100 results about "Objective evaluation" patented technology

The invention provides an NoC low-delay data transmission method based on a dynamic routing algorithm, relates to the technical field of data transmission, and aims to solve the technical problems that an existing NoC routing algorithm cannot adapt to a link dynamic load, is lack of congestion trend prejudgment, is slow in path search convergence and is insufficient in QoS differential scheduling. The method comprises the following steps: constructing a dynamic sensing module containing a double-branch LSTM time sequence prediction model, collecting states such as link bandwidth and queue length in real time, and predicting a congestion trend in 50-100ms in the future; constructing a multi-objective evaluation function taking delay as a core, and dynamically adjusting the weight according to the QoS level; solving an optimal path by adopting an improved ant colony algorithm which introduces a congestion penalty and pruning strategy; during transmission, dynamic reselection is triggered through a pre-calculation + fast matching mechanism; and iteratively optimizing the prediction model based on incremental learning. According to the method, congestion is avoided in advance, the path search efficiency and the transmission reliability are improved, the delay is remarkably reduced compared with a traditional algorithm, and the method is adaptive to delay sensitive scenes such as a high-performance processor and an AI chip.

A system is provided.SOLUTION: A system comprising: means for receiving and pre-processing lesson content; means for operating a generative AI model for automatically generating lesson plans and test questions based on the pre-processed content; means for transmitting the generated lesson plans and test questions to terminals of educational institutions; means for collecting lesson videos through cameras installed in classrooms; means for automatically generating objective evaluations of students; and means for transmitting the generated evaluation results to the terminals of the educational institutions.SELECTED DRAWING: Figure 1

The invention discloses a tactical simulation confrontation training system based on digital twinning and VR. The tactical simulation confrontation training system comprises a digital twinning construction and synchronization module, a virtual reality interaction and presentation module, a tactical confrontation simulation and deduction module, a training process monitoring and data acquisition module and a training effect evaluation and intelligent redisk module. According to the invention, the digital twinborn body synchronously mapped with the real combat unit and the environment is constructed, and is deeply integrated with the immersive VR interaction environment, so that a highly vivid and panoramic immersive tactical training scene is provided for trainees, the system supports red and blue parties or multiple parties to carry out intelligent confrontation in a virtual-real combined dynamic environment, and the training efficiency is improved. The multi-dimensional training data is acquired in real time, and based on the data, the system can perform refined and quantitative objective evaluation and intelligent redisk on the tactical actions, decision-making processes and collaborative effects of individuals and teams, so that the authenticity of training, the intelligence of confrontation and the scientificity of evaluation are remarkably improved.

The invention belongs to the field of power distribution area resource scheduling, and provides a power distribution area adjustable resource complementary potential evaluation method and system fusing subjective and objective, and the method comprises the steps: constructing a power distribution area complementary potential evaluation index system; based on the sorting result of each index, determining the subjective weight of each index by using an order relation analysis method; acquiring actual operation data corresponding to the indexes, and determining objective weights of the indexes based on the standard deviations of the indexes and conflicts among the indexes; performing fusion optimization based on the subjective weight of each index and the objective weight of each index to obtain a subjective and objective evaluation fusion weight; and obtaining actual operation data corresponding to each index, determining cloud model digital features of each index based on subjective and objective evaluation fusion weights, and performing weighted fusion by using a cloud model to obtain an evaluation result. According to the method, the operation characteristics of the multi-type adjustable resources in the power distribution area can be comprehensively depicted from the three aspects of source, load and source-load complementation.

The invention relates to a pipeline defect intelligent detection method based on YOLOv12 improved attention enhancement and multi-scale adaptive fusion. According to the method, an A2C2f enhanced convolution module, an MS-AFFN multi-scale feature fusion module, a double-path attention mechanism module and an intelligent detection decision module are innovatively constructed on the basis of a YOLOv12 architecture, a three-branch parallel feature extraction mechanism is constructed, collaborative learning of attention enhanced convolution and multi-scale self-adaptive fusion is adopted, and the intelligent detection decision module is constructed. And collaborative optimization of local defect feature enhancement and global context modeling is realized. An ACEM adaptive channel enhancement module and an SAM space attention module are innovatively designed, and adaptive fusion of defect features of different scales is realized through a weight dynamic adjustment mechanism guided by global adaptive pooling. The method effectively solves the problems of insufficient small target identification precision, simple multi-scale fusion strategy and difficulty in balancing real-time performance and accuracy in traditional YOLOv12 in pipeline defect detection, mAP50 on a data set containing 22 and 120 labeled samples reaches 99.3%, the mAP50 is remarkably improved compared with a standard YOLOv12 method, the average reasoning time is only 9.4 ms / image, and the method is suitable for large-scale pipeline defect detection. Experimental results show that the model has excellent performance in objective evaluation indexes and subjective evaluation, has a remarkable effect in six types of pipeline defect detection, and has a remarkable technical effect and industrial application value.

The invention relates to a digital hot continuous rolling loop evaluation method. The method comprises the steps that the complete control period of a loop is divided into a steady-state control stage, a loop rising control stage and a loop falling control stage; establishing evaluation parameters according to the working parameters in the process of the steady state control stage, the loop rising control stage and the loop falling control stage; setting a reasonable range and an optimal range for each evaluation parameter, and mapping each evaluation parameter into a single score through a preset scoring function; and obtaining the loop control quality standard according to all the single scores. According to the method, comprehensive digital evaluation is formed by decomposing, standardizing and scoring key parameters of the whole loop control process, objective evaluation, accurate management and control, fault early warning and continuous optimization of the loop control effect are achieved by utilizing the periodic trend of the comprehensive digital evaluation, and finally the out-of-control probability of the loop and production accidents are reduced.

The invention relates to the technical field of parking management systems, in particular to a parking lot service level evaluation method based on the perspective of user experience, and the method comprises the steps: constructing a parking lot service level evaluation index system which comprises a first-level index, a second-level index and a third-level index; acquiring basic data of the parking lot; calculating the score of each third-level index according to the basic data; calculating the weight of each third-level index in the second-level indexes by adopting a standard deviation method, determining the final weight by combining subjective fine tuning, and calculating the score of each second-level index; the weight of each second-level index in the first-level indexes is calculated by adopting a standard deviation method, and after the final weight is determined by combining subjective fine tuning, the score of the parking service index is calculated; and performing qualitative evaluation on the parking lot service level according to the parking service index score. From the perspective of a user, comprehensive and objective evaluation of the service level of the parking lot is realized by scientifically constructing an evaluation index system and calculating scores, reference is provided for the user to select the parking lot, and an operator is guided to optimize the service.

The application discloses a MOFs synthesis route multi-objective optimization method based on machine learning. The method is to collect the synthesis conditions of prepared Ce-UiO-66, and evaluate the defect content and thermal stability thereof through a thermogravimetric analysis curve, as initial data set; the data set is randomly divided into a training set and a test set, eight algorithms are adopted to model each performance of Ce-UiO-66 and select the proxy model for performance prediction; the target achievement probability (PA) value of each performance in the synthesis space is calculated and expanded into a multi-objective evaluation factor; the Ce-UiO-66 material is prepared; the obtained material is subjected to characterization test, if the test data does not meet the requirement, the data set and the proxy model are updated. The application has the advantages of low cost, short cycle and the like in optimizing the catalytic and stable performances of MOFs based on reliable experimental data and machine learning, and can be popularized to the design of synthesis routes of other materials.

The application provides a noise annoyance modeling method for an audio injection system, which comprises the following steps: firstly, according to a target sound, a corresponding control sound is selected to perform time domain superposition, so as to obtain a plurality of superimposed sound samples; then, a subjective evaluation experiment and data processing are performed on the obtained superimposed sound samples; then, psychoacoustic parameters of the target sound, the control sound and the superimposed sound are calculated, and the difference between the corresponding psychoacoustic parameters of the target sound and the control sound is calculated; the difference between the calculated psychoacoustic parameters of the target sound and the control sound is taken as an independent variable, and the corresponding superimposed sound annoyance score value is taken as a dependent variable; the correlation between the independent variable and the dependent variable, the correlation between the independent variables and the collinearity between the independent variables are calculated; according to the calculation results, the modeling independent variable is determined; the determined modeling independent variable and the superimposed sound annoyance score value are taken as the dependent variable, and a multiple linear regression is adopted to realize the audio injection superimposed sound annoyance modeling. The application improves the precision and parameter correlation of the model, and further improves the objective evaluation of the audio injection superimposed sound annoyance.

The embodiment of the present application relates to the technical field of artificial intelligence and natural language processing, and discloses a petroleum large language model evaluation method and device, an electronic equipment and a storage medium.The present application inputs test data corresponding to the task target into the petroleum large language model according to the task target, and obtains a test result; the task target is divided into the following categories: knowledge understanding category, data processing category, task execution category, anti-interference category and process deduction category; the test result is evaluated using a preset evaluation rule corresponding to the category of the task target, so that the petroleum large language model can be objectively evaluated, and the application effect of the model can be quantified and verified.The evaluation task target includes multiple dimensions and indexes, and can comprehensively evaluate the performance of the large language model in the petroleum field.

The present application relates to the technical field of fastener production, and discloses a fastener production combined process determination method based on a parameter matching algorithm. The method comprises: classifying fastener parameters using a multi-dimensional parameter classification method; collecting sample data based on the classification results, and establishing an association matrix of parameters and processes; generating process labels using a constraint clustering algorithm; dynamically adjusting parameter weights for samples to be processed, and generating candidate schemes using a weighted matching algorithm; selecting the optimal process combination through multi-objective evaluation; and determining detailed production parameters. The system comprises parameter classification, data analysis, constraint clustering, weighted matching, multi-objective evaluation, and parameter determination modules. The present application solves the problems of fixed parameter weights, lack of systematic process induction, and ignoring process constraints in existing methods, realizes the intelligentization and optimization of process determination, and improves production efficiency and scheme accuracy.

The invention provides a method and a device for evaluating usability of a panoramic looking-around system, a terminal and a medium. The method comprises the following steps: acquiring a plurality of test tasks of the panoramic looking-around system and test evaluation indexes for evaluating the plurality of test tasks; according to the test task, test environment parameters are set for a whole vehicle optical test platform, subjective scoring data and objective test data are obtained, and objective scoring data are obtained according to the objective test data; analyzing correlation intensity between a subjective evaluation index and an objective evaluation index based on the subjective scoring data and the objective scoring data to obtain objective forward close degree data and net correlation intensity data; and obtaining system usability according to the subjective score, the objective score, the objective forward close degree data and the net correlation intensity data. The fusion degree of technical parameters and user perception is improved, the evaluation dimension of the panoramic looking-around system is improved, and the evaluation effect of the panoramic looking-around system is improved.

The invention discloses a subjective and objective combination weighting evaluation method based on variance optimization, and the method comprises the steps: firstly evaluating the relative importance of each index, and obtaining the subjective weight of each index; constructing a sample-index matrix, calculating the information entropy value of each index, obtaining the weight of each index based on the information entropy, removing redundant indexes through the calculation of the index correlation degree and the correlation degree between the indexes, achieving the de-overlapping, calculating the absolute correlation degree of the indexes, and carrying out the normalization to obtain the objective weight of the indexes; setting subjective and objective weight coefficients, carrying out linear addition processing on subjective and objective weights, and on the premise that the sum of constraint condition coefficients is 1, enabling the variance to be as small as possible to obtain the optimal subjective and objective combined weight of each index; and judging subjective and objective weights of the indexes and the consistency of sorting arrays, and determining final weights of the indexes. On the basis of keeping the advantages of the original subjective and objective evaluation method, the advantage complementation of subjective weighting and objective weighting is realized, and the evaluation effect of the evaluation model is improved.

The invention relates to the technical field of vehicle sound quality, in particular to a vehicle lifting beacon sound quality optimization method and device, and the method comprises the steps: collecting the noise of a plurality of types of corresponding lifting beacons of a plurality of vehicles in the lifting process, and carrying out the analysis to obtain a plurality of psychological acoustic parameters in the lifting process; obtaining a noise subjective score of the rising and falling vertical mark; and analyzing relevancy between the psychoacoustic parameters and the noise subjective score to determine at least one objective evaluation item of the rising and falling standard and a corresponding objective parameter index according to the relevancy so as to optimize an actual index of the at least one objective evaluation item of the rising and falling standard until the objective parameter index corresponding to the at least one objective evaluation item is reached. Therefore, the problems that in the prior art, the engine noise and the NVH noise of the vehicle are optimized by independently depending on the noise reduction sound pressure level, a special evaluation method for the special acoustic characteristics of the lifting beacon is lacked, and the use requirement and the optimization requirement of actual engineering cannot be met are solved.

The application discloses a transformer substation digital twin system maturity evaluation method and system based on a multi-level evaluation system tree, and relates to the technical field of digital twin maturity system evaluation. The method comprises the following steps: defining maturity levels, score intervals and level standard typical constraint signs; constructing a multi-dimensional transformer substation digital twin comprehensive multi-level evaluation system; defining an evaluation index quantifiable calculation method; calculating an expert subjective weight based on FAHP; calculating an objective weight based on IEWM; dynamically combining coefficients λ ; calculating a level combination weight based on WGM λ ; layer-by-layer weighted score calculation; level standard typical sign constraint; rating radar chart and advantage and disadvantage (function clustering) analysis; obtaining iteration dimensions and contents. The application is helpful for realizing objective evaluation, the evaluation data is derived from system implementation, the score is calculated in a quantitative manner, and the application has strong accuracy and reliability.

The invention provides a forging-heat treatment quality energy efficiency collaborative process parameter intelligent decision-making method, and relates to the technical field of hot working process management and control, and the method comprises the steps: collecting data of a forging-heat treatment process, and constructing a sample library containing process parameters, quality indexes and energy efficiency indexes; constructing a multi-target evaluation model fused by the mechanism model and the data-driven model based on the sample library; modeling a forging-heat treatment process as a Markov decision process, and constructing a reinforcement learning environment taking process parameters as actions and taking quality indexes and energy efficiency indexes as multi-objective return; generating a plurality of reference points in the multi-target space, decomposing the multi-target optimization task into a plurality of sub-problems, and associating each reference point with at least one parameterization strategy; parallel sampling and strategy updating are carried out on strategies corresponding to the reference points; and screening the current process parameter solution by using a non-dominated relationship and a reference point neighborhood criterion, updating an elite process parameter file, and outputting a Pareto process parameter solution set.

This invention relates to the field of intelligent screening technology, and particularly to a high-dimensional feature intelligent screening method and system based on large-model semantic understanding. The system includes: a data acquisition module, an initial feature processing module, a large-model feature derivation and screening module, a multi-objective evaluation module, and an optimization result output module. The core innovation of this invention lies in building a large-model-driven, fixed-number-iteration iterative optimization feature screening system based on the OpenEvolve open-source framework, organically integrating traditional statistical quantitative indicator evaluation with the deep semantic understanding capabilities of large-models, and realizing automated and targeted iterative correction of feature screening code.

The invention relates to the technical field of vehicle sound quality, in particular to a hidden outer handle sound quality optimization method and device, and the method comprises the steps: collecting the noise of a hidden outer handle of a vehicle in the opening and closing process, so as to analyze a plurality of psychological acoustic parameters in the opening and closing process; obtaining a noise subjective score of the hidden outer handle; and analyzing relevancy between the psychoacoustic parameters and the noise subjective score to determine at least one objective evaluation item of the hidden outer handle and a corresponding objective parameter index so as to optimize the hidden outer handle until the actual index reaches the objective parameter index corresponding to the at least one objective evaluation item. Therefore, the problems that in the prior art, the noise of the vehicle and the outer handle of the vehicle door is optimized by independently depending on the sound pressure level for reducing the noise, it is difficult to effectively connect with the actual subjective feeling of a user on the noise, targeted optimization cannot be conducted, and the use requirements of a driver and passengers and the optimization requirements in an actual scene cannot be met are solved.

This invention belongs to the field of metal pressure processing and intelligent manufacturing, specifically relating to an intelligent prediction and process optimization method for the forming performance of hot-rolled titanium strip. The invention discloses an intelligent prediction and process optimization method for the forming performance of hot-rolled titanium strip, aiming to solve the problems of low prediction accuracy and lack of closed-loop optimization caused by the complex coupling of hot-rolling process parameters. The method includes: collecting multi-source heterogeneous data from hot rolling and constructing a multi-dimensional feature correlation matrix; establishing a prediction model based on a fusion architecture of deep convolutional neural networks and long short-term memory networks, extracting deep features of microstructure evolution and mechanical response; using the prediction model as the objective evaluation function, employing a non-dominated sorting genetic algorithm to perform collaborative optimization of process parameters throughout the entire process; and achieving real-time feedback correction and closed-loop control through an online learning mechanism. This invention achieves accurate performance prediction and global process optimization, improving prediction robustness and finished product qualification rate, and reducing production costs.

The application relates to the field of intelligent traffic, in particular to a driving data collection and analysis system and method, which comprises a global multidimensional data collection module, an intelligent fusion perception processing module, a data set and scatter module and a backtracking analysis platform; wherein the global multidimensional data collection module collects vehicle external environment, vehicle self state and driver operation state data in the vehicle running process; the intelligent fusion perception processing module performs real-time fusion, structured processing and labeling on the data, forms true value perception and road label data; the data set and scatter module converges the processed data, stores and manages the data, and distributes the data to a subjective and objective evaluation module, generates subjective and objective evaluation results through artificial label dotting and algorithm rule matching; and the backtracking analysis platform is used for backtracking and evaluating vehicle illegal or risk behaviors in combination with the subjective and objective evaluation results, and generating an analysis report. Therefore, the problems of low perception range, insufficient perception accuracy and single evaluation function in the prior art are solved.

The invention provides an image super-resolution processing method and device, and the method comprises the steps: introducing fuzzy kernel non-negative constraint, fuzzy kernel singular value decomposition constraint and fuzzy kernel Schur complementary set constraint in the aspect of fuzzy kernel constraint, and constructing a more precise fuzzy kernel estimation model. According to the method, the visual quality and objective evaluation indexes of the super-resolution image can be improved, and the stability of the training process can be enhanced.

This application discloses a motion standardization counting system and motion counting device, relating to the field of motion recognition technology. The system includes: a motion detection module for acquiring user training motions and filtering target motion data from the training motions based on preset motion data; a motion analysis module for receiving target motion data and comparing it with preset motion requirements to generate standardization analysis results; a counting module for receiving target motion data and counting the motions after each received data point to obtain counting information; and an output module for outputting motion feedback information to the user based on the standardization analysis results and the counting information. Compared to existing counting devices that only accumulate the number of motions, this application performs multi-parameter standardization comparison and real-time analysis of the filtered motion data through preset motion requirements, thereby achieving an objective evaluation and real-time feedback of the user's motion execution quality.

The application relates to a kind of unsupervised domain adaptation street view semantic segmentation method, establish PixEL unsupervised domain adaptation network model;First, the pre-training segmentation model is obtained by training the segmentation network deeplabv2_multi to source domain data set;In target domain, the features obtained by deeplabv2_multi encoder are added to the disturbance of image, and the enhanced version of the features of the image is obtained;The strong enhancement version, weak enhancement version and feature disturbance version of the target domain image are trained by using the segmentation model;In target domain, the closeness of the energy score of the unlabeled sample to the current training distribution is evaluated by calculating the energy score, the sample with higher energy score is selected to generate pseudo label, and the model is further optimized and trained by using the pseudo label.The application compared with the existing unsupervised domain adaptation method obtains better semantic segmentation effect on the benchmark data set from synthesis to reality, whether in subjective vision or in objective evaluation index is obviously improved, effectively solves the unsupervised domain adaptation problem in the field of street view image semantic segmentation.

The invention discloses an automobile sound effect intelligent evaluation system and method based on deep learning, and relates to the field of automobile engineering and artificial intelligence, and the system comprises an audio data collection module, a psychoacoustic feature extraction module, a deep learning evaluation module and an optimization guidance module. The system obtains an original audio signal of a vehicle, extracts psychoacoustic parameters such as loudness and sharpness, and constructs a time sequence feature matrix; performing convolution operation and nonlinear mapping on the feature matrix by using a pre-trained one-dimensional convolutional neural network model, and outputting predicted values of subjective evaluation indexes such as fluency, comfort and luxury; and key parameters influencing the sound quality are identified by calculating the gradient of the predicted value relative to the input matrix. According to the method, automatic objective evaluation of the automobile sound effect is realized, a mapping relation between physical parameters and subjective hearing is established, optimization guidance based on attribution analysis can be provided, and the adaptability of an evaluation model to a dynamic driving scene is ensured.

This invention provides a BIM-based method and system for dynamic management and optimization of project schedule and cost, belonging to the field of project management and optimization technology. It includes constructing a digital twin platform with real-time sensing capabilities based on a BIM model and multi-source sensor data; importing schedule plans and cost baseline data to establish a schedule-cost correlation model, calculating a comprehensive deviation value, and automatically generating multiple sets of optimization schemes containing resource allocation, process organization, and cost control parameters; performing parallel simulations on the digital twin platform to obtain schedule and cost prediction results, conducting multi-objective evaluation, selecting the optimal scheme, and issuing it to the site for execution; continuously collecting actual execution data and comparing it with simulation results; when the deviation exceeds a threshold, updating the schedule-cost correlation model in the digital twin platform using actual data to achieve adaptive evolution and form dynamic collaborative management and control of schedule and cost. This invention can effectively improve the accuracy and efficiency of schedule and cost management and reduce project risks.

The present disclosure relates to an engineering design process data quality evaluation method and device, and relates to the technical field of computers. The method comprises: obtaining target engineering design process data; structurally processing the target engineering design process data to determine a plurality of fields; determining a plurality of evaluation parameters according to the fields; determining evaluation information of the target engineering design process data according to the evaluation parameters; and outputting the evaluation information. The present disclosure can deeply mine and model the proofreading opinion data in the background, evaluate the quality of each process, and achieve more objective evaluation.

The invention discloses a subjective and objective evaluation method for electric drive order noise, and the method comprises the following steps: building a test environment containing a five-point acoustic sensor in a semi-elimination NVH test room, setting transient and steady state working conditions covering multiple rotating speeds and multiple torques, and collecting noise data; extracting objective parameters such as an order noise decibel value and a PR value under a transient working condition and a total sound pressure level average value and linearity under a steady working condition; and in combination with subjective scoring completed by a professional review group according to a 7-level standard, performing weighted calculation to obtain scores of various working conditions and a comprehensive evaluation result. The method solves the problems of large environmental interference, incomplete working condition coverage and disjunction of subjective and objective indexes in electric drive order noise evaluation in the prior art, can accurately identify noise sensitive orders and risk points in the research and development stage, improves the evaluation accuracy and comprehensiveness, reduces the optimization cost, shortens the research and development period, and improves the development efficiency. The method is suitable for NVH performance evaluation and optimization of various new energy automobile electric drive systems.

The invention relates to a space-time semantic dynamic window obstacle avoidance path planning method based on large model enhancement, and the method comprises the steps: constructing a cognition-decision-reaction three-layer collaborative architecture, and constructing a cost map fusing space semantics and time dimension prediction in a reaction layer; and real-time accurate obstacle avoidance is realized by adopting a multi-target evaluation function of which the weight can be dynamically adjusted and a speed sampling space of orbit constraint. And the decision-making layer performs scene understanding and intelligent reasoning by using a visual language model and a large-scale language model, dynamically generates a behavior instruction and adjusts parameters of the reaction layer. And the cognitive layer realizes continuous learning and optimization of the system through digital twin simulation and a knowledge base generated based on retrieval enhancement. According to the method, the cognitive ability, scene adaptability and passing efficiency of the robot in a complex dynamic environment are remarkably improved, and the method is particularly suitable for a track type inspection robot.

The invention belongs to the technical field of vehicle noise testing, and particularly relates to a constant-speed noise subjective and objective evaluation association method, which comprises the following steps of: on the basis of steady-state noise sample data under historical constant-speed and high-speed working conditions, obtaining a subjective evaluation scoring model based on a power exponent and a sound pressure level; in the testing process, after subjective evaluation is carried out on the standard sound source, a selected standard sound source sample plate is obtained; and according to real vehicle test data, obtaining a real vehicle sound pressure level and a power exponent parameter, and inputting the real vehicle sound pressure level and the power exponent parameter into a conversion formula associated with subjective and objective evaluation to obtain a subjective level score. According to the technical scheme, the subjective evaluation scoring model based on the power exponent and the sound pressure level is analyzed and summarized by starting from the steady-state noise sample under the constant-speed and high-speed working conditions and combining the self-power spectrum distribution characteristics and the subjective evaluation grade evaluation score, the accuracy and feasibility of the subjective evaluation scoring model are verified, and the technical problem of subjective objective results is solved.

The application is an underwater image super-resolution reconstruction method based on dynamic window and state space cooperation. The method comprises a shallow feature extraction module, a deep feature extraction module and an image reconstruction module, through the construction of a double-path feature interaction mechanism, the alternative learning of a dynamic window attention module and a visual state space module is adopted, the cooperative optimization of local detail enhancement and global context modeling is realized. A dynamic channel gate feedforward network is innovatively designed, through a channel weight dynamic adjustment mechanism guided by a global average pooling, adaptive fusion of multi-scale features is realized. The method effectively solves the problems of blur and contrast reduction of underwater images caused by scattering effect, the PSNR is 28.21dB in the x3 task of the USR-248 dataset, which is 0.35dB higher than that of SRFormer, the experimental results show that the model has excellent performance in objective evaluation index and subjective evaluation, and remarkable effect is achieved in detail recovery, which has significant technical effect and industrial application value.