1088 results about "Feedforward neural network" patented technology

Disclosed are a complex device fault diagnosis method and system based on multi-dimensional features, and the method comprises the following steps of: acquiring fault data of a target complex device, and carrying out input coding mapping and position coding; utilizing a multi-head attention mechanism and multi-head dilated convolution series connection, and acquiring fault features covering global and local deep information in combination with a feedforward neural network; fusing the extracted fault features to acquire a multi-dimensional feature vector; and acquiring a fault type probability score matrix by a classifier, and outputting a fault diagnosis result of the target complex device. The global modeling advantage of multi-head self-attention and the local multi-scale feature perception advantage of multi-head dilated convolution are fused, and a device fault diagnosis model in which a Transform model and the multi-head dilated convolution are mutually fused is established.

The invention provides a predictive maintenance management method and system for an industrial robot, and relates to the technical field of equipment maintenance management, and the method comprises the steps: obtaining multi-modal heterogeneous data of the industrial robot, carrying out the intelligent preprocessing, obtaining first preprocessing data, carrying out the cross-modal correlation fusion of the first preprocessing data, generating fusion data, and carrying out the prediction of the industrial robot; extracting complementary redundant information and constructing a semantic feature map; constructing a health state prediction model, adding the semantic feature map into the health state prediction model, mapping the semantic feature map to an embedding space through linear transformation, extracting time sequence information, carrying out feature interaction to obtain interaction features, and carrying out transformation and feature enhancement through a feedforward neural network layer to obtain health state probability distribution; comprehensive health state distribution is generated through Bayesian reasoning fusion, an optimal maintenance strategy is generated in combination with a detection model, an abnormal mode and a key turning point, collaborative maintenance is conducted on the industrial robot, a maintenance result is uploaded to a server, and the current state is stored.

A fuel cell fault diagnosis method based on a neural network. The method comprises: for a PEMFC system, acquiring operating state data, wherein the PEMFC system comprises a humidifier and a fuel cell stack; inputting the operating state data into a trained fault pre-diagnosis model, so as to predict a fault diagnosis result of the fuel cell stack at a subsequent moment. The fault pre-diagnosis model is constructed on the basis of a Transformer neural network, and comprises an encoder and a decoder, wherein the encoder comprises a plurality of small encoders, and each small encoder is provided with a self-attention layer and a feedforward neural network layer; and the decoder comprises a plurality of small decoders, and each small decoder comprises a self-attention layer, an attention layer and a feedforward neural network layer. An early warning can be provided for a water management fault of a PEMFC in an operating state, and the accuracy of pre-diagnosis is effectively improved.

The invention relates to the technical field of intelligent scheduling, and discloses a multimodal transport intelligent scheduling optimization method, device and equipment and a storage medium, and the method comprises the steps: carrying out the modeling of a plurality of transport nodes and transport paths, and obtaining a network topology structure; performing time period division on the scheduling period based on the network topology structure to obtain a time window scheduling framework; inputting the historical freight data into a multi-layer feedforward neural network for training to obtain a demand predictor; dynamically adjusting a time window of the time window scheduling framework to obtain a dynamically adjusted time window; inputting the state space and the action space into an improved double-delay depth deterministic strategy gradient algorithm for training to obtain a strategy generator; and performing multimodal transport scheduling decision generation based on the demand predictor, the dynamically adjusted time window and the strategy generator to obtain a multimodal transport scheduling scheme, so that efficient and flexible scheduling decision generation is realized, and the operation efficiency of multimodal transport is improved.

The invention discloses a denoising network model based on a diffusion model and an oracle rubbing image restoration method, which can improve the universality of the restoration method, improve the diversity of restoration results, reduce the calculation overhead and reduce the manual dependence. The denoising network model is adjusted based on an existing U-Net network and comprises a convolutional trunk network, an encoder, a middle layer, a decoder, a deep convolutional layer and a feedforward neural network which are connected in sequence; segmenting the input image by the convolutional trunk network; the encoder comprises a plurality of groups of cascaded paired visual state space blocks and image merging blocks; the decoder comprises a plurality of groups of cascaded paired visual state space blocks and image expansion blocks; performing down-sampling and up-sampling operations on the image merging block and the image expansion block respectively; the visual state space block adopts the architecture of an existing Mama module, and a two-dimensional selective scanning module is selected to replace an S6 module of the Mama module.

The embodiment of the invention provides a power system short-term load prediction method and device, equipment, a medium and a product. The method comprises the following steps: acquiring historical power load data and historical meteorological parameter data in a historical time period and meteorological parameter data of a to-be-predicted day; according to the historical meteorological parameter data and the meteorological parameter data of the to-be-predicted day, determining a time period corresponding to a target similar day of each time period in the to-be-predicted day, and forming a historical day sample set based on the historical power load data and the historical meteorological parameter data of the time period corresponding to the target similar day; training an improved multi-layer feedforward neural network model based on the historical day sample set to obtain a trained target load prediction model; and inputting the meteorological parameter data of the to-be-predicted day into the target load prediction model to obtain a power load prediction value of each time period in the to-be-predicted day, and accumulating the power load prediction values of all time periods in the to-be-predicted day to obtain a power load prediction value of the to-be-predicted day. By using the method, the power load prediction precision is improved.

The invention discloses an electric power system time sequence prediction model training method, system and device based on large language model weight fine tuning and a storage medium. The method comprises the steps of obtaining time sequence data of different fields and performing data cleaning to obtain a data set; performing weight freezing on a feedforward neural network and multi-head attention of the large language model, and performing first-stage training on the large language model by using the data set; reconstructing an input layer of the large language model to enable the reconstructed input layer to generate a corresponding natural language instruction according to the time sequence data; and freezing weight parameters of the large language model, and performing second-stage training on the large language model by using source load data of the power system to adjust parameters of the reconstructed input layer so as to obtain a time sequence prediction pre-training model. According to the method, the prediction effect can be improved while the weight of the large language model backbone network adapts to the specific task of time sequence prediction.

The invention discloses a process parameter optimization control method and system for an automobile injection molding part, and belongs to the technical field of data processing, and the method comprises the steps: collecting sample data to train a feedforward neural network based on the historical processing record of a target injection molding part, and constructing a product quality prediction plug-in; performing parameter fluctuation influence analysis according to the historical processing record, and constructing a fluctuation influence analyzer; determining product demand characteristics of the target injection molded part in combination with the automobile use scene; and by taking the product demand characteristics as expectations, optimizing the process parameters by using the product quality prediction plug-in and the fluctuation influence analyzer, and outputting the optimal process parameters for processing control. The problems of large product quality fluctuation and low production efficiency caused by insufficient injection molding process parameter control precision in the prior art are solved. By combining historical records, product quality prediction plug-ins and fluctuation analysis, precise optimization of process parameters is realized, the product quality is stabilized, and the controllability and efficiency of the production process are improved.

The invention relates to the technical field of electric power engineering purchase demand prediction, in particular to an electric power engineering purchase demand prediction system based on machine learning, and the system comprises the steps: obtaining historical purchase data, construction progress information and electric power engineering design parameters, carrying out the standard stage division and time alignment, and constructing a stage sequence model reflecting the material use rhythm; and a coupling factor matrix is generated based on the material co-occurrence frequency and the stage position relationship, and the modeling capability of the model for the material cooperation relationship is enhanced. And the stage time sequence features, the coupling information and the structured engineering parameter vectors are fused and input into a regression prediction model, so that accurate mapping of material demands and multi-dimensional engineering features is realized, and the purchase prediction precision in a target period is improved. A deviation sequence is constructed based on historical prediction errors, and error correction is performed through a feedforward neural network, so that prediction accuracy and response capability are effectively improved, and resource waste and construction delay are reduced.

The invention belongs to the technical field of hyperspectral image unsupervised, and discloses a hyperspectral image self-supervised clustering method based on a local-global double-branch network, which comprises the following steps: (1) constructing a 3D pixel block; (2) acquiring shallow depth features through a ResNet module; (3) injecting shallow layer features into a dual-path network module; (4) calculating the similarity between the local features and the clustering center to obtain local semantic probability distribution, and then obtaining target distribution; (5) the global features are sent to a feedforward neural network to obtain the probability of each pixel, and global semantic probability distribution is obtained; and (6) constructing a network loss function through a dual self-supervision mechanism to guide the updating of the whole network model. According to the method, the local detail features and the global semantic features of the hyperspectral image are mined by introducing the dual-path network architecture, so that the recognition precision of the hyperspectral ground features is effectively improved.

Predicting the salience of one or more data entities to a particular (target) data entity from among a plurality of data entities may comprise generating a graph of the plurality of data entities and a machine-learned model architecture that predicts the salience of the one or more data entities output by the machine-learned model architecture using the graph. For example, the machine-learned model architecture may comprise a first machine-learned model for generating an embedding using the content of the target data entity, a second machine-learned model for generating a vector using the data type indicated by the target data entity, and a third machine-learned model (e.g., a graph neural network or other feed-forward neural network) for generating a contextual representation of the target data entity to which other contextual representations associated with the plurality of data entities may be compared (e.g., using Euclidean distance, cosine similarity, dot product).

The invention discloses a calculation and measurement fused high-fidelity digital twinning dynamic monitoring method for a cantilever beam structure. Firstly, a real-time updated database is constructed based on a physical entity system; secondly, constructing a three-dimensional digital model by using SolidWorks, and performing multi-physics field simulation analysis in combination with Ansys to obtain high-precision mechanism data of a structural key area; thirdly, performing reduced-order calculation on a simulation result by using MWorks, training and optimizing a reduced-order model through a feedforward neural network, and reducing an error between simulation data and reduced-order data; fusing the one-dimensional system data order reduction model and the three-dimensional field data order reduction model based on a Modelica language; and finally, integrating a mechanism-data hybrid model, constructing a dynamic response characteristic and spatial distribution characteristic integrated high-fidelity digital twinborn body, comparing twinborn body calculation data with sensor actual measurement data in real time, and accurately representing geometric morphology, mechanical properties and dynamic behaviors of a physical system. The method is suitable for high-precision monitoring, real-time state evaluation and performance optimization of the cantilever beam structure.

The invention discloses a large language model accelerator architecture based on a three-dimensional NAND flash memory, and belongs to the technical field of calculation, reckoning or counting. The architecture comprises a three-dimensional NAND flash memory used for executing feedforward neural network calculation; the auxiliary calculation unit is used for executing attention mechanism calculation; the DRAM chip is used for storing attention mechanism related weights and KV cache; and the interconnection resource is used for realizing data interaction among the components. Wherein the three-dimensional NAND flash memory comprises a logic chip and an NAND array chip, the logic chip is used for controlling and executing calculation, and the NAND array chip is used for storing weights and participating in calculation. The invention further provides a scheduling method based on KV cache awareness. The scheduling method comprises decomposition and dynamic allocation of calculation tasks. Through collaborative design of the hardware architecture and the scheduling method, the memory wall bottleneck in large language model calculation can be relieved, the energy consumption is reduced, and the overall calculation performance is improved.

The invention discloses a multi-modal large language model training-free acceleration method based on visual token efficient processing, and the method comprises the following steps: obtaining multi-modal question and answer data, and dividing the multi-modal question and answer data into a verification set and a test set; inputting the verification set into the target multi-modal large language model, and sorting the redundancy of each layer of the model through a search method; according to the redundancy ranking, inputting the test set into the target multi-modal large language model, and obtaining the redundancy of each layer of the target multi-modal large language model; and according to the redundancy, carrying out an acceleration operation about visual token processing on a part of layers of the target multi-modal large language model. According to the method, the characteristic that the computing power required by the visual token is less than that required by a text token is utilized, the redundancy layer in the large language model is positioned, and the self-attention operation and the feedforward neural network operation about the visual token are optimized, so that the model reasoning efficiency is greatly improved.

The invention relates to a hybrid network 4D flight path prediction method based on LSTM and Transform, and belongs to the technical field of flight path prediction. Comprising the following steps: inputting preprocessed flight path data into an embedding layer, mapping the preprocessed flight path data into high-dimensional feature representation, and forming an input sequence; inputting the input sequence into the LSTM network to generate an output representation with a time-dependent feature; performing global feature extraction on the output representation of the LSTM network by using a multi-head attention mechanism so as to capture a dependency relationship and relevance between different trajectory data points; performing normalization processing on global features output by the multi-head attention mechanism, and further enhancing feature expression ability through a feedforward neural network; a mask multi-head attention mechanism is adopted in a decoder, and prediction features are output; and mapping the prediction features output by the decoder to a target space through a linear layer, and generating a final result of flight path prediction. The method is excellent in performance in complex flight path prediction tasks.

The invention relates to a high-density memory chip packaging structure design method, device and equipment. The method comprises the steps of performing three-dimensional modeling on a multi-chip stacking structure of a high-density storage chip based on a current packaging structure parameter combination to obtain a packaging structure model; performing finite element analysis to obtain stress distribution data and a stress-strain curve; performing feature extraction and feature fusion to obtain a fused feature vector; training based on a multilayer feedforward neural network to obtain a structure optimization model; performing dependency analysis, constructing a parameter tree structure, and distributing corresponding priority values; and sorting and optimizing the plurality of packaging structure parameters one by one by utilizing a parameter tree structure and the priority values, and processing constraint conditions by adopting a self-adaptive penalty function method to obtain a target packaging structure parameter combination. According to the implementation of the invention, the design efficiency and performance of the high-density memory chip packaging structure are improved.

The invention discloses a remote sensing image segmentation method fusing frequency modulation and spatial perception, and the method comprises the steps: obtaining and preprocessing an original remote sensing image, and generating a standardized input image; the image is input into a multi-scale frequency domain enhanced feature extraction network, features are extracted step by step according to a plurality of feature levels, each level realizes frequency adaptive semantic enhancement through frequency domain modulation transformation and spatial feature fusion, and deep feature expression is enhanced through feedforward neural network modeling and residual connection output and cross-level residual fusion introduction; the final multi-scale features are decoded through a decoding module, the spatial resolution is recovered, and a pixel-level segmentation result is generated; and constructing a composite loss function containing classification errors, boundary perception and frequency consistency items, and carrying out optimization training on the network. According to the method, semantic complementarity of a remote sensing image in a frequency domain and a space domain is fully mined, so that segmentation precision and robustness of a ground object target in a complex scene are improved, and the method has good generalization ability and engineering practicability.

The invention provides a prediction method and device for operation parameters of a virtual power plant, a storage medium and a computer program product. Comprising the steps of performing data preprocessing on acquired influence factor data; using a time sequence decomposition algorithm to decompose the influence factor data into three-dimensional sequences including a trend component, a season component and a residual component, and segmenting each sequence into sequence blocks and mapping the sequence blocks into time sequence feature vectors; performing K-means clustering on word embedding used for pre-training the large language model, and selecting K clustering centers as semantic anchor points to be spliced with the time sequence feature vectors; finely adjusting the position embedding parameters of the large language model, the weight of the feedforward neural network in residual connection and the parameters of the normalization layer; and according to the semantic enhancement time sequence feature vector, predicting to obtain a trend component, a season component and a residual component of each dimension, and carrying out splicing and reverse normalization processing on the components to obtain a prediction result. According to the prediction method, the accuracy and real-time performance of virtual power plant load and electricity price prediction are improved.

The invention discloses an H-bridge key equipment service life and system reliability evaluation method and system for a cascade network construction type energy storage system, and belongs to the technical field of power system automation. The method comprises the following steps: firstly, extracting task profile parameters under multiple time scales, and constructing a time sequence feature model; secondly, estimating a hot spot temperature sequence of the IGBT device and the capacitor based on a multilayer feedforward neural network; then, in combination with a continuous extreme point paired temperature cycle extraction method and a Miner linear cumulative damage criterion, the damage factor and the residual life of the device are evaluated; then, task profile samples are expanded based on a generative adversarial network with gradient penalty, and life distribution and reliability indexes of key devices under different profiles are calculated; and finally, based on H-bridge series structure mapping device level information, constructing a system level reliability model, obtaining system failure rate, average fault-free operation time and a reliability function, and realizing health state perception and reliability quantitative evaluation of the energy storage system.

The invention relates to the technical field of traffic safety, in particular to a highway traffic safety intelligent early warning system and method based on artificial intelligence. The system is composed of a data acquisition module, a data preprocessing module, an intelligent analysis module, an early warning decision module and an interactive early warning interface, and the data acquisition module collects multi-parameter data through three ways of a sensor, a monitoring system and a meteorological station; the data preprocessing module cleans, standardizes, normalizes and supplements missing values for the collected multi-parameter data, trains the extracted features through the intelligent analysis module by using a feedforward neural network FFNN + self-attention mechanism, constructs a multi-dimensional risk prediction model, automatically identifies the change of a traffic mode, predicts potential dangerous events, and improves the risk prediction efficiency. The interactive response interface displays the early warning information and the multi-parameter data collected in real time. According to the invention, road conditions and traffic risk conditions can be more comprehensively known, powerful data support is provided for decision making of an early warning system, and the early warning accuracy and intelligent level are significantly improved.

The invention belongs to the technical field of biological feature recognition, and relates to a Transform-based gene data mining method, which comprises the following steps of: loading gene expression data from a file, and preprocessing; inputting the data into a designed model for training and evaluation; the designed model comprises a CNN (Convolutional Neural Network) and an improved Transform model; the improved Transform model comprises an encoder and a classifier; the encoder adopts a multi-head self-attention mechanism, a convolutional self-attention mechanism and a feedforward neural network; processing the data by using the trained model, and storing the fusion attention weight obtained by the coding layer in the data processing process; and finally, averaging the fusion attention weights of all batches, calculating the comprehensive attention weight of each gene at different time points, and identifying the most important gene by calculating and analyzing the attention score of each gene. According to the method, the problems of scarcity and imbalance of gene data can be solved, the risk of overfitting is reduced, and the precision of gene data analysis is improved.

The embodiment of the invention discloses an inference method and device for a large language model, which are used for reducing bandwidth consumption of inference of the large language model. The method comprises the steps that a parallel reasoning system receives reasoning tasks of a large language model, wherein the reasoning tasks comprise a calculation-intensive task and a self-attention calculation task; the processor executes calculation-intensive tasks, generates intermediate data and sends the intermediate data to the near-memory calculation module, and the calculation-intensive tasks comprise one or more of the following tasks of the large language model: a feedforward neural network calculation task, a projection task and a layer normalization task. And the near memory calculation module executes a self-attention calculation task according to the intermediate data to generate a near memory calculation result. And the processor generates a reasoning result corresponding to the reasoning task based on the proximity calculation result.

The invention relates to a machine tool spindle thermal deformation prediction method based on a GCN-Transform model. The method comprises the following steps: constructing a sensor network for a machine tool; a space-time diagram data set is generated in the three-dimensional model of the machine tool, and the spatial characteristics of the sensor network are represented through an adjacent matrix; constructing a GCN-Transform model, wherein the model comprises an input layer, a one-dimensional convolution component, a multi-head self-attention mechanism, a feedforward neural network layer, an addition and normalization layer and an output layer; performing prediction, extracting spatial features through GCN, extracting time features through Transform, an adjacent matrix and sensor data as input, and performing thermal error; according to the method, the time characteristics are captured by using the Transform model, and then the spatial characteristics in the topological space of the sensor are captured by using the GCN, so that the time characteristics of the thermal error can be accurately captured, and the spatial information of the sensor can be fully utilized, so that the prediction result is more accurate and robust.

The invention discloses an antibacterial peptide recognition method and system based on a sequence-structure dual-channel neural network, and the method comprises the steps: splicing amino acid features and amino acid-level manual features extracted by ProtT5 to obtain peptide embedding, and transmitting the peptide embedding to a sequence channel composed of a plurality of Transform blocks to extract the sequence features of the peptide; predicting a three-dimensional structure of the peptide by using ESM-Fold to construct an adjacency graph, taking amino acid features obtained by ESM-2 as node features of the adjacency graph, and performing layer-by-layer extraction and enhancement by fusing structural channels of multi-head graph attention, a residual network, layer normalization and a feedforward neural network; and carrying out maximum pooling and splicing on the sequence features and the structural features, and then, carrying out antibacterial peptide prediction. According to the method, a multi-feature fusion strategy is adopted, meanwhile, the three-dimensional structure information of the antibacterial peptide is introduced, and the sequence and the structural features are fused through a two-channel architecture, so that the recognition accuracy of the antibacterial peptide is effectively improved.

Aspects of the disclosure include methods for leveraging retrieval augmented generation (RAG) over a graph neural network (GNN) for edge building and the generation of reason-aware graph recommendations. A method can include constructing a graph neural network from an input graph having a plurality of nodes and one or more edges. The graph neural network includes one or more internal layers, each internal layer having one or more node vectors encoding a K-hop neighborhood for a target node of the plurality of nodes. RAG data including non-graph contextual data is retrieved for each of the plurality of nodes and transformed into embeddings using a large language model encoder. The RAG embeddings are encoded into node vectors of the graph neural network. The graph neural network generates a representation for the target node that is transformed by a feed forward neural network tower into an output vector.

The invention discloses a method for correcting a laser gas concentration sensor, which is used for measuring gas concentration by using laser with different wavelengths and correcting by combining environmental parameters. The method comprises the following steps: extracting laser absorption characteristics by using a first feedforward neural network, extracting environmental matrix characteristics by using a convolutional neural network, fusing the characteristics through an attention mechanism, and predicting the gas concentration through a second feedforward neural network. The loss function not only considers the difference between the predicted value and the true value, but also comprises the physical constraint between the light intensity value and the theoretical light intensity value. According to the method, the measurement accuracy and the model robustness under different environmental conditions are improved, and the influence of measurement errors and environmental fluctuation is reduced through the combination of multi-wavelength measurement and physical knowledge.

The invention provides an in-orbit integrated circuit aging fault diagnosis method and diagnosis system, and relates to the technical field of integrated circuit design, and the method comprises the steps: ground training: collecting characteristic parameters, carrying out the characteristic parameter processing based on an SPI protocol algorithm, a digital signal recovery fixed-point decimal algorithm, a wavelet transform filtering algorithm and a deviation standardization algorithm, and carrying out the fault diagnosis; constructing an aging failure fault data set, training through a back propagation algorithm to obtain feed-forward neural network parameters, and transplanting the feed-forward neural network parameters to a feed-forward neural network constructed based on Pytorch and a feed-forward neural network constructed based on FPGA for joint testing; and in-orbit testing: data enters a data buffer area, training data is provided for a learning training module, feedforward neural network parameters are periodically updated by using an exponential weighted moving average algorithm, and the feedforward neural network parameters are transplanted to a feedforward neural network established based on Pytorch and a feedforward neural network established by an FPGA for joint testing. Continuous learning of on-orbit data is realized, and the method adapts to irregular periodic electromagnetic influence of a space environment.

The invention discloses a fumigation treatment multi-parameter comprehensive monitoring system which comprises a multi-parameter sensor monitoring module, a multi-modal data processing and analyzing module, a system intelligent monitoring and adjusting module, a fumigation environment control module, a remote maintenance and management module and a system interaction and feedback module. The multi-parameter sensor monitoring module is used for collecting parameter data. The multi-modal data processing and analyzing module is used for fusing and analyzing data. The system intelligent monitoring and adjusting module is used for intelligently monitoring and adjusting fumigation strategies. The fumigation environment control module is used for adjusting environment parameters. The remote maintenance and management module is used for system maintenance and management. And the system interaction and feedback module is used for display interaction of data and a control state. According to the fumigation treatment multi-parameter comprehensive monitoring system, a multi-modal data fusion algorithm based on information channel decomposition is provided for multi-parameter data fusion and analysis; and a multi-parameter intelligent monitoring algorithm based on a multi-layer feedforward neural network is provided to intelligently monitor the fumigation environment.

The invention discloses an efficient Transform model reasoning accelerator deployment method based on Versaal ACAP, and belongs to the field of software and hardware collaborative acceleration numerical calculation. Firstly, load analysis is carried out on the model, and meanwhile, according to hardware performance, the types and scales of operators needing to be borne by hardware of all parts are determined; then, hardware is equipped with a data transmitter, a data receiver, a nonlinear operator and other modules; then, the modules are reasonably deployed and connected in hardware, pipeline parallelization is completed, and calculation of a multi-head attention stage and a feedforward neural network stage is achieved; and finally, integrating the two stages to realize accelerated calculation of the whole encoder / decoder layer so as to accelerate the reasoning speed of the upper layer model. Experiments prove that the throughout of the model is effectively improved, the reasoning delay is reduced, meanwhile, the accelerator has huge advantages in the aspect of batch reasoning, and the accelerator has wide application prospects in the aspects of reducing the model reasoning cost and improving the model reasoning speed.

The invention belongs to the technical field of target detection, and discloses an unmanned aerial vehicle image target detection method based on an improved RTDETR model, and the method introduces an FAPPA module in a backbone network, and effectively alleviates the information loss problem of small target features in a deep network. According to the method, a feedforward neural network of an AIFI module in an encoder is replaced by an EDFFN module, and high-frequency features such as edges and textures are selectively enhanced by using a patchwise FFT strategy and learnable frequency domain weights. Besides, according to the method disclosed by the invention, adaptive alignment of multi-scale features is realized by utilizing a Zoomcat module, and local details and global semantics are fused by utilizing a convolution attention double-path architecture of a CAFMFusion module, so that complementary information of different levels is fully utilized. According to the method, the unmanned aerial vehicle image target detection capability can be remarkably improved, and small target features in a complex background can be identified.