a secret key proofreading mechanism is designed in the process that the wireless node applies for a working secret key. The mechanism ensures that an encrypted communication link established between the node and the gateway is encrypted and decrypted using the correct key.

An interface interaction method, apparatus, device and product are provided. The method proposed herein includes: presenting a first message, the first message being from an intelligent system, the first message describing a first object; receiving a first input, the first input being in a natural language form, the first input indicating a second object, the second object being the first object or a part of the first object; and presenting a second message, the second message being from the intelligent system, the second message being related to an order, the order being for obtaining the second object. In this way, the selection of an object involved in a previous conversation and the ordering of the object can be completed in a natural language in a conversation flow with an intelligent system, reducing the operation steps and page jumps of human-computer interaction, improving the efficiency of human-computer interaction and reducing the computing overhead in the interaction process.

The invention discloses a video processing system based on a domestic deep learning processor, which belongs to the field of video processing systems and comprises a multi-dimensional Bayesian optimization module, an unweighted neural network module and a space-time denoising and filtering module. The multi-dimensional Bayesian optimization module uses multi-dimensional Bayesian optimization to explore and optimize the design space of the deep learning processor; the unweighted neural network module is used for introducing an unweighted neural network in neural network design, and replacing traditional weighted connection with an LUT structure, so that the calculation overhead and power consumption of a processor are reduced; and the space-time denoising filtering module provides a group of space-time denoising filters connected with the cache, and filters noise events in the preprocessing of the video stream to improve the signal-to-noise ratio of the video data. According to the invention, by combining the power consumption advantage of hardware neurons and the lightweight design of the unweighted neural network, the calculation and power consumption overhead of the processor is greatly reduced, so that the processor can run on low-power-consumption edge equipment.

This invention discloses an InSAR interferometric network optimization method based on a multi-factor coherence surrogate model. The method includes: constructing a multi-factor feature dataset containing SAR imagery, baseline data, and soil moisture data of the study area; constructing a candidate interferometric pair set, and randomly selecting a subset of sample interferometric pairs from the candidate set; constructing a coherence surrogate prediction model, which uses the multi-factor feature data corresponding to the sample interferometric pair subset and the ground truth values ​​of sample coherence to train the model parameters; inputting the multi-factor feature data corresponding to each interferometric pair in the candidate set into the trained coherence surrogate prediction model and outputting the predicted coherence of each interferometric pair and the optimized interferometric pair network. This invention improves the stability and accuracy of unwrapping and time-series inversion, while also reducing the difficulty of interferometric processing and coherence calculation under large-scale data, and can support the needs of large-scale, long-term, and near-real-time monitoring.