1014 results about "Feature selection" patented technology

The invention relates to the crossing field of hydraulic engineering hydrological monitoring technology and machine learning prediction technology, discloses a sediment concentration prediction method based on deep learning, and aims to solve the problems that in existing sediment concentration prediction, hyper-parameter manual tuning is low in efficiency, key feature attention is insufficient, local and time sequence information is difficult to consider by a single model and the like. Accurate prediction is realized through five core modules: a data preprocessing module performs missing value filling, abnormal value processing and derivative feature generation on hydrological data; the feature selection module screens key features based on mutual information; the time sequence construction module generates time sequence data through a sliding window; the hyper-parameter automatic optimization module adopts Bayesian optimization iteration to obtain an optimal hyper-parameter; the CNN-LSTM-attention prediction module fuses CNN local feature extraction, bidirectional LSTM time sequence dependence capture and multi-head self-attention mechanism key feature focusing capability, is suitable for scenes such as river channels and channels, and provides efficient decision support for hydrological regulation and control.

The invention discloses a camouflage target detection method based on feature selection attention and frequency domain edge guidance. According to the method, four-level features of a camouflage target image are extracted through a backbone network SMT and are respectively screened; the high-level features are input into a semantic information supplement module, and after semantic features are enhanced, the high-level features and the trunk features are sent into a spatial feature enhancement module together. And inputting the obtained fine-grained features into an edge feature sensing module, and finally fusing multi-scale features through a multi-scale jump connection technology to generate a mask pattern with higher discrimination. The method has the advantages that the network parameter quantity is reduced and key information is reserved through a feature selection mechanism; a spatial feature enhancement module is used for enhancing multi-scale feature representation and remote dependence modeling; the dilution of the semantic context is relieved by means of a semantic supplement module so as to improve the positioning precision; and an edge feature enhancement module is adopted to enhance edge semantic perception and improve boundary integrity. According to the method, the camouflage target detection performance is remarkably improved with relatively low calculation cost.

The invention discloses a method, a system and equipment for predicting the stability of integrated circuit test equipment and a medium, and belongs to the technical field of integrated circuit test. The method comprises the following steps: firstly, carrying out preprocessing and feature selection on historical data in an FT test stage, and adopting a Gaussian mixture model (GMM) to cluster and identify different operation condition clusters of a test machine; then, establishing a health state GMM reference for each working condition cluster, calculating a KL divergence value of a normal sample and the reference, and setting a dynamic anomaly detection threshold by 99.73% quantile of the KL divergence value; and finally, in real-time monitoring, calculating a KL divergence value of real-time data and a corresponding working condition cluster benchmark, and comparing the KL divergence value with a dynamic threshold value to realize accurate anomaly marking. The method effectively solves the problem of abnormal detection of the test data of the integrated circuit under complex and changeable working conditions, and improves the monitoring accuracy and working condition adaptability.

The invention discloses a cerebral hemorrhage postoperative gastrointestinal hemorrhage prediction method based on an LGBM model. The method comprises the steps of obtaining a multi-dimensional clinical feature sequence of a target patient, screening out a stable feature subset by adopting a Boruta feature selection algorithm, performing nonlinear relation fitting and integrated decision by utilizing a pre-trained LightGBM machine learning model, and generating an individualized ATH risk probability value; and when the risk probability value exceeds a dynamic risk threshold value, triggering a high-risk early warning signal, and based on a Kaplan-Meier survival analysis model, carrying out association mapping on a prognosis track of poor long-term neural function recovery, and finally generating a comprehensive prediction report. According to the invention, accurate quantitative evaluation of ATH risk is realized, clinical intervention timeliness is improved through a dynamic threshold early warning mechanism, short-term complication risk and long-term function prognosis are organically combined, and a comprehensive and reliable prognosis basis is provided for individualized treatment decision.

The invention relates to a lightweight neural network model implementation method and system for small target detection in a complex aerial photography scene, and belongs to the technical field of computer vision and unmanned aerial vehicle target detection. In order to solve the problem of low detection precision caused by small target size, complex background, easy feature submerging and the like in an aerial image of an existing unmanned aerial vehicle, the method comprises the following steps: constructing a multi-scale adaptive hierarchical feature enhancement module MSAHFE, and combining multi-scale pooling and differential edge enhancement to improve feature sensitivity; constructing a lightweight feature selection module LAFS based on an attention mechanism, and screening high-correlation features by using a space and frequency double-domain attention mechanism; a lightweight detection head LWDeect is constructed, and shared packet convolution and self-calibration convolution are utilized to reduce the calculation complexity. The method has the advantages of being high in detection precision, small in parameter quantity, high in reasoning speed and the like, and is suitable for complex aerial photography and other application scenes needing real-time small target detection.

The invention discloses a teenager psychological sub-health intelligent early warning system based on multi-source heterogeneous data fusion, and the system comprises a data collection layer which collects the behavior, physiological and social three-dimensional data of teenagers through the multi-source channels of campus cards, wearable devices, social media and questionnaires, and builds an original data pool through cleaning, denoising and standardization; the data fusion layer is used for integrating multi-source data by adopting weighted average and Kalman filtering, mining psychological sub-health key indexes in combination with a feature selection and extraction technology, and forming a three-dimensional psychological portrait; the deep learning layer is used for constructing a multi-modal fusion early warning model based on a Transform architecture, and carrying out real-time prediction and dynamic tracking of psychological sub-health risks through large-scale data training and cross validation optimization; and the intelligent early warning layer is used for visually displaying an early warning result, integrating three-party linkage of a management end, a teacher end and a parent end, providing 24-hour online intervention by a built-in AI psychological counseling module, and automatically transferring high-risk cases to professional psychological consultants.

The invention discloses a vertical federal learning feature selection method based on context awareness, and belongs to the technical field of artificial intelligence and data privacy protection. According to the method, firstly, an unsupervised sparse network is utilized at a client to initialize the importance of local features so as to accelerate convergence and reduce calculation complexity; and then, obtaining the embedded representation of each client in a pre-training stage, and screening the embedded representation by combining context features through a server side, thereby indirectly identifying key features. In the feature selection stage, the client side performs local feature screening according to the significant embedded index issued by the server, and the influence of irrelevant features on calculation and communication is further reduced. According to the invention, an attention mechanism is introduced to dynamically evaluate contributions of different participants, so that fair weight distribution is realized. According to the method, through staged joint optimization, the communication and calculation cost in the federation training process is effectively reduced, and meanwhile, the prediction precision and interpretability of the model are improved.

The invention provides a water chlorophyll a concentration inversion method and system based on multi-modal data and a lightweight model, and relates to the technical field of water environment remote sensing evaluation. The method comprises the following steps: firstly, acquiring a Gaofeng No.5 satellite remote sensing image, a sentinel No.3 satellite image and ground actual measurement data, and completing image preprocessing and water body pixel extraction; constructing a hyperspectral index and an aquatic vegetation index, and fusing the hyperspectral index and the aquatic vegetation index with the water body temperature, the pH environmental factors and the spectral reflectivity to form a multi-dimensional feature sample set; a core feature subset is obtained through random forest and XGBoost coupling feature selection, and a lightweight student model is trained based on knowledge distillation; and constructing a to-be-predicted feature sample for the to-be-predicted time phase image and the environment factor, inputting the to-be-predicted feature sample into the lightweight student model to obtain a chlorophyll a concentration predicted value, and generating a spatial distribution map and a quality control map layer. According to the invention, high-precision, low-redundancy and efficient deployment chlorophyll a concentration inversion is realized.

The invention discloses a method for constructing a machine learning model based on cerebellar subregion multi-modal radiomics. According to the method, [18F] FDG PET metabolic features and 3DT1 W MRI structural features of a cerebellar subregion are extracted, a random forest classification model is constructed after feature selection, and high-precision identification of the Alzheimer's disease (AD) and a cognitive normal (CN) is realized in combination with SHAP analysis. According to the method, the multi-modal radiomics characteristics of the cerebellar subregion are integrated for the first time, the accuracy and interpretability of AD early diagnosis are improved, and a non-invasive and efficient diagnosis tool is provided for clinic.

The invention provides an electric quantity prediction method and system fusing physical constraint factors, and relates to the technical field of electric quantity prediction. Historical load, weather, electricity price and calendar data are collected, and a key feature set is constructed through preprocessing and feature selection; a prediction model with the physical information neural network as the core is constructed, the prediction model comprises a recursion sub-module used for short-term prediction and a trend sub-module used for long-term prediction, and a physical constraint loss item based on a physical rule is introduced into model training so as to enhance the generalization ability; a multi-time granularity modeling framework is adopted, uncertainty quantization is achieved through a Monte Carlo Dropout or Bayesian neural network, and a confidence interval of a predicted value is output; and finally, causal reasoning is carried out through a Shapley value algorithm and anti-fact simulation, and key influence factors are identified. According to the method, the precision, stability and interpretability of electric quantity prediction are effectively improved, and reliable support is provided for power grid dispatching and decision making.

The invention relates to the technical field of rotating machinery health monitoring and fault diagnosis, and discloses a diversion sealing intelligent diagnosis method and system.The diversion sealing intelligent diagnosis method comprises the steps that an original monitoring data set is collected; performing working condition self-adaptive preprocessing on the original monitoring data set; extracting multi-scale time-frequency-space cooperation features, and performing dimension reduction by using a feature selection method; training by utilizing physical consistency constraint and a small sample learning method to obtain a small sample diagnosis model; inputting the optimized feature vector into a diagnosis model for anomaly detection, failure mode recognition and severity evaluation; a sealing performance degradation model is established, degradation model parameters are estimated, and the remaining service life is predicted; integrating the diagnosis model and the residual life prediction model into an intelligent diagnosis system, and adopting online real-time diagnosis to obtain a diagnosis report; according to the invention, the problems of incomplete monitoring information, difficult early fault detection and lack of life prediction capability in diversion sealing diagnosis are effectively solved.

The invention provides a cross-omics sparse feature selection system and method based on hierarchical causal modeling, and the system comprises a data input and preprocessing module which is used for receiving multi-omics original data of a multivariate sample; the hierarchical causal structure learning module is connected with the data input and adaptive preprocessing module and is used for constructing a cross-omics hierarchical causal topology; the causal-oriented sparse feature selection module is connected with the hierarchical causal structure learning module; and the model retraining and integration module is used for constructing a three-layer weighted integration discrimination model based on the screened markers, optimizing the fusion weight of each layer through a gradient descent algorithm, and outputting a final prediction result. According to the method, the protein-metabolism biological hierarchy relationship and serum-urine complementary information are fully utilized, and the method has good generalization ability and can be widely applied to marker mining and prediction modeling of cancers, metabolic diseases and the like, so that the accuracy and reliability of precise medical treatment are improved.

The invention discloses an ancient silk damage type automatic identification and classification method and system. The method comprises the following steps: S100, collecting hyperspectral data; s200, positioning a damaged area: processing the reflectivity data cube to generate a binary mask of the damaged area; s300, extracting multi-dimensional features: for each damaged area, extracting features from a plurality of dimensions including geometric morphology, spatial texture, spectral characteristics and physical parameters to form an initial feature vector; s400, performing feature optimization and classification: performing feature selection and fusion on the initial feature vector, performing classification by adopting a hybrid machine learning model, and outputting a damage type and confidence; s500, result visualization and decision support are carried out; according to the method, the four-dimensional feature system fusing the macroscopic morphology, the microscopic texture, the material composition and the internal physical state is systematically constructed in the cultural relic detection field for the first time, and the defect that the single feature representation capability is weak is overcome.

The invention relates to the cross technical field of natural language processing and computer vision, provides a multi-modal relation extraction method and device fusing text to guide image feature selection and bidirectional attention interaction, and aims to improve the accuracy and interpretability of entity relation recognition in an image-text scene and reduce the reasoning complexity. According to the method, a head entity and a tail entity in a text are taken as queries, correlation scoring is carried out on image candidate areas, a Top-k effective area is selected, text-to-image and image-to-text bidirectional attention interaction is established between the text and the selected area, and in combination with gating fusion and a correlation fallback mechanism, the image-to-text bidirectional attention interaction is established. Self-adaptively suppressing visual noise in the image-text weak correlation sample; alignment and reinforcement of multi-modal representation are realized by introducing joint classification and cross-modal comparison loss. The method is suitable for multiple multi-mode application scenes such as knowledge graph construction, information extraction, intelligent question and answer and content auditing, and has good practical value.

The invention discloses an SAR ship image detection method based on an improved YOLOv11 model, an improved C2PSA module C2DyMoETAttn is introduced, the core innovation point is that a PSABlock module is replaced by a DyMoETAttnBlock module, the DyMoETAttnBlock module fuses a Dynamic Tanh activation function, a Mona module, a TSSA attention mechanism and a frequency domain enhancement feedforward network (EDFFN), multi-dimensional modeling and robust enhancement of features are realized, and the detection accuracy is improved. The feature expression capability and the noise suppression performance under the background of small targets and complex sea clutters are effectively improved; in the deep feature fusion stage, a C3k2 module of YOLOv11 is optimized, an ScConv structure is introduced, adaptive fusion of space and channel features is realized through a joint reweighting mechanism of SRU and CRU, and the multi-scale target discrimination capability and feature selectivity are enhanced; on the bounding box regression layer, a Focaler-MPDIOU loss function is provided, and a Focaler-IoU sample difficulty adaptive mechanism is combined with MPDIOU positioning matching constraint, so that the learning ability of the model for small targets and shielded targets is enhanced, and the bounding box positioning precision and convergence stability are improved.

The invention relates to a fine-grained target real-time image segmentation method and system based on a dynamic state modeling network, and belongs to the technical field of intelligent image processing. The method comprises the following steps: extracting multi-scale detail features by using a lightweight backbone network; through a dual-scale two-dimensional selective scanning module, the features are divided into a thin branch and a thick branch, and local scanning and global scanning are executed respectively; a dynamic cross-scale feature selection and aggregation module is adopted, redundancy is suppressed through reweighting and statistical filtering, and key target responses are highlighted; at a decoding end, local details and global semantics are fused through jump connection and an edge extractor; and finally, introducing a form-guided pseudo label hierarchical supervision strategy, and improving the structure learning ability of the model by using a coarse-to-fine morphological prior. According to the method, the segmentation precision, the boundary integrity and the tiny target recall rate of the fine-grained target under the scenes of ore separation, industrial defect detection, pavement crack recognition and the like are remarkably improved.

The invention relates to the technical field of computer mode recognition, and discloses a multi-source heterogeneous data fusion processing and key feature extraction method and system, and the method comprises the steps: achieving the adaptive caching and granularity normalization of streaming data through a dynamic buffering queue and a time alignment window; generating a structured vector of a unified space-time reference by using a structured analysis module; a high-dimensional fusion feature tensor is constructed through two-stage convolutional coding and a cross-source attention interaction network; and a key feature channel is screened based on gradient sensitivity through a differentiable channel pruning module. The system comprises a multi-source data access unit, a dynamic buffer management unit, a time alignment unit, a synchronous resampling unit, a structured analysis unit, a primary fusion coding unit, a cross-source attention interaction unit, a time sequence dependence modeling unit, a feature importance evaluation unit, a key feature screening unit and the like. According to the method, efficient, accurate and low-overhead multi-source heterogeneous data real-time fusion and task-oriented key feature extraction can be realized.

The invention discloses an algae abundance dynamic feature selection method based on multi-agent reinforcement learning, which belongs to the technical field of artificial intelligence and data mining, and comprises the steps of multi-source heterogeneous data acquisition and convergence, time sequence data cleaning and standardization processing, driving factor identification and weight calculation key driving factor set establishment, and dynamic feature selection. Convergence and determination of an optimal feature subset selected by multi-agent reinforcement learning and dynamic features: an MARLN system performs multi-round iteration and interactive learning, and each agent continuously optimizes a decision strategy thereof according to a global reward signal fused with a specification item; according to the method, the problems that a traditional static feature selection method cannot adapt to data changes and neglects interaction among features are solved, the most critical driving factors, namely the features, for algae abundance prediction are automatically recognized from multi-source heterogeneous data through a dynamic feature selection method, and the accuracy of algae abundance prediction is improved. And an optimal feature subset is constructed to improve the precision and robustness of the prediction model.

The invention discloses a fine-grained image classification method based on large model enhancement, which guides a model to learn a specific judgment mode of a task by directly introducing prior knowledge. A series of image descriptions are generated by performing question and answer interaction with a multi-modal large language model (MLLM). In order to filter illusion information and redundant content existing in description, a dual-guide text feature optimization module is introduced, and the quality of text features is improved through task-guided feature selection and similarity-guided feature pruning. And finally, adopting a multilayer network structure based on an attention mechanism to realize vision-language fusion for final classification prediction.

A method for previewing the terrain in front of an emergency rescue vehicle in a geometric feature degradation scene relates to the technical field of road surface recognition, realizes accurate alignment of LiDAR point cloud and IMU data through timestamp synchronization and linear interpolation, and combines a spherical projection model and a degradation perception complementary feature selection algorithm to realize the previewing of the terrain in front of the emergency rescue vehicle. Converting the three-dimensional point cloud into a robust intensity image and extracting gradient significant features; based on a double-observation secondary filter frame, a motion state is predicted by utilizing IMU forward propagation, a point cloud geometric residual error and an intensity image luminosity residual error are synchronously fused, timestamp deviation is compensated through back propagation, and a multi-source observation model under a global coordinate system is constructed. Finally, the error state is iteratively optimized to realize collaborative output of the high-precision odometer and the three-dimensional terrain map, and the problems of data asynchronism, feature degradation and dynamic interference in a complex scene are effectively solved.

The invention relates to a lightweight neural network target detection model optimization method for an embedded device, and belongs to the technical field of target detection, and the method comprises the steps: determining the hardware constraint and detection task boundary of the embedded device; adjusting a network infrastructure based on hardware and task characteristics; network parameter redundancy is eliminated; designing a scene adaptive dynamic feature selection mechanism; compressing parameter storage precision; optimizing calculation-intensive operation; adjusting a memory access mode; an embedded deployment framework is adapted; and constructing a closed-loop iterative optimization process. The method is beneficial for realizing lightweight and efficient operation of the model, improving detection precision and stability, reducing memory occupation and power consumption, and meeting hardware constraints of embedded equipment.

The invention discloses a speech emotion recognition method and system based on multi-modal feature fusion, and relates to the technical field of speech emotion recognition. The speech emotion recognition method and system based on multi-modal feature fusion comprises the following steps: S1, collecting a speech emotion data set, and carrying out label unified coding and normalization processing; s2, frame-level acoustics and construction of frequency spectrum, rhythm and sound quality emotion features are carried out; s3, a speech emotion representation generation method fusing multi-sub-mode depth coding and a gating cooperative attention mechanism; s4, performing random forest weight initialization and two-order variation grey wolf mapping evaluation; and S5, voice emotion recognition and operation feedback adaptive updating are carried out. According to the method, the feature selection efficiency and the emotion classification accuracy in voice emotion recognition are effectively improved, and the problems that existing voice emotion feature selection is single in stage and single in index, emotion retention and real-time performance are difficult to consider while dimension reduction is performed, and the overall performance is limited are solved.

A system obtains high-resolution data of actual over-ground gait rehabilitation interactions through a custom-made wearable system for identifying abnormal gait patterns and the therapists' assistance strategies. The system implements an impedance learning algorithm with feature selection and a goal-directed attractor definition reproduces therapist assistance in a way that integrates clinical insights into the control of lower-limb exoskeletons.

The invention relates to a P91 pipeline aging prediction method, and belongs to the field of P91 pipeline aging prediction. The method comprises the following steps: acquiring a micro-magnetic signal (such as Barkhausen noise) of a P91 pipeline through a micro-magnetic detector, and extracting characteristic parameters related to an aging state; and inputting the parameters into a pre-trained aging grade prediction model, and directly outputting the aging grade (1-5 grade) of the pipeline. The prediction model is constructed based on a machine learning algorithm (such as ReliefF feature selection and a BP neural network), and training data comes from an accelerated aging sample which is simulated through specific heat treatment and has a clear microstructure (represented by EBSD) and macromechanical property correlation. According to the method, rapid, lossless and quantitative prediction of the aging state of the P91 pipeline is realized, and a reliable basis is provided for safety assessment and residual life prediction of the in-service pipeline.

The invention discloses a cloud resource dynamic scheduling method, system and device and a storage medium, relates to the technical field of cloud computing resource management, and provides basic data support for resource demand prediction by collecting multi-dimensional core resource real-time data. The input feature weight is dynamically adjusted by means of an LSTM neural network model fused with a dynamic feature selection mechanism, and the short-term fluctuation and long-term trend of the resource demand are respectively captured through an LSTM structure, so that the time sequence analysis better fits the multi-scale characteristics of the load, and the accuracy of resource demand prediction is improved; a prediction result and a current resource configuration state are integrated through a benefit function, overall consideration of economic constraints and service quality requirements is realized, iterative solution is carried out in combination with an optimization algorithm, and a resource scheduling scheme can balance multiple objectives; meanwhile, the model design naturally has adaptive capacity to complex load changes, the deviation between resource configuration and actual requirements is reduced, and accurate prediction and efficient dynamic configuration of cloud resource requirements are achieved.

The invention relates to the technical field of spectral analysis and biomedical detection, and discloses an SERS (Surface Enhanced Raman Scattering) spectrum quantitative detection method and system based on interpretable stacked ensemble learning. The method comprises the following steps: acquiring SERS spectral data of serum tumor marker standard substances with different concentration gradients; performing baseline correction and normalization preprocessing on the data, performing sparse feature selection by using an LASSO algorithm, and screening out key spectral features to construct a sample data set; constructing an interpretable stacking integration model, wherein the model adopts a base learner layer and a meta learner layer; training the model by using the training set, optimizing model hyper-parameters by using a cross validation strategy, and establishing a mapping relationship between spectral features and tumor marker concentrations; and collecting SERS spectral data of a to-be-detected serum sample, extracting key spectral features, inputting the key spectral features into the trained interpretable stacked integrated model, and outputting a concentration predicted value of the tumor marker in the to-be-detected serum sample. The method has the advantages of high precision, universality and molecular level interpretability.

The invention relates to an automatic measurement method and device, computer equipment, a medium and a program product. The method comprises the following steps: in response to a measurement parameter configuration request for a target workpiece, obtaining and displaying a standard workpiece image corresponding to the target workpiece; in response to image positioning feature selection information, extracting image positioning features from the standard workpiece image, and constructing and displaying a standard image coordinate system corresponding to the standard workpiece image; in response to the measurement point configuration information for the standard workpiece image, determining at least one standard measurement point included in the standard workpiece image and a target measurement mode corresponding to the standard measurement point; performing feature matching in the target workpiece image according to the image positioning features, and constructing a target image coordinate system corresponding to the target workpiece image; and measuring the target measurement point according to the target measurement mode to obtain a measurement result corresponding to the target workpiece. By adopting the method, the workpiece measurement efficiency can be improved.

Techniques are disclosed relating to feature selection based on feedback-assisted optimization models. In various embodiments, for example, the disclosed techniques include accessing a training dataset that includes a plurality of data samples that include data values for a plurality of features, and a set of labels corresponding to the plurality of data samples. In some embodiments, a computer system performs feature-selection operations to select, from the plurality of features, a subset of features to include in a reduced feature set. For example, in some embodiments the feature-selection operations include processing the training dataset based on an optimization model, where an objective function utilized in the optimization model utilizes performance feedback information corresponding to machine learning models that are trained based on candidate feature sets. Based on the feature-selection operation, the computer system may generate an output value that indicates the subset of features to include in the reduced feature set.

The invention relates to the technical field of intelligent fault prediction, and provides a multi-dimensional time sequence anomaly detection method for the process industry, which aims at historical and real-time multi-source data of the process industry, performs adaptive feature extraction to obtain better features through a feature selection module fusing mRMR and a self-attention mechanism, and improves the detection accuracy. Multivariable time series data double-fusion prediction calculation is completed through a sequence prediction model, the incidence relation of data at different time points and the incidence relation of different features at the same time are mined, and after cross entropy loss and optimizer training optimization are carried out, abnormal state judgment is finally achieved in combination with a dynamic threshold value. The invention further discloses a system used for the method, the method and the system can mine mechanism information implied by equipment from historical multi-source data in the process industry, time and features are subjected to relevance fusion respectively, and therefore the method and the system can be more suitable for the unsteady state and strong coupling conditions of the process industry; and particularly, a remarkable effect is achieved in a slag grinding system.

The invention discloses a lightweight SDN (Software Defined Network) attack detection method based on multi-scale iterative attention, relates to the technical field of network security, and solves the problem that an SDN attack detection method based on deep learning in the prior art is insufficient in feature selection static state and spatial modeling and gives consideration to both lightweight and high precision. The method is based on a feature contribution degree evaluation mechanism, the most critical features for attack discrimination are screened out in real time, redundant information is eliminated, and the calculation burden is reduced. Moreover, the attack feature map is generated through normalization, time window overlapping slicing and multi-channel space coding, so that the perception capability of a complex attack mode is improved. Besides, a multi-scale iteration attention mechanism is embedded in a lightweight network architecture, key features are highlighted and redundant information is suppressed through multi-granularity convolution extraction and iteration weight fusion, and both lightweight and high-precision detection are realized, so that the method is suitable for real-time network environment and edge device deployment.