209 results about "Spectral domain" patented technology

The invention discloses a self-adaptive frequency spectrum monitoring and interference suppression method for a railway power transformer. The method comprises the following steps: S1, collecting original multi-source signal data; s2, performing high-order filtering and Z-score normalization processing on the original multi-source signal; s3, inputting the original multi-source signal into a multi-scale residual fusion time-frequency transformation network, and extracting a time-frequency feature tensor; s4, inputting the time-frequency feature tensor into the interference identification network fused with the attention mechanism; s5, dynamically activating an interference suppression module according to an identification result; s6, constructing a multi-dimensional tensor data structure, extracting sparse dictionary morphological features and spectral domain statistics, and generating a composite feature vector set; s7, inputting the composite feature vector into a health state evaluation module; and S8, uploading the diagnosis result to a remote monitoring platform through the embedded communication module. According to the invention, multi-dimensional perception and adaptive modeling are fused, and intelligent identification and remote monitoring of railway transformer faults are realized.

The invention provides a modified metal surface defect detection method and device and a medium, and the method comprises the steps: obtaining a multi-angle reflection image sequence of a modified metal surface under the irradiation of a multi-spectral light source, and generating a defect sensitive parameter set based on a preset modified metal material characteristic database and in combination with the spectral reflectivity distribution information of the multi-angle reflection image sequence; performing cooperative feature enhancement on the multi-angle reflection image sequence and the defect sensitive parameter set, and enhancing the feature contrast of a defect area and a normal area through weight distribution to obtain an enhanced defect feature set; joint anomaly detection of a spatial domain and a spectral domain is carried out on the enhanced defect feature set, a potential defect region of the modified metal surface is obtained through identification, and a defect region feature descriptor is generated; and performing defect morphological quantitative analysis according to the defect region feature descriptors, and determining the type, position and severity level of the modified metal surface defect. According to the invention, the comprehensiveness and reliability of a defect detection result can be improved.

The invention discloses a liquid crystal display screen gray scale control method based on adaptive brightness compensation, which relates to the technical field of liquid crystal display screen control, and comprises the following steps: collecting environmental spectrum distribution information and a pixel gray scale voltage response curve of a liquid crystal display screen in real time, and carrying out fusion processing to form a time sequence joint feature vector; through multi-mode synchronous sampling and time sequence joint feature vector generation, ambient light and screen content changes are accurately captured, target adjustment values of backlight power and gray scale voltage are dynamically predicted, and response lag and overshoot are avoided. Through spectral domain balancing and a self-adaptive damping scheduling mechanism, cooperative control of backlight and gray scale is optimized, high-frequency oscillation is suppressed, and linear response of a low-frequency band is ensured. Meanwhile, Gamma mapping and global color gamut calibration are dynamically refreshed, accurate adjustment of area-level brightness and gray scale is achieved, and continuity of the display effect and visual experience of a user are improved.

The invention relates to the technical field of tunnel crack detection and artificial intelligence edge calculation, and provides a tunnel apparent disease detection method based on deep learning and knowledge distillation, which comprises the following steps: step 1, introducing spectral domain information enhancement to an original tunnel image, the edge texture features of the disease area in the image are enhanced through methods such as multi-scale wavelet transform and small-scale enhancement. Step 2, constructing a high-performance teacher model, introducing a flexible up-sampling structure to adapt to feature recovery requirements of different levels of semantic information, introducing an efficient visual coding module to enhance feature fusion capability of different scale channels, and designing a scale adaptive weighted loss function at the same time; by introducing a frequency spectrum enhancement mechanism, structural features of disease areas with low contrast, fuzzy edges and the like are remarkably enhanced in an image preprocessing stage, clearer information input is provided for a model, and the stable recognition capability of a system in environments of uneven illumination, complex background and the like is enhanced.

The invention discloses a remote sensing image multi-scale segmentation method based on frequency spectrum information processing and Mama spatial modeling, and the method comprises the steps: carrying out the preprocessing of a remote sensing image, and obtaining an input image; inputting the input image into the trained remote sensing image segmentation model; the remote sensing image segmentation model comprises an initial convolutional layer, a spectral domain information processing unit branch, a Mama layer branch and a segmentation head; performing feature extraction on the input image through the initial convolutional layer to obtain initial image features; combining a frequency spectrum domain information processing unit branch and a Mama layer branch, and performing feature extraction on the initial image features to obtain fusion features; and performing up-sampling decoding and pixel-level classification on the fusion features through the segmentation head, and outputting a multi-scale segmentation result corresponding to the remote sensing image. According to the method, the spectral domain and the spatial domain are combined for feature extraction, and the common problems of fuzzy details, unclear boundaries, insufficient multi-scale target expression and the like in remote sensing image segmentation are solved.

The invention discloses an optical cable micro-disturbance identification method based on physical simulation and self-supervised time sequence decoupling, and relates to the technical field of optical cable identification, and the method comprises the steps: constructing a physical digital twin simulator, and generating a high-fidelity training set; constructing a deep learning model, wherein the deep learning model adopts a lightweight time sequence decoupling network; training the model by adopting a staged training strategy, and sequentially carrying out self-supervised noise distribution pre-training, simulation supervised training and spectral domain physical consistency fine tuning operation; inputting DAS time sequence data collected in real time into the trained model, and outputting the data as an optical cable identity ID and a physical position; the lightweight time sequence decoupling network comprises a physical guide preprocessing module, a lightweight U-Net separation module, a sparse gating module and an intelligent parallel decoding module. Through the technical means of simulation-driven data generation, staged training strategies and the like, the defects of the prior art in the aspects of reducing the data cost, improving the detection capability in a low SNR environment, realizing multi-source blind source separation and the like are overcome.

The invention discloses a multi-channel dynamic hypergraph sentiment analysis method and a multi-channel dynamic hypergraph sentiment analysis network fusing time sequence consistency, belongs to the field of artificial intelligence and multi-modal sentiment calculation, and aims to solve the problems existing in the existing sentiment analysis technology. The method comprises the following steps: S1, a multi-channel feature extraction step: extracting multi-channel features of a text mode and an audio mode through a heterogeneous pre-training model; s2, a local time sequence context fusion step based on a video number: fusing short-term emotional fluctuation based on a local context mechanism of the video number, and capturing long-range dependence across time dimensions through Transform; s3, a single-modal-multi-modal hypergraph collaborative prediction step: dynamically constructing a single-modal hypergraph and a multi-modal hypergraph in a training batch, and modeling a high-order relationship by adopting spectral domain-spatial domain hybrid convolution; and S4, a multi-level multi-branch supervision step: outputting a final emotion prediction result through joint optimization of an early MLP branch and a late hypergraph branch.

The invention relates to the technical field of optical processing, and discloses an imaging depth adjustable laser optical processing system, which comprises a laser emission unit, an optical conduction unit, a spectral domain optical coherence tomography imaging unit, a laser galvanometer scanning unit, an indication light imaging unit, an image processing display unit and a control unit, according to the spectral domain optical coherence tomography imaging system, time scanning of high output wavelength is realized by utilizing a wavelength scanning mode that the scanning telescopic rotating module drives the spectral domain optical coherence tomography imaging system, and position information or image information of a sample from the surface to the interior is provided in real time in combination with an image processing technology; the three-dimensional galvanometer scanning unit scans the position of a sample in real time and records the position and orientation information of a scanning point at the same time, real-time image information and different depths, widths and axial positions of the sample are scanned at the same time, the focusing depth and wavelength of an imaging laser beam and the focusing position and energy of the scanning laser beam are fed back and adjusted, and the laser beam is positioned to the sample for machining.

A system for controlling an operation of an induction motor is provided. The induction motor includes circuitry and a memory having instructions stored thereon that, when executed by the circuitry, causes the system to collect time-domain measurements of a stator current of the induction motor operating under varying conditions. The system transforms the time-domain measurements into a spectral domain using a sequence of STFTs based on sliding time windows over the time-domain measurements. The system performs spectral analysis in the spectral domain of the stator current to determine harmonics of different types present in the stator current of the induction motor and stabilize the determined harmonics to a shape of corresponding harmonics of the induction motor when operating under steady-state conditions. The system performs one or a combination of control, fault detection, and / or monitoring of the induction motor based on the stabilized harmonics of the induction motor.

The invention provides a wind, light and water storage double-layer robust optimization scheduling strategy construction method and system, and relates to the technical field of energy management, and the method comprises the steps: firstly obtaining multi-source data of a wind power plant, a photovoltaic power station and a cascade hydropower station; performing feature extraction and integrated prediction based on the multi-source data to generate a wind-solar power prediction result; combining hydrodynamic coupling modeling analysis to obtain a hydropower station power generation capacity and adjustable dynamic water volume sequence; the method comprises the following steps: constructing a multi-layer energy coupling graph and performing spectral domain structure optimization to form a power coupling network representing system space-time correlation and ecological sensitivity; and finally, adopting double-layer distributed robust optimization, synchronously optimizing the economic target and the ecological risk constraint, and generating a collaborative scheduling strategy. According to the method, the time sequence matching problem between the wind and light volatility and the hydroelectric time lag is effectively solved, the renewable energy consumption capability and the water resource utilization efficiency are improved, and the ecological operation risk is reduced.

The invention provides a DCIM-based method for rapidly calculating a green function of a uniaxial anisotropic layered medium and an electromagnetic simulation system. The method comprises the following steps: modeling a uniaxial anisotropic layered medium to be simulated; determining a uniaxial anisotropic dielectric material of each layer, and assigning transverse parameters and longitudinal parameters; determining longitudinal coordinates of field point and source point distribution required to be solved, fitting a uniaxial anisotropic medium Green function spectral domain expression under the field source distribution by using a DCIM method, and solving an analytical expression in a spatial domain by using Somerfeld identical equation properties; solving the size of the electromagnetic field on the plane according to the fitted analytical expression; compared with a direct numerical integration method or a finite element method numerical solution using the Green function, the method has the advantages that the airspace analytical expression corresponding to the field source distributed on the plane is directly fitted, so that the solution of the airspace Green function is greatly accelerated, and the calculation efficiency and the calculation precision are both good in an actual example.

The invention belongs to the technical field of remote sensing, and discloses a wolfberry planting distribution high-precision remote sensing monitoring method based on multi-temporal fusion and spectral feature optimization. According to the method, a multi-temporal and multi-spectral satellite image is used as a data source, and preprocessing and multi-temporal image fusion are firstly carried out; the method comprises the following core steps: establishing a time sequence characteristic curve according to a unique phenological period (such as bare soil characteristics in a dormancy period and high vegetation coverage in a rapid growth period) of wolfberry; in a spectral domain, screening out a characteristic spectrum dimension combination with the highest discrimination degree between the wolfberry and other crops through a characteristic wave band optimization algorithm (such as vegetation index difference degree and red edge characteristics); and in combination with an object-oriented classification or deep learning classification model, constructing a space-time coupling classifier, and performing high-precision extraction and distribution mapping on the Chinese wolfberry planting region. The method can effectively solve the problem of confusion classification of Chinese wolfberry and similar ground features (such as other shrubs and orchards), and realizes rapid and accurate monitoring of Chinese wolfberry planting area and spatial distribution.

The invention discloses a vector flow velocity measuring device and method based on a bidirectional scanning spectral domain optical coherence tomography technology. Comprising a super light-emitting diode light source, an optical fiber coupler, a polarization controller, an interferometer comprising a sample arm and a reference arm, and a collimating lens, the collimating lens, the grating, the focusing lens group and the linear array camera are sequentially connected through a light path; the linear array camera, the computer and the two-dimensional galvanometer are sequentially connected through a line; according to the method, the fixed bias vector velocity is introduced by utilizing the bidirectional scanning technology, the dynamic light scattering optical coherence tomography method and the Doppler optical coherence tomography method are combined, decoupling of each parameter of the vector flow velocity is efficiently and accurately realized, and a quantitative index is provided for vector flow velocity analysis. The vector flow velocity measuring device based on the bidirectional scanning spectral domain optical coherence tomography technology is simple in system setting and low in cost, the time resolution is improved by about 2.5 times, and the measuring time is shorter.

The invention discloses a hyperspectral image super-resolution reconstruction method based on double-domain information enhancement, and relates to the technical field of new-generation information. According to the method, a to-be-predicted low-resolution hyperspectral image and a to-be-predicted high-resolution multispectral image corresponding to the to-be-predicted low-resolution hyperspectral image are input into a hyperspectral image super-resolution reconstruction network model based on double-domain information enhancement, forward propagation is performed once, and then a predicted reconstructed high-resolution hyperspectral image can be obtained. According to the method, when prediction is carried out based on the low-resolution hyperspectral image and the high-resolution multispectral image, spatial domain information and spectral domain information in different data sources can be fully mined, and the spatial domain information and the spectral domain information are fully fused to form complementary advantages; and a reconstructed high-resolution hyperspectral image containing more texture detail information and spectral information is obtained. Moreover, tests show that the reconstructed high-resolution hyperspectral image is closer to a true value image.

The invention relates to the technical field of big data, in particular to a personalized recommendation method and device based on structure and behavior layering and a medium, and the method comprises the steps: extracting global features and graph structure information of interaction data of a user and an article; modeling a personalized portrait of the user, and generating a filter coefficient; carrying out hierarchical coefficient fusion, and executing mixed spectral domain atlas filtering to obtain a personalized atlas filtering score; obtaining a basic recommendation score through multi-source score fusion, and executing dynamic score modulation to obtain a final recommendation score; and generating a recommendation list based on the final recommendation score, and returning the recommendation list to the user. According to the method, through layered and two-dimensional personalized modeling, a recommendation engine can dynamically adjust the signal response of each user in a low frequency-high frequency and global-personalized spectral domain, so that the diversity, novelty and interpretability of a recommendation result are remarkably improved while the recommendation precision is ensured.

The invention provides a space-spectrum Mama hyperspectral image classification method based on a pyramid structure. The problem that an existing method is insufficient in performance in classification of complex backgrounds and fine-grained ground objects is mainly solved. Comprising the following steps: 1) acquiring a hyperspectral image, and constructing a training set and a test set; 2) designing a double-branch model based on multi-scale space-spectrum adaptive fusion, and extracting multi-scale space and spectrum feature information in parallel; 3) respectively designing a spatial feature extraction module and a spectral feature extraction module, capturing target information in a spatial domain and a spectral domain, and optimizing fusion of spatial spectral features by using a multi-scale adaptive weighting mechanism; 4) constructing a space spectrum interactive fusion module for deep interactive fusion of features extracted by space and spectrum branches; and 5) training the model until convergence, and obtaining a final classification result by using the model. The method can effectively improve the processing capability of the complex background in the image, enhance the ground feature classification precision, and significantly improve the hyperspectral image classification performance.

The invention relates to the technical field of deep sleep management, in particular to a deep sleep staging management method and system based on electroencephalogram feedback. Comprising the steps that electroencephalogram feedback of a target user is collected, an actual deep sleep waveform is extracted and judged with an expected spectral domain of an ideal state transfer stage, and an unreached stage is locked; in an unreached stage, constructing a signal mode prediction model, and determining an optimal guide window stage by combining with intervention entry point migration analysis; performing wavelet basis decomposition to obtain an excellent wave band and a clutter band of the optimal guide window period; when the signal-to-noise ratio is high, crosstalk coupling analysis is executed, and white noise is generated by using the rhythm coding library for guide management; when the signal-to-noise ratio is low, light environment and temperature response regulation is planned, and deep sleep is managed in combination with a color coding chain and a signal excitation result. According to the method and the system, guide management of different decisions can be executed in the stage of entering the deep sleep of the target user, and the deep sleep quality, stability and continuity are improved.

The invention discloses a data enhancement method for hyperspectral target detection and a related device, and relates to the field of target detection, and the method comprises the steps: precisely segmenting pixels into a target, a background and a boundary buffer region based on a hyperspectral image truth value annotation graph, calculating the average spectrum of the target and background regions, and the like. And constructing a distance gradient field by using the boundary buffer region, and calculating abundance values of the target and the background at set pixels according to the distance gradient field and a nonlinear abundance function. And determining an adaptive abundance matrix by physical constraints, and mixing the spectrums of the two regions according to abundance values by using a spectrum mixing model to generate mixed pixel simulation data. And finally performing enhancement processing on the data in a spatial domain, a spectral domain and a time phase domain. According to the invention, intelligent generation of high-quality hyperspectral training data is realized.

The invention discloses a self-adaptive presswork binding method based on pressure feedback, and relates to the technical field of presswork binding. The method comprises the following steps: firstly, acquiring data such as a paper laminated structure and stress distribution by utilizing equipment such as a terahertz time-domain spectrometer, and analyzing paper components and coating thickness; secondly, combining a quantum annealing model with a digital twinning technology to optimize binding parameters, and predicting a wrinkle risk; during binding, the system establishes a pressure-displacement curve, piezoelectric ceramic micro-vibration friction reduction is adopted, and ultrasonic local softening is triggered if pressure is abnormal; after binding, detecting the quality by means of spectral domain OCT and Raman spectrum, and embedding a digital watermark; and finally, constructing a knowledge graph based on a graph neural network and a block chain, rapidly adapting to a new paper type through meta-learning, and realizing edge end deployment by using knowledge distillation. According to the scheme, the binding qualification rate is greatly increased, the equipment loss is reduced, the new paper adaptation time is shortened to 5 minutes from 2 hours, and the production efficiency is increased by nearly two times.

The invention relates to the technical field of point cloud registration, in particular to a multi-view point cloud data registration method and system. The method comprises the following steps: acquiring single-view point cloud data to perform adjacent field configuration construction to obtain local structure chart data; performing resonance mode analysis on the local structure diagram data to obtain diagram embedded data; performing geometric potential energy mapping according to the graph embedded data to obtain morphological stability distribution graph data; performing spectral domain supporting point family extraction according to the morphological stability distribution diagram data to obtain a skeleton topology node set; constructing cross-view structure chart data according to the skeleton topology node set corresponding to the multiple pieces of single-view point cloud data; performing semantic perception fingerprint generation on the cross-vision-field structure chart data to obtain structure recognition cluster data; obtaining different-source structure coupling graph data according to the structure identification cluster data; and performing low-degree-of-freedom rigid body calculation according to the heterogeneous structure coupling graph data to obtain point cloud registration data. According to the invention, the stability and precision of multi-view point cloud registration are improved.

The invention discloses an extreme short and temporary rainfall forecasting method and system based on wavelet domain decoupling and multi-course learning, and belongs to the technical field of meteorological big data processing and artificial intelligence deep learning. According to the invention, a meteorological image sequence is decomposed into a low-frequency approximate component and a high-frequency detail component through frequency domain decoupling, a double-branch deep neural network is adopted to extract features and predict corresponding frequency domain coefficients, and a rainfall prediction map is output through fusion of a spectral domain reconstruction and refining module. In addition, a multi-task course learning strategy is introduced during model optimization, and the loss weight is dynamically adjusted. The method is characterized in that modeling is moved to a wavelet domain to explicitly protect high-frequency details, local severe convection is accurately captured in combination with a double-branch architecture and a feature pyramid, the problem of traditional prediction fuzziness is effectively solved, the prediction definition and the extreme rainfall early-warning capacity are remarkably improved, high-fidelity and high-precision minute-level short-temporary rainfall prediction is achieved, and the prediction efficiency is improved. And reliable technical support is provided for meteorological disaster prevention and reduction.

The invention discloses a high-fidelity visual character image generation method, system and device and a storage medium, which are corresponding schemes, in the scheme, spatial domain and spectral domain feature mutual learning is realized based on a double-domain font coding mechanism, and the problem of character structure disorder is solved; based on a frequency domain perception refining mechanism, a frequency domain signal is dynamically modulated to improve stroke detail precision, collaborative optimization of character details and background quality is realized through an instance-level scaling coefficient, and the bottlenecks of character distortion and background instability in the prior art are broken through. Besides, the method supports multi-language accurate generation, can be widely applied to commercial scenes such as advertisement design, product packaging, film and television posters, teaching materials and the like, and can greatly improve creation efficiency and reduce manual correction cost; and meanwhile, the feasibility of cooperatively improving the multi-modal generation efficiency by the cross-domain features is verified, a high-quality weak supervision training sample in a scarce scene can be provided for a character recognition model, the blank of related data is effectively filled up, and the method has technical advancement and market landing value.

The invention belongs to the technical field of electric power, and provides a user load increase and probability boundary prediction method, which comprises the following steps of: firstly, realizing unified projection and deep fusion of multi-source heterogeneous data by adopting a topological adaptive space-time fusion mechanism; secondly, physical constraint comparative learning and hypergraph clustering are utilized to generate a load-sensitive topology mode according with a power grid rule; a CPT-Mama model is constructed by injecting physical rules, meteorological codes and lightweight NTK features, and spatial correction is performed by means of a graph attention network to improve the precision of load increment prediction, so that efficient prediction of the multi-modal load increment is realized; and finally, innovatively introducing a net rack spectral domain built-in operator, reconstructing a residual error and quantifying uncertainty, and accurately describing a probability boundary of load increase. According to the method, the medium and long term load prediction precision is improved, the crossing from point prediction to probability prediction is realized, and reliable technical support is provided for power grid dispatching, risk prevention and control and planning decision.

This invention discloses a method for camouflaged target detection, belonging to the field of camouflaged target detection. This method constructs a BiSAM based on the SAM network. The SAM encoder, CNN encoder, and dual-domain cascaded attention module BCAM work collaboratively to enrich multi-scale feature information and extract multi-source features from the input image. Specifically, BCAM operates collaboratively in the spatial and spectral domains, selectively recovering and enhancing high-frequency object boundaries while suppressing irrelevant background noise. The multi-source features are fused using AKM to guide the SAM decoder in decoding the ViT image embedding. The method replaces the original position encoding of the cross-attention key vector in the SAM decoder, injecting task-specific knowledge into the frozen SAM decoder. By replacing the original position encoding, the general attention key vector is transformed into a task-aware attention key vector, specifically adapted to camouflaged target detection scenarios, thereby improving the accuracy of camouflaged target detection while maintaining generalization ability.

An imaging system is provided. A first imaging system captures initial sensor data in a form of visible domain data. A second imaging system captures subsequent sensor data in a form of second domain data, wherein the initial and subsequent sensor data are of different spectral domains. A controller subsystem detects at least one region of interest in real-time by applying a machine learning technique to the visible domain data, localizes at least one object of interest in the at least one region of interest to generate positional data for the at least one object of interest, and autonomously steers a point of focus of the second imaging system to a region of a scene including the object of interest to capture the second domain data responsive to the positional data.

The invention provides a'vision + behavior 'information transmission method, which is characterized in that interactive nodes on a data chain are regarded as intelligent agents, a machine observes behaviors of other machines through a photoelectric sensor and receives and accepts information transmitted by the other machines, and data chain information flow is established on a'vision + behavior' information transmission mode. According to the invention, the interaction mode of the data link is innovated, the serious dependence of the data link on radio frequency communication is reduced, and the novel consensus capability of electromagnetic-wave-free communication of the data link is realized. The data chain works in a spectral domain comprising visible light and invisible light, is not decoy and interfered by electromagnetism, and can face strong electromagnetic interference or transmit information under complete rejection. The problem that use of the data chain is limited in a denial environment is innovatively solved, thoughts are expanded for the data chain to seek electromagnetic breakthrough, and the concept category of the data chain is developed.

The invention discloses a dual-drive electromagnetic interference suppression method and device based on random PWM and impedance matching, and relates to the technical field of power electronics and electromagnetic compatibility, and the method comprises the steps: constructing an instantaneous PWM frequency modulation model based on a random PWM mechanism; determining a frequency dynamic disturbance boundary constraint function; constructing a digital controller with a spectral domain feedback function; based on an instantaneous PWM frequency modulation model, a frequency dynamic disturbance boundary constraint function and a digital controller, double-drive electromagnetic interference suppression is carried out on the high-frequency and high-density power electronic system based on multi-harmonic observation and feed-forward compensation. By constructing a frequency spectrum decoupling-impedance reconstruction-collaborative suppression system, the technical bottlenecks of large size and narrow coverage range of single random PWM modulation of a traditional passive filter are broken through, and the filter has the advantages of wide frequency band coverage and high suppression precision.

The invention discloses a camouflage target detection method, and belongs to the field of camouflage target detection, and the method comprises the steps: constructing a BiSAM, an SAM encoder, a CNN encoder and a double-domain cascade attention module BCAM on the basis of an SAM network, carrying out the cooperative work, commonly enriching the multi-scale feature information, and extracting the multi-source features of an input image; wherein the BCAM cooperatively operates in a spatial domain and a frequency spectrum domain, and selectively recovers and enhances a high-frequency object boundary while suppressing irrelevant background noise; the multi-source features are fused through AKM to obtain the multi-source feature fusion module, and the multi-source feature fusion module is used for guiding an SAM decoder to decode ViT image embedding; an original position code of a cross attention key vector replacing an SAM decoder is adopted, so that specific knowledge of a task is injected into the frozen SAM decoder, a universal attention key vector is converted into an attention key vector perceived by the task by replacing the original position code, and the method is specially adapted to a camouflage target detection scene. Therefore, the generalization ability is maintained while the accuracy of the camouflage target detection task is improved.