130 results about "Feature evaluation" patented technology

The invention relates to the technical field of energy consumption interval prediction, and particularly discloses a building energy consumption interval prediction method, system and equipment, and the method comprises the steps: collecting historical energy consumption data, meteorological data, building characteristic data and use mode data of a target building, and carrying out the hierarchical processing, thereby obtaining a hierarchical data set; performing wavelet transform decomposition on the historical energy consumption data, and decomposing the energy consumption time sequence into a trend term, a periodic term and a random term to obtain multi-scale decomposition features; constructing a feature evaluation model, calculating a contribution degree and dynamically allocating weights to obtain a weighted feature set; constructing an integrated prediction framework based on the weighted feature set, and generating a prediction interval through quantile regression; and performing adaptive adjustment according to the historical prediction deviation and the environmental factor change to obtain a final prediction interval. Through multi-scale feature processing, dynamic weight optimization and prediction interval adaptive calibration, the accuracy and reliability of the prediction interval are improved, and more accurate support is provided for building energy management.

The invention relates to the technical field of data processing, and discloses a multi-source data acquisition processing method and system based on artificial intelligence, and the method comprises the steps: carrying out the priority evaluation of the data of a data source based on a data multi-dimensional feature evaluation system and an LSTM network model, determining an association result of the data in an acquired data set based on an association rule algorithm, and carrying out the calculation of the association result. Whether correlation verification is passed or not is judged according to the reproduction frequency, a correlation result is directionally modified, a data correlation identifier is generated, the correlation result is supplemented based on the data correlation identifier and a graph theory method, data without correlation in the collected data set is classified through a three-level classification system, and a classification result is determined. And determining a data degree score for the associated data set and the classification result based on a priority evaluation result, verifying the integrity of the storage according to the hash value, and adjusting the storage position of the data storage according to the call page view. According to the invention, the real-time performance, integrity and accuracy of multi-source data acquisition and processing are ensured.

The invention discloses an image restoration method, system and device, a medium and a product, and belongs to the field of power grids, and the method comprises the steps: obtaining a to-be-restored image of a power scene under an extreme weather condition, inputting the to-be-restored image to an image restoration model, and obtaining a target image, the image restoration model is obtained by guiding an initial restoration model under a semi-supervised framework through an evaluation model to carry out iterative training, in each iteration, the label-free data is input into the first restoration model to obtain a restoration result, the restoration result is input into the evaluation model to carry out quality evaluation to obtain an evaluation signal, and the evaluation signal is sent to the semi-supervised framework; dynamically adjusting model parameters of a current second repair model based on the evaluation signal to guide the current second repair model to learn key features of the power equipment under extreme weather conditions, the evaluation model being obtained by performing all-parameter supervision fine tuning training on a multi-modal large language model based on power scene data; therefore, the generalization ability and the repairing effect of the repairing model can be improved by implementing the method and the device.

The invention belongs to the field of intelligent medical treatment. The invention provides a two-channel feature screening method for cognitive impairment recognition modeling and a modeling method thereof, and the method comprises the steps: taking multi-site cognitive impairment screening data as input data, and carrying out the preprocessing of the input data; constructing a traditional robust feature screening channel and an LLM knowledge enhancement screening channel, and performing feature screening on the preprocessed input data; integrating dual-channel feature screening results, performing bias perception joint score optimization, and obtaining a feature set; and performing semantic alignment and version normalization on the feature set to obtain a high-quality feature set. And applying the high-quality feature set to a classifier for cognitive impairment recognition modeling, and generating a clinical decision support result. According to the parallel feature screening method based on combination of a large language model and traditional feature screening, the problem of data heterogeneity bias in cognitive impairment recognition is systematically solved by constructing a dual-channel collaborative feature evaluation architecture, and organic unification of statistical robustness and clinical interpretability is achieved.

A task distribution rationality judgment method and device, equipment, storage medium and program product are disclosed, a target task area is divided into at least two sub-areas, the target task area is the overall task range of large model training under a cloud edge collaboration architecture; the area parameters of each sub-area are obtained respectively, the area parameters include at least two types of feature data; for the area parameters of each sub-area, the area parameters are subjected to quantitative operation respectively, and at least two types of feature evaluation values corresponding to each sub-area are obtained; based on the feature evaluation values corresponding to all sub-areas, the collaborative distribution rationality value of the target task area is calculated; when the collaborative distribution rationality value is greater than a preset reasonable distribution threshold value, it is determined that the task distribution of the target task area is in an abnormal state. By adopting the embodiment of the present application, the task multi-element characteristics and the all-around state of the cloud edge node can be comprehensively evaluated, so that the rationality of the large model training task distribution can be accurately judged, and the large model training efficiency is improved.

The embodiment of the invention provides a clinical intervention decision-making method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the steps that secretory otitis media clinical data of a to-be-evaluated user are acquired, and the secretory otitis media clinical data comprise multiple etiological clinical features; obtaining a target secretory otitis media disease probability prediction model; on the basis of a target secretory otitis media disease probability prediction model, performing feature evaluation on the clinical features of the multiple etiologies to obtain multiple user feature scores; based on the target secretory otitis media disease probability prediction model and the plurality of user feature scores, performing disease probability prediction on the to-be-evaluated user to obtain user disease prediction data; and based on the user disease prediction data, performing clinical intervention decision on the to-be-evaluated user. According to the application, the secretory otitis media can be effectively prevented, the accuracy of clinical intervention decisions can be improved, and in addition, the treatment effect of the secretory otitis media can be effectively evaluated.

The invention relates to the technical field of artificial intelligence, in particular to an automatic management method for object feature assessment based on a supply-demand relationship, which flexibly takes sample data as an overall assessment effect and reference coordinates of products and commodities, and can assess an object development trend, namely centralization and discretization, through the logic and the data. The basic composition generated by spiral rising and sinking is used for evaluating damage and repairing. A complex multi-cause and multi-fruit problem is ingeniously evaluated through the relationship between supply and demand and products and commodities, and a good methodology is provided for the development and change of things; furthermore, an ideal society is simulated and selected according to supply and demand, damage and repair are promoted, and a natural and social framework combination is evaluated. In an intelligent modern society, an effective novel method is provided for automatically realizing an ideal society and an ideal nature, and impact, damage and destruction of an intelligent robot to the human society and the nature are effectively prevented.

The invention provides a multi-modal image fusion medical adhesive applicability evaluation system, and the system comprises a multi-modal image collection module which is used for collecting CT, MRI and ultrasonic image data; the wound surface geometric feature extraction module is used for extracting geometric features of the wound surface from the image data; the hemodynamic parameter extraction module is used for extracting hemodynamic parameters from the image data; the deep learning feature fusion module is used for fusing multi-modal features; and the applicable scene classification module is used for evaluating the applicability of the medical adhesive in embolism, spraying and smearing scenes according to the extracted features and outputting recommendation indexes. According to the technical scheme, the intelligent auxiliary decision-making system for automatically evaluating the application scene of the medical adhesive is realized.

The invention discloses a Fresnel lens calibration method and a Fresnel lens, particularly relates to the technical field of industrial visual inspection, and is used for solving the problem that an existing focal length calibration method based on machine vision is easily interfered by inherent annular textures of the Fresnel lens, so that the measurement accuracy and reliability are insufficient. A focusing light spot image sequence is collected by moving an image sensor along an optical axis, then a radial intensity distribution gradient feature evaluation value and an annular texture edge gradient saliency evaluation value of each image are calculated, and a candidate focus area is identified by analyzing the correlation of the two evaluation value sequences; then, the convergence of the position change trajectory of the preset feature points is analyzed in the candidate region to determine an optimized candidate region, the symmetry of the evolution trajectory of the morphological descriptor is further analyzed to accurately position and optimize the focus position, and finally the focal length is calculated; the interference of annular textures can be effectively suppressed, and the precision and robustness of focal length verification of an industrial vision system on an automatic production line are improved.

The application discloses a marketing strategy generation method and system based on user behavior sequence, relates to the technical field of intelligent marketing, and comprises the following steps: obtaining the original behavior sequence of a target user, extracting a behavior object identifier, constructing an object association structure, fusing and encoding time sequence information and the object association structure to generate an enhanced semantic representation, identifying a marketing response intention type based on the enhanced semantic representation and outputting an initial confidence degree, calibrating the confidence degree by calculating the distribution deviation degree of the enhanced semantic representation from a reference semantic representation set, evaluating a conversion potential score in combination with a behavior active feature, identifying a behavior cycle mode based on the time sequence information and predicting a marketing triggering time, and finally using the conversion potential score to determine a resource allocation quota, selecting matched marketing content, and generating a personalized marketing strategy. The application improves the accuracy and effectiveness of the marketing strategy through semantic enhancement representation and multidimensional evaluation.

The application discloses a batch service template generation method and device. The method comprises the following steps: obtaining target batch service data to be processed, determining a corresponding target service scenario, and obtaining target commodity data of the scenario; identifying all first key field features in the target batch service data and the commodity data to form a first key field feature set; analyzing the features by using a batch service template generation model corresponding to the service scenario to evaluate a first importance score of different key fields on the success of batch service processing, wherein the model comprises a random forest sub-model and a logistic regression sub-model; determining target key field features according to the score, and generating a target batch service template. The application solves the technical problem that the batch service template in the related scheme mainly depends on manual configuration, has low efficiency, is prone to errors, and has poor applicability.

The invention discloses a model training method and device, and relates to the technical field of artificial intelligence. A specific embodiment of the method comprises the following steps: pre-configuring a plurality of initial features of a specific article of a target category, and determining an evaluation score of the plurality of initial features in at least one evaluation index according to a pre-trained feature evaluation model; screening out at least one key feature from the plurality of initial features according to the evaluation score, and training a pre-established grading model by using the training sample data of the specific article based on the key feature; the grading model is a random forest model comprising a plurality of composite decision-making trees, any composite decision-making tree comprises a top decision-making tree and two bottom decision-making trees, and the top decision-making tree and the bottom decision-making trees are binary trees; the root nodes of the two bottom-layer decision trees in the same composite decision tree are leaf nodes of the top-layer decision tree. According to the embodiment, automatic and refined grading of the specific article can be realized based on the artificial intelligence model.

The present disclosure relates to a multi-modal information fusion representation exoskeleton gait feature recognition method and device, electronic equipment and storage medium. The method comprises: calibrating and initializing the sensor of the exoskeleton wearer coupling system; collecting exoskeleton sensor signals, calculating multi-order derivatives and storing as time series; constructing a multi-layer stack denoising sparse auto-encoding neural network model; constructing a parallel neural network model for space-time domain feature hybrid extraction; based on a genetic-particle swarm hybrid optimization method, the weight vector of the cross-entropy function in the above model is optimized and searched, the mode search of the gait feature label vector predicted by the model in the previous n time is carried out, and the abnormal value is filtered and recognized as the gait feature evaluation output of post-processing. The present disclosure realizes that the recognition accuracy of the label reaches the target optimal value, effectively avoids the oscillation step caused by the recognition abnormal value, and effectively improves the recognition accuracy, robustness and generalization ability of the exoskeleton for gait features.

The invention relates to the technical field of communication quality detection, and discloses a signal quality detection method based on video telephone real-time communication, which comprises the steps of data acquisition, feature extraction, feature evaluation, quality analysis and real-time feedback, and realizes accurate detection of video telephone real-time communication signal quality. According to the method, communication videos are collected at fixed time intervals, a plurality of key feature data such as a video frame rate, an audio sampling rate, a video code rate, an audio code rate, a packet loss rate and delay time are extracted, and the quality condition of communication signals can be comprehensively and accurately reflected. And performing evaluation processing on each piece of feature data to obtain a corresponding evaluation value. According to the quantitative evaluation mode, the scientificity and objectivity of detection are improved, interference of human factors is reduced, and the evaluation result is more reliable.

The invention relates to a large model resource protection method based on static feature evaluation and dynamic decoding control, which comprises the following steps of: receiving a text input by a user, and carrying out preprocessing and static feature analysis on the text to obtain a risk score; based on the risk score, a decoding tendency mode is set, large model reasoning is started, and a dynamic control module is initialized; the dynamic control module dynamically calculates reasoning characteristics and executes real-time decoding intervention in the reasoning process according to the reasoning characteristics; according to a decoding intervention result, if reasoning is terminated in advance, an interrupt signal is sent to a reasoning rear end of the large model, Token generation is forcibly stopped, reasoning fallback is carried out, and a reasoning fallback result is fed back to the user; according to a decoding intervention result, if decoding is inferred to be normal and a termination Token is generated, a complete response is output and fed back to the user. Compared with the prior art, the resource depletion attack can be efficiently recognized before a large number of computing resources are occupied, and deep defense before reasoning and in reasoning is achieved.

The invention provides a business processing method which can be applied to the technical field of artificial intelligence. The business processing method comprises the following steps: in response to received user operation participating in a current activity, obtaining feature data of a user; the feature data are input into a feature evaluation model, a comprehensive weight score of each feature is obtained, the feature evaluation model is obtained by training a historical data set, and the historical data set comprises user features, intervention tags and conversion result tags; according to the comprehensive weight score of each feature, screening out at least one feature from the feature data as a target feature; based on the target feature, determining and executing a business processing flow corresponding to the current activity; wherein the training process of the feature evaluation model comprises the step of adjusting model parameters by fusing the prediction importance measurement value and the lifting importance measurement value of each feature. The invention further provides a business processing device, equipment, a storage medium and a program product.

The application discloses an industrial malodor online monitoring system based on big data, which comprises a data acquisition module, a pretreatment module, a feature fusion module, a multi-source perception weight analysis module, a dynamic risk assessment module, a self-adaptive correction module and a real-time feedback module; data interaction between the modules adopts a block chain encryption transmission protocol, and a feature evaluation mapping chain with a time stamp is established in a distributed database; the application builds a new paradigm of industrial malodor monitoring with environmental intelligent adaptability through the innovative design of heterogeneous sensor network collaborative perception, cross-modal feature organic fusion, pollution diffusion three-dimensional dynamic deduction and credible data governance architecture; four technical effects complementarily support from four dimensions of data acquisition credibility, feature modeling scientificity, risk assessment accuracy and system decision robustness.

The application discloses a biomass classification method, an electronic device and a storage medium for biochar yield and calorific value prediction, and belongs to the technical field of biomass classification. In order to more scientifically utilize biomass classification, the precision of biochar yield and calorific value prediction is improved. The application collects biochar yield data and biochar calorific value data of biomass raw material samples, carries out data cleaning and missing value processing on data in a biomass data set for biochar yield and calorific value prediction, carries out sample classification, obtains different types of biomass data sets for biochar yield and calorific value prediction, divides the data sets into a training set and a prediction set, designs a biomass prediction model for biochar yield and calorific value prediction, trains the biomass prediction model for biochar yield and calorific value prediction, and carries out feature evaluation on the trained biomass prediction model for biochar yield and calorific value prediction. The application can obtain better prediction results by inputting fewer parameters.

The invention relates to the technical field of psychological assessment, and discloses a multi-modal psychological feature modeling and assessment method based on artificial intelligence, and the method comprises the steps: carrying out the preprocessing of a house tree person drawing image, and extracting image features from the preprocessed house tree person drawing image; a score vector model is generated, image features are converted into structured vectors, a Sigmoid function is optimized and trained, scores of different psychological dimensions are calculated through the Sigmoid function, accurate quantitative evaluation can be provided for the psychological state, emotional tendency and the like of an individual, meanwhile, an SHAP method is used for interpretable analysis, the prediction result of the model can be analyzed, and the prediction accuracy of the model is improved. And the contribution of each feature to the result is quantified, so that the transparency of the model is enhanced, a tool for understanding the model decision process is provided for a user, the credibility of the model is improved, and the psychological feature evaluation accuracy is improved.

The application relates to the technical field of data processing, and discloses a multi-source data acquisition processing method and system based on artificial intelligence, which comprises the following steps: preferential priority of data of a data source is evaluated based on a data multi-dimensional feature evaluation system and an LSTM network model, an association result of data in an acquired data set is determined based on an association rule algorithm, whether the association result passes an association verification is judged according to a recurrence frequency, the association result is directionally modified and a data association identifier is generated, the association result is supplemented based on the data association identifier and a graph theory method, data in the acquired data set that does not exist in association is classified and a classification result is determined by adopting a three-level classification system, a data degree score of the associated data set and the classification result is determined based on a result of the preferential priority evaluation, the integrity of storage is verified according to a hash value, and a storage position of data storage is adjusted according to a calling access amount. The application ensures real-time performance, integrity and accuracy of multi-source data acquisition processing.

The invention discloses a boiler combustion early warning method based on an improved deep forest algorithm. The method comprises the following steps: obtaining an initial quantitative index of a combustion state of combustion equipment; determining a dynamic fluctuation mode of the combustion process of the combustion equipment under the multi-parameter coupling condition; constructing a multi-level feature evaluation structure by adopting an improved deep forest algorithm to obtain a credibility score of the combustion features; if the credibility score of the combustion feature is lower than a preset threshold value, obtaining an adjusted intermediate layer risk feature; judging a triggering condition of a potential coking risk; the source position of abnormal combustion of the combustion equipment is determined; and generating a real-time early warning signal according to the source position of the abnormal combustion of the combustion equipment. The problems that a traditional scheme is insufficient in combustion abnormity early warning capacity and lagged in response under the multi-source complex working condition are effectively solved.

The invention discloses a brain fatigue multi-modal evaluation method and system for a video operator, and belongs to the technical field of multi-modal data processing, and the method comprises the following steps: video data preprocessing, two-dimensional feature evaluation and monitoring result determination. The method comprises the following steps: collecting face video data of a target operator through a non-contact camera deployed in a specified operation area, preprocessing the video data, then carrying out two-dimensional feature extraction on the face video data subjected to the video data preprocessing to obtain a bimodal feature parameter, and meanwhile, carrying out two-dimensional feature evaluation on a brain fatigue state to obtain a brain fatigue state. And finally, according to a result of two-dimensional feature evaluation, judging a brain fatigue state monitoring result, constructing a dynamic self-adaptive and personalized evaluation criterion, providing operable and hierarchical decision support, and solving the problems of relatively low multi-modal data accuracy, relatively high reliability and the like in the prior art. And the reliability of mental fatigue assessment is not high.

The invention relates to a multi-modal sentiment analysis method. The method comprises the following steps: acquiring a multi-modal sentiment analysis data set; constructing a multi-modal sentiment analysis model, wherein the model comprises a multi-modal feature coding module, a layer specificity fusion feature generation module and a self-adaptive strategy selector; constructing a training set to train a layer specificity fusion feature generation module and an adaptive strategy selector in the multi-modal sentiment analysis model to obtain a trained multi-modal sentiment analysis model; and obtaining a to-be-analyzed multi-modal sample, and sending the to-be-analyzed multi-modal sample into the trained multi-modal sentiment analysis model to complete analysis. According to the method, the adaptability of each fusion strategy to the current sample is evaluated through the self-adaptive strategy selector in combination with the input sample characteristics and the interaction characteristics generated by the fusion strategies, and a plurality of candidate strategies are dynamically selected. By fusing the reasoning results of the candidate strategies, the system can effectively improve the recognition accuracy of the complex emotional state.

The invention discloses a marketing strategy generation method and system based on a user behavior sequence, and relates to the technical field of intelligent marketing, and the method comprises the steps: obtaining an original behavior sequence of a target user, extracting a behavior object identifier, constructing an object association structure, and carrying out the fusion coding of time sequence information and the object association structure to generate an enhanced semantic representation; identifying a marketing response intention type based on enhanced semantic representation and outputting an initial confidence coefficient; the confidence coefficient is calibrated by calculating the distribution deviation degree with the reference semantic representation set; a transformation potential score is obtained by combining behavior activity feature evaluation; identifying a behavior period mode based on the time sequence information and predicting a marketing triggering opportunity; and finally, the conversion potential score is used for determining a resource allocation quota, and matched marketing content is selected to generate a personalized marketing strategy. According to the method, through semantic enhancement representation and multi-dimensional evaluation, the accuracy and effectiveness of a marketing strategy are improved.

The application discloses a sparse positive sample risk discrimination method and system based on cluster analysis, relates to the technical field of data processing, and comprises the following steps: obtaining positive samples and negative samples of a historical clinical data set, performing double evaluation on the historical clinical data through a feature evaluation model to obtain a target feature subset; projecting the negative samples to the target feature subset to obtain a feature space, performing cluster analysis to determine a plurality of data subgroups and a plurality of cluster centers, determining a corresponding anomaly detection model for each data subgroup, determining the local anomaly scores of the corresponding data subgroups through the anomaly detection model; calculating the Mahalanobis distance between the to-be-tested sample and each cluster center, determining the main subgroups and adjacent subgroups corresponding to the to-be-tested sample; determining the anomaly scores corresponding to the main subgroups and the adjacent subgroups; determining a verification set according to the positive samples, performing index maximization processing according to the verification set to determine a decision threshold, comparing the anomaly scores with the decision threshold, and obtaining a risk discrimination result.

The invention relates to the technical field of automobile detection, and particularly discloses an automobile instrument desk display abnormity automatic detection system based on big data, automatic detection of instrument desk backlight display abnormity and data fusion display abnormity is realized through cooperative work of multiple modules, and a data acquisition module acquires state data of an instrument desk in real time; the data processing module is used for cleaning noise, abnormal values and repeated data for sensor data, carrying out standardization processing, denoising image data, extracting characteristic parameters and carrying out standardization; the data analysis module is used for analyzing the processed data and extracting relevant features of backlight display and data fusion display; and the abnormality evaluation module evaluates whether backlight display abnormality and data fusion display abnormality occur in the instrument desk according to the features extracted by the data analysis module, multi-dimensional abnormality detection is realized in combination with real-time sensor data and an image processing technology, and the detection efficiency and reliability of the display abnormality of the instrument desk are remarkably improved.

The invention relates to the technical field of artificial intelligence, in particular to an electric power scientific research cluster-oriented strategy recommendation method based on knowledge graph driving, and the method comprises the steps: obtaining multi-source heterogeneous data, and carrying out the preprocessing of the multi-source heterogeneous data to obtain normalized data; constructing a scientific research resource knowledge graph based on the standardized data, and performing versioning management by adopting an incremental updating mechanism including candidate layer submission and consistency verification; taking the core technology entry as a query object, and performing candidate entry retrieval in combination with the relation path constraint and the representation similarity; calculating policy source capability characteristics, literature measurement leading edge characteristics and policy time sequence compliance characteristics of the candidate entries in a time sliding window; according to the power scientific research strategy recommendation method based on the knowledge graph, after the features are normalized, the rule scores, the representation learning scores and the constraint weights are integrated to perform weighted fusion scoring, and the strategy recommendation result is output according to the score sorting, and by constructing the dynamically updated knowledge graph and fusing multi-dimensional feature evaluation, precision and adaptive optimization of power scientific research strategy recommendation are achieved.

The application discloses a method and system for identifying a drilling formation type, comprising the following steps: according to real-time logging data under a drilling condition in a well to be evaluated, the real-time logging data is pretreated, then key features are extracted from the pretreated real-time logging data, and key feature evaluation data is obtained; and according to the key feature evaluation data, a preset formation identification model is used to predict a formation type. The application can use less types of data information to realize quick identification and high-precision identification of the formation type while drilling through real-time data of a comprehensive logging instrument in the process of drilling, and has practical application significance and value.

The invention provides an unmanned aerial vehicle grid airspace construction method based on multiple attributes, and the method comprises the steps: 1, constructing the overall grid airspace of an unmanned aerial vehicle, and quantitatively designing the structural parameters of the airspace based on the technical parameters of the unmanned aerial vehicle; 2, based on the airspace structure parameters, the unmanned aerial vehicle grid airspace is subdivided into three sub-layers, and attribute parameter distribution in each sub-layer is calculated to serve as a grid feature evaluation index of the layer; 3, determining an importance degree sequence of each attribute in an unmanned aerial vehicle grid airspace according to a grid attribute index system evaluation standard, and constructing a multi-attribute grid feature comprehensive evaluation model by adopting a combined weighting method to obtain weight values of grid feature evaluation indexes of three sub-layers; and step 4, performing multi-layer fusion to obtain a final unmanned aerial vehicle grid airspace, and assisting in decision-making airspace planning in an actual environment. The method provides a theoretical basis for the airspace operation environment and flight path planning of the unmanned aerial vehicle, and has practical significance.

The application discloses an intelligent inspection method and system for a construction site, and belongs to the technical field of image processing. The application firstly processes a remote sensing image of a cement pavement of the construction site, and positions a pixel area of the pavement; then, relative difference values of each pixel point are calculated according to a pixel mean value of the area, and a difference value distribution image is generated; then, maximum values and minimum values of each window area of the distribution image are extracted, and high difference value and low difference value compression images are obtained respectively. After that, non-pavement pixel areas are marked as abnormal points, and the two types of compression images are divided into grids, difference characteristic values are calculated based on the abnormal points in the grids, and high and low difference characteristic matrices are formed; then, joint deviation degrees of the difference characteristic values of each grid are calculated, and high and low joint deviation matrices are obtained; finally, the above matrices are processed through a pavement defect characteristic evaluation network, and a pavement quality score is output. The application effectively improves the precision of pavement evaluation through multi-dimensional feature extraction and intelligent evaluation.