13040 results about "Frequency domain" patented technology

The invention discloses a thunder and lightning monitoring and early warning method and system based on multi-source data fusion, and relates to the technical field of thunder and lightning early warning, and the method comprises the steps: extracting electric field time domain and frequency domain features, magnetic field change features, lightning activity modes and meteorological change features through obtaining atmospheric electric field, magnetic field, lightning activity and meteorological environment data in real time; and constructing multi-source feature data. A time sequence analysis and Bayesian fusion technology is adopted to calculate a correlation weight between data sources, and a fusion feature vector is generated. And establishing a weighted regression model based on the vector, calculating thunder and lightning occurrence probability through dynamic weight distribution, and generating a risk distribution map in combination with geographic information. The method has a closed-loop feedback optimization mechanism, model parameters and weights can be adaptively adjusted according to prediction errors and early warning accuracy, the accuracy, timeliness and environmental adaptability of lightning early warning are improved, and the method is widely applied to the fields of electric power, aviation, buildings and the like.

The invention relates to the technical field of bearing fault detection, and discloses a bearing fault detection method and system based on a health state index. The method comprises the following steps: collecting multi-source sensing signals at least comprising a vibration signal, a temperature signal and an acoustic signal during bearing operation; respectively performing time domain feature extraction and frequency domain feature extraction on the multi-source sensing signals, and performing normalized fusion on the extracted time domain features and frequency domain features to generate a multi-dimensional health state index sequence; constructing a long-short-term memory network model based on an attention mechanism, inputting the multi-dimensional health state index sequence into the model for training, and outputting a bearing health state prediction sequence; and calculating a dynamic early warning threshold according to the historical health state prediction sequence, comparing the current prediction value with the dynamic early warning threshold in real time, and generating a fault early warning signal. The method can improve the accuracy of bearing health state evaluation and fault early warning, and is suitable for complex operation conditions.

The invention relates to the technical field of data processing, and discloses a comprehensive evaluation method and system for the health degree of a distribution network cable. The method comprises the following steps: collecting cable joint multi-source monitoring signals, normalizing the monitoring signals to obtain a degradation degree feature vector, correcting multi-physics field coupling model parameters, obtaining a recessive degradation index through finite element calculation to obtain an enhanced feature vector, and performing time-frequency domain decomposition to extract multi-scale feature parameters to obtain a comprehensive feature matrix; a double attention mechanism calculates a feature weight and a time sequence correlation degree to obtain a deterioration trend prediction value, and fuzzy integral is fused with a multi-classifier output probability to obtain a health degree evaluation grade and an early warning result. According to the invention, the early defect identification accuracy and the degradation trend prediction precision are improved.

The invention relates to the technical field of computer vision detection, in particular to an unmanned aerial vehicle target detection method based on frequency-space joint attention and dynamic fusion, and the method comprises the steps: obtaining an unmanned aerial vehicle image data set, carrying out the preprocessing, and dividing a training set and a test set; constructing a target detection model, inputting the training set into the target detection model to extract image features, sequentially performing frequency domain detail enhancement, spatial domain salient region extraction and multi-scale feature adaptive fusion based on the image features, and establishing a feature sequence; screening the feature sequence to obtain an initial target query, and finishing target classification and positioning on the initial target query through a decoder; training a target detection model by using the training set, and inputting the test set into the trained target detection model to generate a detection result; on the premise that the real-time reasoning advantage of RT-DETR is kept as much as possible, the problems that in an unmanned aerial vehicle scene, a target is prone to missing detection, the scale change is large, the background is complex, and the target is fuzzy are effectively solved, and the detection precision is improved.

The invention discloses an unmanned aerial vehicle image small target detection method based on dynamic filtering and an adaptive sparse Transform. According to the method, an end-to-end target detection framework is adopted, a dynamic filtering module is introduced into a backbone network, global feature interaction is achieved through data-dependent frequency domain operation, and linear calculation complexity is maintained. For feature interaction in a scale, an adaptive sparse Transform module is introduced to enhance the capability of focusing key information on high semantic hierarchy features of a model, and noise interference and feature redundancy are effectively suppressed at the same time. Through the combination of dynamic filtering and adaptive sparse Transform, the model can extract image foreground information more effectively on the premise of not significantly increasing the calculation burden, and the problem that a traditional target detection model is susceptible to complex background interference is significantly relieved.

The invention relates to the technical field of aircraft structure detection, and discloses an aircraft structure crack intelligent identification method based on deep learning. The method comprises the following steps: acquiring original vibration response signals and electromagnetic field distribution data on the surface and inside of an aircraft structure in parallel through a multi-source sensor network; synchronously processing the data by using a multi-scale convolutional neural network, and extracting time-frequency domain abnormal fluctuation features and space magnetic field distortion features; constructing a cross-modal correlation model, analyzing a topological dependency relationship of the two types of features through a graph attention mechanism, and generating a fused damage sensitive feature vector; inputting the vector into a pre-trained deep belief network to obtain a probability distribution mapping relation for different crack types; and according to the mapping relation, carrying out adaptive weighted fusion on original multi-sensor data, inhibiting environmental noise and structural background interference, and separating and reconstructing an accurate three-dimensional morphology map of the target crack. According to the method, multi-source data information can be effectively fused to improve the accuracy of aircraft structure crack identification.

The invention discloses a light guide plate defect detection method and system based on a neural network, and particularly relates to the technical field of machine vision detection, and the method comprises the following steps: aiming at the problem of image instability of a light guide plate in a dynamic transmission or rotation process, continuously collecting an image sequence and extracting time domain features; and performing interference judgment in combination with the inter-frame consistency prediction coefficient and a first threshold to realize accurate identification of the abnormal image frame. For an abnormal image frame, further correcting the recognition credibility of the abnormal image frame by adopting a confidence adjustment and fusion mode, and meanwhile, introducing a frequency domain transformation and image enhancement strategy to compensate detail loss caused by motion blur; according to the method, inter-frame consistency analysis, confidence fusion regulation and control and frequency domain fuzzy recognition and compensation mechanisms are introduced, abnormal judgment and image quality restoration of the light guide plate image in the dynamic scene are realized, the recognition accuracy and stability of the neural network model on the defect type, position and confidence are improved, and the false detection and omission ratio is effectively reduced.

The invention discloses a multi-modal remote sensing semantic segmentation method and system for learning frequency domain fusion. The method comprises the following steps: respectively extracting multi-scale features of two modal input images by adopting a double-branch encoder; sequentially executing frequency domain decoupling and fusion, mutual information constraint-based feature optimization and low-frequency guided cross-modal fusion processing on each scale feature to generate a fused semantic feature; and performing up-sampling and feature refining on the fused features through a decoder, and outputting a full-resolution segmentation prediction map. According to the multi-modal remote sensing image semantic segmentation method, modal sharing information and specific details are effectively separated through frequency domain decoupling, feature representation is optimized through mutual information constraint, adaptive feature fusion is achieved in combination with an attention mechanism, and the accuracy and robustness of multi-modal remote sensing image semantic segmentation are remarkably improved.

The invention discloses a semantic segmentation method for a low-resolution road scene, and aims to solve the problems of difficulty in small target recognition, fuzzy details, texture information loss and the like existing in a low-resolution image in the conventional semantic segmentation technology. The method comprises the following steps: (1) collecting a low-resolution road scene image and a corresponding semantic tag; (2) constructing a semantic segmentation model consisting of an edge guidance module (BGM), a double-domain feature decomposer (DDFD), a domain alignment attention fusion module (DAAFM) and a double-layer attention context aggregation module (HACAM); (3) designing a joint loss function to carry out multi-scale supervision on semantic regions, edges and middle features; (4) carrying out model training by utilizing the road scene image; and (5) outputting a semantic segmentation result map and an edge prediction map. The boundary perception capability is enhanced by introducing learnable pixel difference convolution, the extraction precision of a small target and a global structure is improved by combining frequency domain and spatial domain feature alignment, and context semantic relationship expression is optimized by fusing a channel and a spatial attention mechanism. The method effectively improves the semantic segmentation precision and boundary restoration capability of the model in a low-resolution complex road environment, and is suitable for intelligent analysis tasks of road images in scenes of automatic driving, intelligent traffic, severe weather and the like.

The invention relates to the technical field of optical fiber communication monitoring, in particular to a passive optical fiber multi-parameter digital twin drive abnormal root cause positioning method and system. Collecting temperature, stress, acoustics and polarization signals, and constructing a time domain, frequency domain and energy domain coupling feature tensor and time sequence data set; performing nonlinear dimension reduction and feature decoupling by using a beta-VAE model, and dynamically quantifying contribution of each parameter to anomaly by using an integral gradient to form a contribution degree vector group; constructing a PINN digital twinborn model embedded with heat conduction and elasto-optical effects, and predicting a normal fluctuation range under contribution vector weighting and data-physics dual constraints; modeling measurement and prediction deviations under the guidance of contribution vectors, and outputting an abnormal measurement score, confidence, a position and a time sequence; and a causal graph neural network is constructed, topology and abnormal events are fused for tracing causes, a fault source is identified, and the model is subjected to closed-loop calibration. According to the invention, data driving and a physical mechanism are fused, and accurate detection and root cause positioning of the abnormity of the optical fiber system are realized.

The invention belongs to the technical field of turbulent flow field prediction and deep learning, and discloses an incompressible turbulent flow field prediction method based on a potential diffusion model. The method comprises the following steps: acquiring original turbulence data; processing the turbulence data; constructing a turbulence prediction model; model training; and evaluating the model and the like. The model of the technical scheme of the invention specifically comprises the following steps: designing a multi-scale Fourier auto-encoder for extracting multi-scale space and frequency domain features in a turbulence field and obtaining a global structure and a local scale structure of turbulence; a novel accelerated sampling method is proposed and introduced in the diffusion process, namely a diffusion probability model solver greatly shortens the reasoning time in a potential space and keeps high fidelity in long-time-sequence prediction; a physical constraint loss item based on a partial differential equation is introduced, and a Navier-Stokes equation (N-S) is explicitly introduced into a training process, so that the physical consistency of results is effectively improved, and errors are remarkably reduced.

The invention relates to the technical field of electric power monitoring analysis, and discloses an intelligent analysis system for electric power monitoring data based on a mutual inductor. The system comprises a mutual inductor data acquisition and structuring module which acquires a current waveform, a voltage waveform and a harmonic component in real time, and generates a standardized monitoring data unit through preprocessing, field mapping and association identification establishment; the theoretical monitoring value calculation module constructs a dynamic calculation model according to historical power data, and an input standard data unit outputs a theoretical value; the rule conformity verification module is used for matching the equipment type with the operation rule and generating single compliance judgment; a multi-dimensional difference analysis module compares a theoretical value with a measured value from time domain deviation, frequency domain deviation and waveform distortion, and generates an equipment-level difference coefficient matrix; the association network construction module is used for constructing a multi-monitoring-point association map according to the equipment identifier and the position information; and the anomaly positioning and strategy generation module combines the map, compliance judgment and a difference matrix, calculates a risk score, generates an anomaly probability distribution map, positions anomaly and matches a monitoring strategy.

The invention discloses a power grid equipment state sensing driving dynamic response method based on the Internet of Things technology, and relates to the technical field of power system automation and informatization, and the method comprises the following steps: S001, collecting original multi-dimensional state signals of a plurality of sensors deployed in a power grid equipment state sensing channel, constructing an electromagnetic disturbance recognition model, and carrying out the recognition of the original multi-dimensional state signals; frequency domain and time domain feature extraction is carried out on the signals, and a feature comparison parameter set used for distinguishing electromagnetic interference and real faults is generated. Frequency domain and time domain features are extracted through an electromagnetic disturbance recognition model, dynamic threshold judgment and response strategy adjustment are achieved in combination with multi-source sensing data, interference and faults can be accurately distinguished, early warning and protection logic can be corrected in real time, protection actions can be accurately triggered, closed-loop control is achieved, the delayed fault tolerance and multi-source verification capacity is achieved, and the method is suitable for large-scale popularization and application. The false operation rate and the false stop risk are effectively reduced, and the intelligence, the safety and the stability of power grid operation are improved.

The invention discloses a multi-sensor fusion heat pump full life cycle AI maintenance early warning system, and relates to the technical field of new energy utilization, and the early warning system comprises a data collection module which obtains operation parameters in a heat pump full life cycle based on a sensor array, and constructs a data set after preprocessing the parameters; the operation parameters comprise temperature, pressure, flow and micro vibration; the data fusion module is used for extracting trend correlation characteristics and parameter coupling characteristics from temperature, pressure and flow parameters by adopting a dynamic sliding window adaptive to a working condition, and preserving core nonlinear information through KPCA dimension reduction; the micro-vibration signal extraction comprises frequency domain and time domain features. According to the method, features are extracted through a working condition adaptive dynamic sliding window, then through cross-space mapping and a life cycle-working condition double-attention mechanism, the analysis and early warning module depends on a core feature mapping library and a two-dimensional dynamic baseline, through instantaneous and accumulated deviation double judgment, abnormal accurate recognition and stage division are achieved, and early warning perspectiveness is high.

The invention discloses a machine vision-based intelligent detection method for steel galvanized surface defects, which comprises the following steps: S1, acquiring and preprocessing a steel galvanized surface image to obtain a standardized image; s2, constructing a specular reflection probability graph according to the brightness distribution and the gradient magnitude, and calculating a reflection intensity value; s3, calculating a structure tensor matrix, determining a main direction angle and an anisotropic consistency coefficient, and generating a direction feature matrix; s4, establishing a multi-scale direction adaptive phase kernel function, and performing phase modulation in a frequency domain by adopting an improved phase stretching transformation algorithm; s5, inverse Fourier transform is executed, and a phase response matrix is extracted; s6, performing weighted fusion to obtain a comprehensive phase response diagram; and S7, setting a threshold value according to the noise variance and the statistical characteristics, executing binarization and morphological processing, and outputting a defect region and boundary coordinates. According to the invention, high-precision identification and boundary positioning of steel galvanized surface defects are realized.

The invention provides an ocean intelligent forecasting large model construction method, and relates to the field of ocean forecasting, and the method specifically comprises the following steps: obtaining multi-source ocean observation data, and constructing a high-resolution ocean analysis data set which is subjected to quality control, space-time registration, standardization and data set division processing through multi-source observation data fusion and numerical mode assimilation; constructing a basic prediction model, and performing multi-scale fusion on the frequency domain enhanced features and the spatial local features by using the basic prediction model; the trained basic prediction model is operated in a set area range, and an output result of the basic prediction model is recovered to an original physical quantity value through inverse standardization; and comparing the rolling output of the basic prediction model with observation data or a high-resolution mode result through a correction module, learning an error, outputting a correction quantity, and superposing the correction quantity with an original prediction result to obtain a prediction field. According to the technical scheme, the problem that the ocean forecasting model in the prior art cannot meet the requirement of a complex application scene is solved.

The invention provides a training method of a document image tampering detection model and a tampering detection method and device.The training method of the document image tampering detection model comprises the steps that multi-scale visual domain features are extracted from a sample document image, and multi-scale frequency domain compressed sensing features are extracted from frequency domain information; acquiring tampered area edge mask data from the document image; fusing the multi-scale visual domain features and the multi-scale frequency domain compressed sensing features to obtain multi-modal fusion features; performing semi-supervised training on the multi-scale sensing network by taking the multi-scale visual domain feature as a sample feature of a first prediction head, taking the multi-modal fusion feature as a sample feature of a second prediction head, taking a real label or a pseudo label as a sample label and taking joint loss as a loss function to obtain a document image tampering detection model; according to the method provided by the invention, document image tampering pixel-level detection under low labeling cost is realized, and the detection precision of a document image tampering detection model is improved.

The invention relates to the technical field of metal defect detection, and discloses an aluminum alloy round aluminum rod surface defect image super-resolution method. The method comprises the following steps: acquiring a low-resolution original image sequence of surface defects of the aluminum alloy round aluminum rod, and synchronously acquiring gray value distribution at different illumination angles through a multi-channel optical sensor; constructing a dynamic degradation model according to pixel displacement of adjacent frames in the original image sequence, extracting cross-scale defect features in the original image sequence, and taking output parameters of the dynamic degradation model as spatial constraint conditions of a feature extraction network; and a high-resolution defect image is generated through the multi-stage residual error reconstruction network, the high-resolution image output by the reconstruction network is fed back to the dynamic degradation model, and the frequency domain response coefficient of the spatial fuzzy kernel function is updated to form closed-loop optimization. The identification degree of defect features is improved, and a reliable image data basis is provided for accurate detection of the surface defects of the aluminum alloy round aluminum rod.

A wireless transmit / receive unit (WTRU) may receive subband non-overlapping full duplex (SBFD) configuration information. The SBFD configuration information may be associated with subbands for uplink transmission and subbands for downlink reception. The WTRU may receive scheduling information associated with physical uplink shared channel (PUSCH) transmissions. The scheduling information may comprise a first frequency domain resource allocation (FDRA). The WTRU may transmit a first PUSCH transmission using a first frequency resource. The WTRU may determine that at least a second PUSCH transmission is to be sent using at least one OFDM symbol. The WTRU may determine that the first frequency resource overlaps. The WTRU may receive one or more of a second FDRA or a frequency offset for the second PUSCH transmission. The WTRU may determine a second frequency resource for transmitting the second PUSCH transmission. The WTRU may transmit the second PUSCH transmission using the second frequency resource.

The invention provides a gas ultrasonic transducer fast matching method, device and equipment and a storage medium, original measurement data such as flight time are obtained by deploying an ultrasonic transducer in a gas ultrasonic flowmeter in a gas conveying pipeline, and multi-dimensional data acquisition is carried out in combination with other sensor parameters. Wavelet noise reduction and dynamic time warping processing are carried out on collected signals, signal quality and time sequence consistency are improved, time domain, frequency domain, environment and statistical features are extracted, and multi-dimensional feature vectors are formed. Modeling is carried out through a time sequence feature branch and an environment feature branch, weighted fusion is carried out by adopting an attention mechanism, and the recognition capability of the model on key features is enhanced. A machine learning model is trained based on fusion features, a multi-objective loss function and a data enhancement strategy are adopted, the prediction precision and generalization ability of the model are improved, and a compensation value is output to calibrate the original traffic in real time. According to the method, the accuracy and stability of gas flow measurement in a complex environment are effectively improved, and the method has a good engineering application prospect.

The invention discloses an electrical equipment defect identification and alarm method and system based on deep learning. The method comprises the following steps: synchronously collecting and registering visible light and infrared thermal imaging images on the surface of power equipment, and constructing an instance segmentation network comprising a lightweight feature extraction network, a multi-scale feature fusion network and a frequency domain mask prediction branch; enhancing the diversity of training samples by adopting a generative adversarial strategy; based on the graph neural network, analyzing the incidence relation between the defects and the equipment topology and historical records, and deducing the defect causal relation and the risk level; generating interpretable alarm information including the thermodynamic diagram, the natural language report and the maintenance suggestion; real-time detection and deep analysis are realized by adopting an end-side cloud collaborative architecture; and the system performance is continuously improved through a closed-loop optimization mechanism. According to the method, high-precision defect detection under multi-modal data fusion is realized, the robustness and interpretability are high, and the operation and maintenance intelligence level of power equipment is remarkably improved.

The invention discloses an industrial part defect sample accurate generation method based on a conditional diffusion model, and belongs to the field of image processing and artificial intelligence. The method forms a closed-loop cooperative system by constructing four deep coupling modules of physical constraint noise scheduling, multi-scale feature coupling, double-domain feedback optimization and adaptive weight adjustment; a defect physical forming mechanism is converted into a dynamic noise scheduling strategy, deep interaction between condition information and a feature map is established at multiple levels of a diffusion network, quality closed-loop optimization is achieved through dual evaluation of a pixel domain and a frequency domain, and training weight is dynamically adjusted according to defect scarcity. And multi-scale accurate control is realized, a quality guarantee closed loop is established, the problem of data imbalance is effectively solved, and the performance of an industrial defect detection model is remarkably improved.

The invention discloses a radar echo extrapolation method and system based on frequency domain enhancement, and the method mainly comprises the following steps: obtaining and preprocessing a historical radar echo grayscale image sequence, generating a sequence sample through a sliding window, and dividing the sequence sample into a training set, a verification set and a test set; the method comprises the following steps: constructing a frequency domain enhanced U-Net network comprising an encoder-decoder structure, introducing a multi-scale deep convolution structure into an encoder and a decoder, and enhancing frequency domain features by using a frequency domain dynamic attention mechanism in jump connection; inputting the training set into the model for training by adopting a composite loss function comprising intensity weighted loss, frequency domain consistency loss and structural similarity loss; and inputting the test set into the trained model, and outputting a radar echo prediction result at a future moment. The method can be effectively applied to the fields of short-term and temporary weather forecast, severe convection monitoring and the like, and provides more accurate and reliable radar echo prediction support for meteorological disaster early warning.

The invention relates to the technical field of precision manufacturing, and discloses a multi-mode laser galvanometer calibration method and device. The method comprises the steps of obtaining a smooth angle sequence, comparing the smooth angle sequence with a calibration model, extracting an abnormal component if a comparison deviation is abnormal, and performing smooth processing to obtain a deviation vector; fusing the statistical characteristics of the light beam position and the deviation vector, predicting the light beam position, and generating a calibration parameter set if the light beam position exceeds a deviation range; extracting a key compensation item and optimizing a driving sequence, fusing frequency domain interference to generate a simulation track, and determining a stability index; and comparing the environment feature set, adjusting parameters to obtain an enhanced path model, and determining a final calibration scheme in combination with real-time feedback. According to the method, accurate calibration of the galvanometer in a multi-mode dynamic scene can be realized.

The invention discloses a bearing defect detection method and system based on machine vision and ultrasonic detection, and particularly relates to the technical field of industrial automatic detection, and the method comprises the steps: S1, a synchronous collection module: carrying out pulse triggering synchronous collection, and generating a time-space reference table; s2, a feature extraction module: performing image noise reduction segmentation and ultrasonic frequency domain decomposition, and outputting a defect feature vector; s3, a fusion identification module: performing cross-modal feature alignment fusion to generate a defect classification conclusion; s4, a size measurement module: performing contour fitting to calculate inner and outer diameters, and outputting a size deviation value; and S5, a comprehensive judgment module: carrying out threshold comparison logic judgment, and generating a multi-modal detection report. According to the method, a space-time reference is established through an encoder, images are segmented in a self-adaptive mode, features are extracted through wavelet decomposition ultrasound, feature weights are re-calibrated through a parallel network and an attention mechanism, composite defects are recognized through cross-modal fusion, comprehensive judgment is conducted in combination with dimensional deviation, and a multi-dimensional quality evaluation system is achieved.

A system and method are disclosed for low-light image enhancement using hierarchical adaptive wavelet decomposition with cross-scale feature fusion. The system analyzes a raw input image to determine image characteristics and preprocessing parameters. A hierarchical adaptive wavelet decomposition process creates a variable-depth decomposition tree comprising frequency domain nodes, with decomposition depth determined by local image complexity. Cross-scale feature fusion implements attention mechanisms between nodes at different decomposition levels, enabling bidirectional information flow across scales. A dynamic network pool allocates specialized neural networks to process nodes based on their frequency characteristics, with weight sharing between similar nodes for efficiency. An adaptive reconstruction engine traverses the decomposition tree using learned filters and multi-scale residual learning to produce an enhanced image. The hierarchical approach enables superior low-light image enhancement by allocating computational resources based on content complexity, achieving better quality than fixed decomposition methods while maintaining compatibility with existing image signal processing pipelines.

The invention discloses a cross-modal target detection method based on learnable Fourier transform, and mainly solves the problem of insufficient fusion of a visible light image and an infrared image in a complex scene due to inter-domain difference in the prior art. According to the implementation scheme, the method comprises the steps that bimodal features are extracted through a double-flow CSPDarknet53 network; a target position guiding module is utilized to enhance target area representation and suppress background interference; the features are converted to a frequency domain, and amplitude texture information of the visible light image and phase contour information of the infrared image are adaptively enhanced through a learnable frequency domain feature enhancement module; suppressing noise through global filtering and then inversely transforming back to a spatial domain; and finally, outputting a target detection result of the multi-modal image by the detection head. According to the method, frequency domain physical characteristics are fully utilized, full complementation and adaptive fusion of cross-modal features are realized, the detection precision and robustness of vehicles, pedestrians and other targets under low-illumination and complex backgrounds are remarkably improved, meanwhile, high calculation efficiency is kept, and the method can be applied to the fields of automatic driving, intelligent monitoring and the like.

The invention provides an end-to-end tiny target detection method, and aims to solve the problems of missing detection and false detection of tiny targets caused by interference of sparse features, halo, noise and the like. According to the method, a TINYDETR model is constructed, and the TINYDETR model is composed of an HGNetv2 backbone network, an LGFSI module, an SO-CSFF module and a decoder with an auxiliary prediction head. Wherein the LGFSI module realizes global-local information interaction through joint modeling of a frequency domain and a spatial domain, and background interference is effectively suppressed; the SO-CSFF module enhances the fusion of shallow details and deep semantics through a bidirectional feature flow mechanism, and enhances the feature expression of a tiny target. After the model is trained and optimized, high-precision detection of a tiny target can be realized.

The invention relates to the technical field of computer vision and deep learning, in particular to an aerial image target detection method based on frequency domain decoupling multi-scale feature fusion, which comprises the following steps of: acquiring an aerial image of an unmanned aerial vehicle, establishing a data set, and performing preprocessing and data division; an aerial image target detection network is constructed, and the aerial image target detection network receives an input image and outputs a target category and a bounding box position; loss functions are determined, wherein the loss functions comprise classification loss representing matching quality, coordinate loss representing prediction coordinate relevancy and bounding box regression loss representing bounding box positioning accuracy; training the aerial image target detection network based on the data set and the loss function; inputting a to-be-detected aerial image into the trained aerial image target detection network to obtain a to-be-detected target category and a bounding box position; the method can improve the feature fusion degree, retains high-frequency details, and enhances the small target recognition rate.

The invention provides an automatic early warning method for sudden weather in a target area, which belongs to the technical field of weather early warning, and comprises the following steps of: establishing a primary dense matrix by adopting adaptive filtering processing and a frequency domain signal separation technology, and generating a secondary dense matrix by applying a marine meteorological recognition model of a spiral progressive network structure; a dynamic statistical equation is used to calculate the physical coupling relationship of each parameter to establish a multi-scale weather process balance matrix, a maximum flow and minimum cut algorithm is used to optimize a weather system coupling relationship network to calculate a coupling degree matrix, and a dynamic threshold adjustment mechanism is established according to coupling strength parameters to adjust the early warning detection frequency. And based on a comparison result of the coupling degree moment order maximum characteristic value and a preset risk threshold value, establishing a grading early warning system and outputting a corresponding early warning signal to control an offshore oil and gas platform emergency response system. The technical problem that a multi-time scale weather process coupling relationship cannot be effectively processed is solved.