247 results about "Data diversity" patented technology

Multi-modal models learn a joint latent space for relating data points across different modalities. To more effectively learn multi-modal models with reduced training requirements and greater benefit from limited multi-modal training data, a multi-modal model may be trained with fixed or pre-trained unimodal encoders that generate data representations in respective latent spaces. The multi-modal model is trained to learn a shared latent space while fixing the unimodal encoders, enabling training without storing the unimodal encoders in memory. Limited multi-modal data may also be augmented by generating synthetic data between commonly-labeled pairs in the respective modality's latent spaces. The effect of data diversity can also be determined by generating a diverse data set with respect to the data points in latent space, enabling measurement of performance of the multi-modal model on limited training data.

Multi-modal models learn a joint latent space for relating data points across different modalities. To more effectively learn multi-modal models with reduced training requirements and greater benefit from limited multi-modal training data, a multi-modal model may be trained with fixed or pre-trained unimodal encoders that generate data representations in respective latent spaces. The multi-modal model is trained to learn a shared latent space while fixing the unimodal encoders, enabling training without storing the unimodal encoders in memory. Limited multi-modal data may also be augmented by generating synthetic data between commonly-labeled pairs in the respective modality's latent spaces. The effect of data diversity can also be determined by generating a diverse data set with respect to the data points in latent space, enabling measurement of performance of the multi-modal model on limited training data.

Multi-modal models learn a joint latent space for relating data points across different modalities. To more effectively learn multi-modal models with reduced training requirements and greater benefit from limited multi-modal training data, a multi-modal model may be trained with fixed or pre-trained unimodal encoders that generate data representations in respective latent spaces. The multi-modal model is trained to learn a shared latent space while fixing the unimodal encoders, enabling training without storing the unimodal encoders in memory. Limited multi-modal data may also be augmented by generating synthetic data between commonly-labeled pairs in the respective modality's latent spaces. The effect of data diversity can also be determined by generating a diverse data set with respect to the data points in latent space, enabling measurement of performance of the multi-modal model on limited training data.

The invention discloses an intelligent question answering system construction method and system based on an LLM large language model, and the method comprises the steps: carrying out the corpus aggregation from a plurality of heterogeneous data sources, and dynamically capturing the text data related to a target domain; preliminarily screening the text data, and based on semantic similarity and a text quality scoring rule, removing redundant and low-quality text data or text data irrelevant to a target field; performing diversity analysis on the text data, and generating diversified text data with high similarity with real data through an adversarial generation technology; performing multi-level semantic annotation on the text data, and generating a training sample by analyzing context information in the annotated corpus; and converting the training sample into a standard format sample, and training an LLM large language model by using the standard format sample to serve as an intelligent question-answering system of a target domain. By utilizing the embodiment of the invention, the intelligent level of the intelligent question-answering system in various application scenes can be effectively improved, and further development of the intelligent question-answering technology is promoted.

The invention relates to the field of Raman spectrums, and discloses a Raman spectrum nonlinear generation method driven by mixed machine learning, which is used for meeting the requirements of a modern Raman spectrum analysis technology on mixture spectrum data with wide coverage, complete types and reliable quality. Comprising the following steps: performing weighted linear combination on pure component spectrums based on a concentration ratio to generate a primary mixture spectrum; the data diversity is enhanced; generating a prediction spectrum; generator parameters are optimized through multi-level feature comparison and adversarial training of the discriminator; matching and deleting false peaks, and reconstructing physical reasonable characteristic peaks; and dynamically adjusting the intensity of the characteristic peak, and outputting high-fidelity spectrum data. According to the method, by fusing physical constraint and deep learning, the peak position precision, intensity consistency and noise robustness of the generated spectrum are remarkably improved, quantitative analysis of a complex mixture in a wide concentration range is supported, and an efficient solution is provided for Raman spectrum database construction and chemical detection.

Aiming at the problems of insufficient data diversity, data sparsity and model generalization ability in an industrial time sequence prediction model, the invention provides a method for establishing a time sequence prediction model oriented to industrial multi-modal data, which comprises the following steps: acquiring multi-source heterogeneous industrial time sequence data, and cleaning the acquired data; performing data enhancement on the cleaned data to generate an enhanced time sequence data set; fusing the multi-source heterogeneous time series data in the obtained enhanced time series data set by using a heterogeneous data fusion sensing method of multi-dimensional space-time mapping; using the fused data to construct a prediction model oriented to the multi-modal industrial time series data based on an AI large model, and performing model training; and according to the trained prediction model oriented to the multi-modal industrial time series data, predicting the change trend of the key indexes in the industrial process in a period of time in the future.

The invention belongs to the technical field of power grid big data, and particularly relates to a power grid regulation and control data asset value evaluation method and device, equipment and a medium. According to the method, key factors (data integrity, data diversity, data timeliness, data scale and data relevance) influencing the value of the power grid regulation and control data assets are determined, the weight coefficient of each factor is calculated by using an analytic hierarchy process, and then the value of the power grid regulation and control data assets is calculated. According to the scheme, a unified data value evaluation system is established, and a unified value measurement standard is provided for power grid regulation and control data assets by determining the key influence factors and the weight coefficients. Through data value evaluation, the value and distribution condition of each data asset can be clearly known, and efficient management and reasonable utilization of the data assets can be realized. Meanwhile, through data value evaluation, the potential value of the data assets can be found, and possibility is provided for value increment and achievement of the data.

The invention discloses an information transmission attack event intelligent prediction model training optimization method, which comprises the steps of a data preparation stage, a feature engineering stage, a model selection and construction stage, a model training stage and a model evaluation and optimization stage, and is characterized in that data are collected from multiple channels, noise is cleaned and removed, missing values and abnormal values are processed, and accurate labeling is carried out; then key features are extracted, redundant irrelevant features are removed through selection, standardized transformation is carried out, a proper model is selected based on data features, an integrated learning combination model is adopted, then a training set, a verification set and a test set are divided, a proper optimizer is selected, parameters are adjusted to be over, training is carried out through the training set, and over-fitting is prevented through the verification set; and finally, using index evaluation of accuracy and recall rate, and adjusting hyper-parameters and continuous optimization of the model structure through cross validation. According to the method, data collection, data imbalance processing and data noise interference processing are adopted, data diversity is improved, and interference of data noise is reduced.

The present invention relates to the field of image processing technology, and discloses a method for automatic segmentation and recognition of ultrasound images based on deep learning, including the following steps: S1, data preparation and preprocessing, S2, self-supervised learning pre-training, S3, transfer learning and incremental learning, S4, multi-task joint learning, S5, image quality enhancement and super-resolution reconstruction: S6, multi-modal information fusion, information fusion by combining modal images, S7, model training and optimization, increasing data diversity by using data enhancement technology. By fusing ultrasound images with modal image information, the segmentation accuracy and lesion recognition ability of ultrasound images are improved, and different modal images provide complementary features, showing significant advantages in the recognition of complex lesions, complex locations, complex morphology and early lesions. At the same time, by fusing images of different modalities through a deep learning model, image quality can be improved and diagnostic accuracy can be improved.

The invention relates to the technical field of data management, in particular to an artificial intelligence-based automatic data grading management method, which comprises the following steps of: acquiring to-be-graded data through a multi-source data acquisition channel, and performing data preprocessing on the to-be-graded data to generate a training data set; constructing a feature extraction network based on an improved Transform architecture, extracting data features in the training data set in combination with a cross-modal mechanism, and generating a dynamic feature mapping matrix; and constructing a time-space domain fusion model based on a graph neural network by using the dynamic feature mapping matrix, calculating a data sensitivity index, and performing adaptive classification. According to the method, complex cross-modal data can be processed, time-space domain features are fused, and efficient hierarchical management is realized through an improved deep learning model, so that the problems of data diversity, real-time performance and accuracy of an existing data hierarchical method are effectively solved.

The invention discloses an ice surface defect detection and quantification system and method based on deep learning. The system comprises a data acquisition module, a data enhancement module, a neural network training module, a binocular stereoscopic vision-based detection module and an information fusion and analysis module. The method comprises the following steps: firstly, acquiring image data of ice surface defects by using various camera devices through a data acquisition module, and covering different illumination conditions and defect types; and then, the data enhancement module performs rotation, overturning, contrast adjustment and noise addition on the acquired data so as to improve data diversity and generalization ability of the neural network. And then, a multi-scale feature extraction module (MSTG-Net) and a detail enhancement convolution module (MSDSC) are introduced into a neural network training module, and the segmentation feature extraction of the ice surface defect is realized by training the improved YOLOv8 neural network. The detection module based on binocular stereo vision acquires an ice surface image through a binocular camera, and obtains defect mark information and a depth map in combination with the trained neural network model. And finally, further determining the three-dimensional position, the surface area size and the spatial morphological characteristics of the defect by combining the information fusion and analysis module with a binocular stereoscopic vision distance measurement result. According to the method, high-precision identification and positioning of ice surface defects are realized, reliable data support is provided for subsequent ice surface repair planning, and the method has the advantages of high precision, strong robustness, wide applicability and the like.

The invention discloses a wheat organ three-dimensional point cloud segmentation method based on dynamic voxel and category perception. Three-dimensional reconstruction is carried out on a wheat sample based on a three-dimensional Gaussian splash method, and dense point cloud data are obtained; performing data amplification on the dense point cloud data by adopting a dynamic voxel rasterization random sampling method combining dynamic voxel division and a random farthest point sampling strategy, and performing standardized point cloud quantity processing to obtain a standard point cloud data set; and based on the standard point cloud data set, carrying out key organ segmentation by adopting a category perception segmentation network. According to the data enhancement method, key geometric features are maintained while data diversity is expanded, a feature extraction module of the category perception segmentation network solves the problem of ignoring important points during sampling by fusing DCA and a weighted farthest point sampling strategy, and an improved loss function combining a point cloud space structure and information is adopted to improve the accuracy of data enhancement. And the segmentation precision of the wheat complex point cloud data is effectively improved.

The present invention proposes a wind power equipment quality assessment method based on multi-sensor data fusion, comprising: acquiring real-time data of multiple sensors on the wind power equipment, and adopting a five-point cubic smoothing algorithm to denoise the real-time data to obtain standard sample data; performing time domain, frequency domain and entropy domain feature extraction on multiple sensor data at the same time in the standard sample data in time sequence to obtain a multi-sensor high-dimensional feature data set; performing feature fusion on feature vectors in the multi-sensor high-dimensional feature data set at the same time to obtain an initial multi-modal fusion feature map; constructing a multi-scale deep feature fusion network model based on a ResNet network, inputting the initial multi-modal fusion feature map, and outputting a multi-scale fusion feature vector; through multi-sensor data fusion and deep feature extraction, the data diversity and complexity problems in the quality assessment of wind power equipment are solved, and the accuracy and robustness of the assessment are improved.

The invention discloses a wolfberry screening method and system based on image analysis, and particularly relates to the technical field of wolfberry screening. Lycium barbarum samples with different growth environments, maturity, varieties and drying degrees are selected, images are shot under various illumination conditions, and data diversity is ensured; the image is preprocessed, surface texture roughness features and color distribution uniformity features are extracted, and support is provided for accurate classification; constructing a data prediction model based on the extracted features, evaluating the accuracy of the algorithm for identifying the appearance difference of the Chinese wolfberry fruits, and if the algorithm is high in accuracy, directly applying to a production line to realize automatic sorting of qualified and unqualified Chinese wolfberry fruits; and if the accuracy is low, the problem of misjudgment caused by natural differences of the Chinese wolfberry fruits is effectively solved by predicting the abnormal degree of the accuracy, dynamically adjusting the identification strategy and improving the screening precision, the accuracy and the automation level of Chinese wolfberry fruit screening are improved, the waste of high-quality Chinese wolfberry fruits is reduced, and the enterprise income and the brand reputation are remarkably improved.

The invention provides a ship noise multi-feature classifier data enhancement method and system based on a multi-fine-grained conditional diffusion model. And compressing a waveform to a potential space through VQ-VAE, extracting a ship type / ship name cross semantic vector by using ResNet, and optimizing clustering in combination with a loss function. And a one-dimensional U-Net conditional diffusion model is constructed, unconditional / conditional model output is dynamically weighted and fused, and the weight is adaptively adjusted according to training loss. In the generation stage, a semantic prototype is constructed by using a high-fine-granularity label, parameters are determined by using low / medium-granularity mean value sampling and Bayesian optimization, and fine-granularity controllable waveform generation is realized. After the generated data is converted into multiple features such as MFCC and Lofar, the generated data and original data are combined to train a classifier, and a virtual class strategy relieves class imbalance. Experiments show that the MSE of generated data and real data is reduced, the classification accuracy is improved, the data diversity and the model generalization ability are remarkably enhanced, and the method is suitable for scenes such as underwater target recognition.

The invention belongs to the technical field of aero-engine health management, and discloses an aero-engine remaining service life prediction method based on a diffusion model and a space-time attention mechanism. The invention provides a novel learning framework which is used for RUL prediction of an aero-engine. The framework fuses a diffusion process and an attention-based time module to achieve end-to-end accurate estimation from sensor signals to RUL. Aiming at the problem that the scarcity problem of aero-engine data limits the application of a deep learning network in the aspect of improving the RUL prediction precision, the method adopts a diffusion process to enrich the data diversity and improve the data quality. In order to improve the accuracy of RUL prediction, an attention mechanism is used to extract time features in samples and between the samples, and time information is effectively utilized, so that the capture capability of the aero-engine performance degradation process is enhanced. The RUL prediction method provided by the invention provides more reliable support for health management and maintenance decision of the aero-engine.

The invention provides a bearing small sample data expansion method and system based on a variational auto-encoder, and belongs to the field of deep learning and data enhancement. The problems that a traditional generation model has limitation in bearing small sample data, feature fuzziness and distortion are prone to occurring, and the data set quality is poor are solved. According to the method, a deep VAE framework is constructed, and a dimension reduction module, a data expansion module and a dimension raising module are used in a potential space; dimensionality reduction is performed on high-dimensional data by adopting a UMAP algorithm, so that the topological structure of the data is effectively reserved, and the extraction efficiency and quality of data features are improved; a Gaussian mixture model combining regularization and particle swarm optimization optimization is used for fitting distribution of scattered small sample data, new data with fusion features are expanded through sampling, and data diversity is increased; a radial basis function is used for nonlinear data dimension raising, new data can be ensured to be accurately mapped back to a high-dimensional space, meanwhile, the relation between features is reserved, and defect data with fusion features is reconstructed through a decoder.

The invention relates to the technical field of advertisement marketing analysis, in particular to a data diversity identification advertisement marketing system, which comprises a data diversity identification module, a double-loop control module and a cross-layer information sharing channel module, and is characterized in that the data diversity identification module is used for calculating dynamic diversity indexes of multi-source heterogeneous data in real time; the double-circulation control module comprises an outer circulation strategy optimization layer unit and an inner circulation real-time control layer unit, the outer circulation strategy optimization layer unit operates at an hour-level time scale, a reinforcement learning algorithm is adopted to carry out long-term advertisement marketing strategy optimization, and an optimization objective function of the outer circulation strategy optimization layer unit is fused with a dynamic diversity index; the internal circulation real-time control layer unit operates at a second-level time scale, and adopts a parameter self-adaptive control algorithm to adjust an advertisement putting strategy in real time; and the cross-layer information sharing channel module is used for realizing strategy parameter transmission between the outer circulation and the inner circulation and applying strategy consistency constraint conditions so as to collaboratively optimize an advertisement marketing strategy.

The invention discloses a big language model knowledge extraction method and device based on difficulty perception, and aims to solve the problems that an existing knowledge extraction method is high in training cost and lack of pertinence in sample selection. According to the method, the difficulty of a distillation difficulty score evaluation sample is introduced, a difficult sample with learning value for a student model is identified, a distillation data set is dynamically adjusted in combination with a hierarchical data updating strategy, the difficult sample is preferentially reserved, a simple sample is removed, and meanwhile, data diversity is kept. Besides, a bidirectional difference loss function is provided, KL divergence and inverse KL divergence are combined, the optimization process is stabilized, more attention is paid to difficult samples, and gradient explosion or disappearance is avoided. Experimental results show that the performance of a student model is effectively improved, the method even exceeds a teacher model under some conditions, meanwhile, the training cost is remarkably reduced, and the method is suitable for task-independent instruction following and specific tasks and has wide application prospects.

The invention relates to the technical field of image processing, in particular to a multi-modal image processing method and system based on deep learning, and the system comprises a self-adaptive perception and generation module which is used for exploring the new characteristics of image data, enhancing the perception capability through a generation and simulation technology, and discovering that details are difficult to capture in an existing processing mode. According to the invention, through reverse modal generation, cross-modal characteristic unification, dynamic modal reconstruction and data simulation experiments, data diversity expansion and accurate characteristic analysis are realized; through causal reasoning, knowledge migration and adversarial learning, inter-modal characteristic consistency and analysis performance are optimized; according to the multi-modal image analysis method and system, the multi-modal image analysis efficiency and accuracy are remarkably improved by combining the task-driven model switching and federated learning technology and supporting distributed model training of dynamic adaptation and privacy protection, and meanwhile, the application range of medical image analysis in a resource-constrained environment is widened.

The invention discloses a power system multi-physical-quantity data alignment method, and relates to the technical field of power system data alignment, and the method comprises the following steps: collecting multi-physical-quantity data in real time through a plurality of sensors and devices in a power system, and calibrating the original data according to the types and installation positions of the sensors, and all data sources are ensured to provide accurate time sequence information. By calibrating and synchronizing sensor data, time sequence consistency of multi-physical-quantity data is ensured, and errors caused by equipment difference and transmission delay are eliminated. Abnormal values are automatically detected and corrected by adopting a deep learning algorithm, the data quality is improved, and manual intervention is reduced. Meanwhile, the generative adversarial network is used for synthesizing and augmenting data, the data diversity and coverage are enhanced, the generalization ability and prediction precision of the large electric power model are improved, especially in a complex scene, the overfitting problem is avoided, and the accuracy and high efficiency of the model in practical application are ensured.

The invention discloses a face privacy protection-oriented generated image detection method and a face privacy protection-oriented generated image detection system. The method comprises the following steps of: firstly, preparing face and text pairing data, and finely adjusting a diffusion model; secondly, on the basis of the diffusion model after fine tuning, potential vectors are extracted and clustered, and text prompts and center vectors obtained through clustering form a dictionary; and finally, based on the obtained dictionary, obtaining a pseudo image and a label through the fine-tuned diffusion model, and outputting a detection result through a classifier. According to the method, potential spatial clustering and conditional diffusion generation are combined, privacy protection and data diversity are taken into consideration, and the security and generalization ability of forged face image detection are remarkably improved.

The invention relates to the technical field of image recognition, in particular to an animal and plant image recognition method based on deep learning and electronic equipment, and the method comprises the steps: obtaining and preprocessing sample image data, and constructing a sample image database; and constructing a data model, setting training parameters, and inputting image data and labels for training. And evaluating the model by using the verification set, and outputting after optimization. And evaluating the performance by using the test set, and deploying the model for intelligent identification. According to the invention, data diversity is increased by using a data enhancement technology, a model structure and a training strategy are optimized, and an under-fitting problem is solved. And by increasing the complexity of the neural network, selecting a proper weight initialization method and a proper learning rate strategy, and increasing training rounds, the generalization ability of the model is improved. Meanwhile, different attitude changes are simulated by using a data enhancement technology, multi-attitude image data are collected, an attention mechanism and an attitude estimation model are introduced, and differential recognition of animals and plants under different attitudes is realized by combining a multi-scale feature extraction and fusion strategy.

The invention discloses a hospital terminal threat behavior recognition method based on interpretable deep learning, and particularly relates to the technical field of network security, threat intelligence analysis and interpretable artificial intelligence. According to the method, a multi-modal hospital terminal operation data set fusing time sequence images, logs and other data is constructed, and different modal data are effectively integrated by adopting a multi-modal data fusion method; integrating a dynamic attention mechanism into a long short-term memory network model for adaptively extracting key features in a terminal user behavior sequence; and a double-layer interpretable framework based on gradient weighted class activation mapping and a local interpretable model irrelevant interpretation technology is introduced, so that high-precision recognition and interpretable analysis of the threat behavior of the hospital terminal user are realized. According to the method, complex threat behaviors in the hospital terminal can be efficiently and accurately identified, the limitation of a traditional method in the aspects of data diversity, interpretability and hidden threat detection is overcome, and the transparency and efficiency of threat analysis are enhanced.

The invention belongs to the technical field of data processing, particularly discloses a bridge crack data set construction method based on multi-modal data enhancement and a GAN network, and aims to solve the key problems of small sample learning, insufficient data diversity, high-resolution image generation and the like in a bridge crack detection task. A solution integrating multi-modal data enhancement and deep symmetry GAN is provided, real scene interference is simulated through a combined enhancement strategy, 98% of interference types of a real scene are covered, the problem that a traditional enhancement technology cannot simulate complex shooting conditions is solved, a deep symmetry GAN network is designed, and the real scene interference can be simulated through the deep symmetry GAN network. According to the method, the chessboard effect of the generated image is eliminated, the high-frequency detail retention capability is improved, a 640 * 640-resolution image is generated, a high-quality bridge crack data set is constructed, bridge crack data is expanded, crack images in different forms are generated, and the problems of scarcity of the crack image data set and the like in bridge crack detection under a complex background are solved.

The invention discloses a large-scale language model fine tuning method, device and equipment and a storage medium, and relates to the technical field of large-scale language model fine tuning. According to the method, a preset check point is firstly subjected to large-scale language model fine tuning based on training loss performance, sample embedding space distribution and sample reply score performance of a current model; and performing multi-dimensional self-reference diagnosis on the training data set, identifying a suboptimal sample which is not matched with the capability of the current model, and then processing the suboptimal sample through an adaptive optimization engine to complete dynamic evolution of the training data set. Therefore, dynamic evolution of the data set adaptation model can be realized, the problem of redundancy or insufficient adaptation of static data is avoided, and the training efficiency is improved; meanwhile, data waste is reduced, data diversity is reserved, the model generalization ability is promoted, manual intervention is not needed, and the data optimization cost is reduced.

The invention relates to the field of data analysis, in particular to a database information acquisition and analysis method for pharmacological analysis of heat stroke, which comprises the following steps: determining a data state according to data diversity and data stability of rat input data, and determining a data processing mode according to the data state; in analog data supplementation, determining a supplementation mode according to a comparison result of the effective standard index proportion and a preset effective standard index proportion; in the keyword analysis, a use state is determined according to the matching frequency and the spacing distance of the effective keywords, and a corresponding search mode is determined according to the use state; in the effective keyword search process, determining a page distribution state according to the page similarity and the information relevancy of the target website, and determining an acquisition adjustment mode as track simulation adjustment or search simulation adjustment according to the page distribution state; the searching and matching efficiency of database information acquisition and analysis in pharmacological analysis related to heat stroke is improved.

The invention is suitable for the technical field of credit risk assessment, and provides a federated learning-based credit risk assessment method, which comprises the following steps of: obtaining a local model accumulation gradient and a global model gradient, and constructing a graph relationship between clients according to the similarity between the local model and a central server-side global model obtained by the local model accumulation gradient and the global model gradient; obtaining the difference of clients based on the graph relationship between the clients, calculating an aggregation weight, and forming global model update; and the central server side sends the aggregated global model update to each client side, and the client side continues the next round of local training until the model reaches the expected performance. The method has the effect of improving the global model generalization ability, and can be effectively applied to financial credit risk assessment. Through cooperative training, the problems of reduced accuracy, poor robustness and the like of the model effect due to insufficient data volume, data category or data diversity can be avoided, so that a credit risk assessment model with higher generalization ability is constructed.

The invention relates to the technical field of land type classification, in particular to a land type classification method based on territorial resource data, and the method comprises the following steps: S1, data preparation; s2, determining a classification target, a classification standard and a classification range; s3, land feature extraction and analysis; s4, data preprocessing and cleaning; s5, land feature extraction and selection; s6, formulating a classification method and algorithm selection; s7, land type division and marking; s8, determining a classification method and technology; and S9, formulating a classification standard and index system. Through land feature extraction and analysis, data preprocessing and cleaning and land feature extraction and selection, land data diversity understanding and effective management can be achieved, classification precision and efficiency are improved, and it is ensured that the classification result is more accurate and reasonable through verification and precision of the classification result; the reasonable utilization and protection level of land resources is improved through result optimization, and the accuracy and applicability of land classification are improved.

The invention discloses an improved YOLOv12-based aircraft surface damage detection algorithm and system, and relates to the technical field of aircraft surface damage, and the method comprises the following steps: S1, data collection and preprocessing: employing an unmanned plane and an unmanned vehicle to collect the related data of the aircraft surface damage, precisely marking the aircraft surface image, and increasing the data diversity; s2, model improvement: improving a YOLOv12 aircraft surface damage detection model according to an aircraft surface detection task; a high resolution input size is employed to capture small defect features. According to the method, the YOLOv12 aircraft surface damage detection model is improved, the unmanned aerial vehicle and the unmanned vehicle can comprehensively record key structural components on the surface of the aircraft under various illumination conditions, the data integrity is guaranteed, and through the CLAHE algorithm, diversified data and the imaging effect in various environments, the detection accuracy of the aircraft surface damage detection model is improved, and the detection accuracy of the aircraft surface damage detection model is improved. And the robustness of subsequent model training can be obviously improved, so that the detection algorithm has higher environmental adaptability.