256results about How to "Reduce redundant information" patented technology

The invention provides a riverbed evolution prediction method based on multi-source data fusion, and belongs to the technical field of hydrological resource monitoring. The method comprises the following steps: acquiring and fusing multi-source data by using different devices, performing data preprocessing to form unified three-dimensional grid data, and then generating a multi-scale feature map based on a designed feature extraction network; based on the designed multi-channel convolutional neural network, learning and extracting features on multiple channels at the same time, and performing hierarchical expression and fusion on different information sources; and finally, based on the designed time sequence regression network, extracting a complex dependency relationship in the time sequence through a double-layer LSTM structure, learning residual error information through an adaptive residual error correction module to improve prediction precision, and obtaining a final prediction result. The method shows relatively high applicability and robustness in a complex river channel environment, and can cope with various river bed form changes caused by natural factors and human activities through multi-source data fusion and an innovative feature extraction and regression method.

The invention provides a multi-modal sentiment analysis method based on unwrapping representation learning, which comprises the following steps: acquiring a multi-modal sentiment analysis data set, and extracting features of texts, audios and videos; constructing a representation learning network, extracting private features and shared features, and processing through adversarial learning optimization and time smoothness constraint; under the guidance of different language scales, the private features of the audio and the video are fused; further fusing text, audio and video features through cross-modal attention and a gating mechanism; and finally, inputting the fusion features into an emotion analysis module, and generating an emotion prediction result in combination with an emotion classification or regression task. According to the method, the consistency of the audio and video features in continuous time is considered, redundant information is effectively reduced, the time dimension consistency is ensured, deep fusion of the audio and video features is realized under text guidance, and the cross-modal fusion effect is remarkably improved.

The invention discloses a satellite remote sensing target data identification method based on artificial intelligence, which belongs to the technical field of data processing, and realizes distributed model training and data mining of a multi-node joint model by performing satellite remote sensing target identification on an acquired spectral vector and adopting a model training architecture of distributed federated learning; carrying out manual annotation on the acquired spectral vector data; a generative adversarial network based on multi-scale spectral decomposition is adopted for sample generation, and then data expansion is achieved; and performing data identification by adopting a satellite remote sensing target data identification model. According to the method, the generative adversarial network based on multi-scale spectral decomposition is adopted for sample generation, and the data set is expanded. A self-encoder algorithm is used for carrying out feature dimension reduction, feature independence is enhanced through a nonlinear conversion layer, and redundant information between features is reduced. A kernel extreme learning machine algorithm based on conjugate gradient compensation is adopted, a kernel function is mapped to a high-dimensional space, and the classification accuracy and efficiency are improved.

The invention discloses a laser radar pose estimation method based on multilayer iterative extended Kalman filtering, and belongs to the technical field of radar navigation. The method aims at solving the problems that an existing laser radar pose estimation method is low in operation efficiency and poor in estimation precision. Comprising the following steps: calculating a state predicted value # imgabs0 # and a covariance matrix predicted value # imgabs1 # according to a state value xk-1 and a covariance matrix Pk-1, obtaining an observation matrix # imgabs2 # by adopting a scan-to-submap matching method insensitive to an initial value, updating the predicted value to obtain updated # imgabs3 # and # imgabs4 #, and performing iterative distortion correction on preprocessed point cloud data by combining a High-level motion compensation method; performing iterative updating on the updated state value # imgabs5 # by combining a low-level state updating method, and obtaining a current state optimal value xk after convergence; and then the updated covariance matrix # imgabs6 # is updated, and the current covariance matrix Pk is obtained. The method is used for pose estimation.

The invention discloses a multi-modal traffic flow prediction method based on multi-source data feature fusion. The method comprises the following steps: acquiring traffic data; dividing a training set and a test set according to the traffic data; constructing a traffic prediction model, and training the traffic prediction model by adopting the training set; verifying the traffic prediction model by using the test set; the traffic prediction model comprises a cloud image encoder used for carrying out feature extraction on cloud image data to obtain cloud image features; the space-time encoder is used for performing feature extraction on the space-time data to obtain space-time features; the fusion module is used for fusing the cloud image features and the spatial-temporal features to obtain fusion features; and carrying out addition operation on the cloud picture features, the spatial-temporal features and the fusion features, and inputting into a spatial-temporal decoder for prediction to obtain a final prediction result. According to the method, multi-azimuth and multi-dimensional information is fused, redundant information in the spatial-temporal feature extraction process is simplified, efficient and high-precision prediction is achieved, and the method can be widely applied to the technical field of traffic flow prediction.

The invention discloses a data processing and quality evaluation method for efficiency test data of an electric drive system. The method comprises the following steps: acquiring an original electric drive efficiency test data set; performing noise reduction on the original electric driving efficiency test data set through a probability distribution-based data noise reduction method to obtain a noise-reduced electric driving efficiency test data set; carrying out the first abnormity detection and processing of the noise reduction electric drive efficiency test data set through a DBSCAN model, and obtaining a preliminary screening electric drive efficiency test data set; carrying out the second anomaly detection and processing of the preliminary screening electric drive efficiency test data set through an LOF algorithm model and an iForest algorithm model, and obtaining a preprocessed electric drive efficiency test data set; and performing quality evaluation on the preprocessed electric drive efficiency test data set through the constructed six-tuple data quality model. According to the invention, effective noise reduction and abnormity detection can be carried out on the efficiency test data of the electric drive system, so that the data quality is improved.

The invention discloses an information retrieval and guidance method and device based on a big data software system, relates to the technical field of data processing, and solves the technical problem that retrieval results often contain a large amount of irrelevant or redundant information due to lack of deep analysis of retrieval problems and effective integration and screening of the retrieval results. According to the method, the retrieval problem is finely split according to the subject content, the retrieval intention of the user can be more accurately understood, compared with a traditional retrieval mode based on overall keywords, the retrieval pertinence and accuracy are improved, the situation that the retrieval result is disordered and inaccurate due to general use of the keywords is avoided, and the retrieval efficiency is improved. According to the semantic-based analysis method, information with similar contents can be identified more accurately, the limitation of matching only based on vocabularies traditionally is overcome, redundant information is effectively reduced, the retrieval result is more refined and valuable, the efficiency of obtaining useful information by a user is improved, and the user experience is improved.

The invention discloses a soil organic matter content inversion method based on a near infrared spectrum, and relates to the technical field of soil detection. Compared with a previous soil organic matter content measurement technology, the problems that in the aspect of soil organic matter content measurement, an existing near infrared spectrum technology is complex in data preprocessing, difficult in feature selection and insufficient in model generalization ability are solved; the near infrared spectrum range is optimally selected, key characteristic wave bands related to soil organic matter changes are effectively focused, and the accuracy and reliability of measurement are improved; a differential transformation preprocessing technology is used, and the data quality is effectively improved; efficient feature selection is carried out by adopting a stable competitive self-adaptive reweighted sampling method, a feature wave band which can reflect the soil organic matter content is extracted, and the prediction capability of the model is enhanced. And a plurality of machine learning algorithms and parameter optimization technologies are combined to optimize the model structure and parameters so as to adapt to soil samples of different areas and different types, and the prediction precision and stability are improved.

The invention discloses an active learning-based electric power knowledge graph construction method and system, and relates to the technical field of electric power market transaction data analysis, and the method comprises the steps: collecting data, cleaning the collected data, generating word prediction task data of an electric power transaction scene, and training an electric power field word embedding model; generating an entity relationship training sample, inputting a word embedding model, performing entity relationship label prediction through an entity relationship joint extraction model, extracting a high-quality sample based on active learning to perform part-of-speech tagging, and training the model based on a tagged sample; and applying the trained entity relationship joint extraction model to power transaction data to construct a knowledge graph, performing manual correction and entity unification, and storing the knowledge graph into a Neo4j database. According to the method, active learning and deep learning are combined, so that the sample labeling cost is effectively reduced, and meanwhile, the model performance is ensured.

The invention relates to the technical field of data processing, and particularly discloses a model optimization method and system based on big and small model collaborative intelligence, and the method comprises the steps: obtaining multi-source heterogeneous data, carrying out the feature extraction of the multi-source heterogeneous data based on a big language model, and obtaining a plurality of pieces of key feature information; obtaining entity information according to the key feature information, and performing large model training on all the entity information to obtain a large model training knowledge graph; and obtaining a first entity evaluation index and a first change trend corresponding to each piece of entity information according to the large model training knowledge graph, and screening the entity information according to the first entity evaluation index to obtain a key feature entity. According to the method, multi-source heterogeneous data is acquired, key feature information is extracted from unstructured and structured data by using a large language model, a knowledge graph is constructed, entity relationship reasoning, information completion and semantic understanding are performed, and the comprehensiveness and accuracy of large and small models are improved.

The invention relates to an early warning event processing system of an oil and gas pipeline optical fiber early warning system, and the system comprises a preprocessing module which is used for receiving and preprocessing the real-time early warning data of a plurality of early warning events, and obtaining the preprocessed real-time early warning data; the initial merging module is used for determining a first similar early warning event in the plurality of early warning events according to the preprocessed real-time early warning data, and performing initial merging on the real-time early warning data of the first similar early warning event to obtain a merged early warning event; the dynamic upgrading and downgrading processing module is used for acquiring a multi-dimensional feature vector of the combined early warning event; using a deep learning model to adjust the initial early warning level of the combined early warning event according to the multi-dimensional feature vector to obtain an adjusted early warning level; and the output module is used for outputting the combined early warning event and the adjusted early warning level thereof. According to the system, repeated and redundant information can be reduced, and the early warning event processing efficiency, response speed and dynamic flexibility are improved.

The invention belongs to the field of electroencephalogram emotion recognition, and provides an emotion recognition method and system for a multi-scale convolutional network based on bidirectional attention, and the method comprises the steps: segmenting an electroencephalogram signal according to a time window, extracting differential entropy (DE) and power spectral density (PSD) features, introducing an electroencephalogram map to embed spatial information, and capturing a spatial dependency relationship in the signal. Then, the preprocessed features are input into an MSA-TCN model, and the model combines the attention in the bidirectional features, the attention between the bidirectional features, a multi-scale convolution network and a time sequence convolution network to effectively extract dynamic time features and cross-channel spatial features; the invention aims to overcome the limitation of the prior art in emotion recognition through an innovative bidirectional attention mechanism and a multi-scale convolutional network structure (MSA-TCN).

The invention discloses a dam effect quantity prediction method and system, a medium and equipment, and relates to the technical field of safety engineering, dam monitoring data is acquired, influence factors of a dam effect quantity are determined, and effect quantity influence factor samples are constructed; performing dimension reduction processing on the influence factor sample through a principal component analysis method to obtain an influence factor matrix after decoupling and dimension reduction; coupling the long and short-term memory neural network with an artificial fish swarm algorithm, constructing a dam effect quantity prediction model, performing parameter setting on the dam effect quantity prediction model, and optimizing the initial weight and the number of neurons of the long and short-term memory neural network by using the artificial fish swarm algorithm; training the optimized long and short term memory neural network until convergence, and obtaining a trained dam effect quantity prediction model; and performing prediction and anti-normalization processing to obtain an effect quantity prediction result. And the accuracy, the stability and the efficiency of dam effect quantity prediction are improved.

The invention discloses a wind turbine generator variable pitch system fault diagnosis method for data balance. The method specifically comprises the following steps: acquiring operation data and fault data of a wind turbine generator through a monitoring and data acquisition system equipped on a wind power plant; based on the obtained data, screening sensitive features related to the fault of the variable pitch system, reducing the dimension of a data sample and reducing the input of redundant information; according to the quantitative relation, dynamic balance factors are distributed for various types of samples, the dynamic balance factors are introduced into a multi-stage diagnosis model Dualformer based on double Transformers, and a DBFDformer data balance fault diagnosis model is constructed; and training a DBFDform model by using the screened feature data, and completing dynamic updating of balance factors and optimization of model hyper-parameters through a snow ablation optimization algorithm. The wind turbine generator variable pitch system fault diagnosis method can effectively improve the diagnosis precision and performance of the wind turbine generator variable pitch system fault under the unbalanced data, has excellent generalization ability and robustness, provides a new thought for the fault diagnosis of the wind turbine generator variable pitch system under the unbalanced data sample, and is helpful for improving the utilization rate of wind energy.

The invention discloses a tree coding-based long video generation optimization method, which comprises the following steps of: constructing a video frame sequence, and obtaining frame features; clustering the frame features to obtain a correlation score between the key frame and the video query; calculating the number of high-correlation clusters, and outputting a corresponding frame feature set and a correlation score; key frames of all the high-correlation clusters serve as first-layer nodes of a video feature tree, the high-correlation clusters are clustered into w sub-clusters respectively, a second layer is constructed based on the sub-cluster key frames, the correlation score of the second-layer nodes is obtained by subtracting one from the correlation score of the first-layer nodes, each layer is constructed according to the method, and the first-layer nodes and the second-layer nodes are obtained; the relevance score of the last layer of nodes is 1; the large language model outputs an optimization reasoning result based on the speculation sampling principle and the draft model; and inputting the optimized reasoning result and the video frame sequence into a preset video generation model to obtain an answer video. According to the method, the reasoning accuracy and efficiency are improved by inquiring self-adaptive and hierarchical long video understanding.

The invention relates to the field of industrial equipment monitoring, and discloses a multi-sensor integrated system for remote monitoring of industrial equipment, which comprises a data acquisition module, a data processing module, a fault prediction module, a self-adaptive control module and a monitoring feedback module, the data acquisition module acquires multi-dimensional data from sensor arrays of a plurality of devices; the data processing module carries out noise reduction processing on the data and extracts key features; the fault prediction module predicts a short-term fault of the equipment and predicts a long-term operation trend; the self-adaptive control module adjusts control parameters of the equipment according to the prediction result and optimizes a control strategy; and the monitoring feedback module generates real-time fault alarm information by realizing remote monitoring and feedback of the equipment. The system can improve the safety and stability of equipment operation, optimize equipment management and reduce the equipment maintenance cost, and has high real-time performance, accuracy and intelligent level.

The invention requests to protect a rolling bearing fault diagnosis method based on envelope spectrum knowledge distillation. The method comprises the following steps: obtaining a rolling bearing vibration signal, and carrying out sample division and normalization processing; the spectral kurtosis of the source working condition sample is calculated, the center frequency and bandwidth of envelope demodulation are determined according to the spectral kurtosis, envelope demodulation is carried out, and the demodulation sample is sent to a teacher network phi for pre-training; respectively inputting an original vibration sample and an envelope spectrum sample under a source working condition into a student network psi and a pre-trained teacher network phi, and extracting bearing health state representations learned by the two networks to establish envelope spectrum knowledge distillation loss; calculating the prediction loss of the student network, the cross-working-condition sample probability distribution alignment loss and the diversity loss represented by the health state; carrying out multi-loss back propagation, and optimizing the weight and bias of the fault diagnosis network by adopting an Adam optimizer; and finally, inputting the target working condition sample into the trained fault diagnosis model for performance testing.

The invention provides a video processing method and device based on multi-modal information fusion, equipment and a medium, and relates to the field of video processing. According to the method, the voice text in the video is processed through the large language model, the voice information can be accurately recognized and transcribed, the unstructured voice data is converted into the structured text data, and operable text information is provided for subsequent analysis. The image key frames are extracted by using the image features of the video data and the text features of the obtained voice recognition text, the most representative frame with the maximum information amount can be screened out from a large number of video frames, redundant information is reduced, and meanwhile, the synchronism of visual information and voice information is ensured. Based on the time correspondence, the image key frame and the speech recognition text are aligned and fused, and the multi-modal fusion text is output, so that the time sequence association and context information of the video can be captured more accurately, and the accuracy and depth of video understanding are improved.

The invention belongs to the technical field of stratum recognition, and particularly relates to a shield parameter geological recognition calculation method and system based on an unsupervised clustering algorithm, and the method comprises the steps: selecting a plurality of groups of data containing factors affecting stratum category recognition; pCA converts a group of complex related variables into fewer and irrelevant variables, and the new variables are called as main components and can capture most of information in original data; determining the clustering number of K-means + + by adopting an elbow method; determining the clustering number of a hierarchical clustering method by adopting a hierarchical tree; k-means + + and hierarchical clustering are respectively used for clustering the data, and evaluation indexes such as identification precision or contour coefficients of the data and the data are calculated, so that potential modes in the data are analyzed and mined more comprehensively. The invention aims to deeply explore the internal relevance between shield tunnel parameters and geological types, provides an innovative and robust solution for real-time detection of geological features in shield construction by overcoming the limitation of the prior art, and has wide application potential.

The invention belongs to the field of computer vision, and particularly relates to an edge detection technology based on Transfomer. Firstly, according to the method, a scale enhancement technology is adopted, and an input image is divided into a plurality of scales with different sizes, so that detail information in the image can be better processed; secondly, according to the method, a feature enhancement technology is adopted, and the features extracted by the network are enhanced, so that feature information is utilized more fully, and the accuracy and robustness of edge detection are improved. And finally, a feature fusion technology is introduced in the method, and features of different scales and different stages are fused, so that the feature expression capability is further enhanced, noise generation is effectively inhibited, and a clear edge is generated. Through comprehensive application of the processing technology, the edge detection method can effectively avoid the problems of loss of partial details of the edge, rough lines, disordered content, noise and illumination interference and the like when a complex background is processed, the accuracy and reliability of edge detection are improved, and meanwhile, the efficiency of edge detection is improved.

The invention discloses a self-adaptive point cloud sampling adjustment method. The method comprises the following steps: training a feature analysis model; sampling a workpiece at an initial sampling rate to obtain point cloud data of the workpiece, inputting the point cloud data into the feature analysis model to obtain regional features, and dividing a local region based on the regional features; and adjusting the sampling rate of the local region based on the region features. The local structure of the point cloud can be reflected more finely based on the division of the regional features, so that the subsequent sampling rate adjustment is more accurate. The refined division mode is beneficial for reserving important features such as edges and angular points in the point cloud, removing redundant information at the same time, and providing a more valuable data basis for subsequent point cloud processing tasks.

The invention belongs to the technical field of unit fault diagnosis, and discloses a fault diagnosis method and system for a fault large-scale unit based on sharing and specific representation learning. The method not only can be applied to multi-measuring-point fault diagnosis of a large-scale unit, but also has wide applicability. In other multi-sensor systems or complex equipment, the fault detection effect can also be improved through multi-view representation learning. The invention provides a weighted sharing-specific multi-view depth representation learning fault diagnosis method for fault diagnosis of multi-measuring-point data of a large-scale unit. The method consists of two parts, namely sharing and specifically representing a learning network and weighted multiple similarity loss and orthogonal loss. A multi-measuring-point data fault diagnosis problem of a large-scale unit is solved by sharing a specific representation learning network, and weighted multiple similarity loss and orthogonal loss are used to further improve the feature extraction capability of the network and reduce redundancy.

The invention relates to the technical field of data processing, in particular to a data compression and block storage processing method based on deep learning. The method comprises the following steps: acquiring integrated data to be compressed; analyzing the effective to-be-compressed integrated analysis data according to the to-be-compressed integrated data; constructing a heterogeneous data feature analysis tree model; extracting to-be-compressed feature data of the effective to-be-compressed integrated data through the heterogeneous data feature analysis tree model; designing a recursive sparse coefficient evaluation objective function based on the to-be-compressed feature data; performing adaptive sparse compression processing on the effective to-be-compressed integrated data based on the recursive compression sparse coefficient evaluation objective function to obtain compressed data; performing block processing on the compressed data to generate block compressed data; and performing compressed data block storage optimization processing on the partitioned compressed data to generate an optimized storage compressed data block. According to the method, data features are analyzed through a deep learning algorithm, so that efficient processing of data compression and block storage is realized.

The invention discloses an acne grading method, system and equipment based on multi-modal fusion and a medium, belongs to facial acne severity grading in the technical field of artificial intelligence, and aims to solve the technical problem of poor facial acne severity grading effect in the prior art. A constructed facial acne grading model comprises a VGG feature extraction module, a multi-modal feature fusion module, an acne feature sensing module and a comprehensive grading module, the VGG feature extraction module extracts facial features in different modals, and the multi-modal feature fusion module models the relation between the facial features in different modals through cross attention for fusion. The acne feature sensing module enhances the sensing of acne features through a cross attention mechanism, and the comprehensive grading module makes a comprehensive and accurate acne severity grading result. According to the method, comprehensive grading of the severity of the acnes is realized by combining the multi-modal face pictures of the patient, and the grading effect of the severity of the facial acnes is greatly improved.

The invention discloses a short video information detection method, device and equipment and a storage medium, and relates to the technical field of network space security information detection.The method comprises the following steps that multi-modal information in a target short video is obtained, and multiple modal features corresponding to the modal information are extracted; performing feature alignment on the title features and other modal features to obtain a plurality of key features most related to the title; modeling the consistency among a plurality of key features most related to the title to obtain a first fusion feature representation and a second fusion feature representation; splicing the first fusion feature representation, the second fusion feature representation and the title feature to obtain a multi-modal fusion feature; and classifying the multi-modal fusion features to obtain a detection result of the multi-modal information in the target short video. According to the method, modeling is carried out based on feature alignment and semantic consistency, key information in the short video is better mined and fused, and the accuracy of an information detection result is greatly improved.

The invention discloses a self-adaptive differential event detection method for an infrared moving target, which relates to the technical field of infrared detection and infrared space, and comprises the following steps: S101, initializing a current frame and a threshold value; s102, infrared differential event detection of a self-adaptive threshold value is carried out; s103, preprocessing an infrared event frame; s104, target event feature extraction is carried out; and S105, abnormal events are removed. In order to balance a detection rate and an error triggering rate for a space-based detection background with complex change and a dark moving target observation task, the invention provides a method for adjusting an event triggering threshold value based on triggering rate feedback, self-adapting to a strong radiation background with high dynamic change, flexibly and efficiently obtaining an infrared differential event frame, reducing static background redundant information, and improving the detection efficiency. And the detection efficiency and the moving target discovery capability are improved. And furthermore, the time-space characteristics of the target event are utilized to extract the centroid coordinates of the target area, so that high detection rate and low false alarm rate are realized, and the method has real-time performance and stability and relatively high engineering application value.

The invention provides a video compression method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the frame-by-frame disassembly of a source video uploaded by a user, and generating a continuous frame sequence; performing image semantic recognition on each video frame in the frame sequence to obtain a semantic word set corresponding to each video frame; according to the semantic word sets, key frames and non-key frames in each video frame are judged frame by frame, the semantic word sets corresponding to the key frames are compared, and repeated frames identical to the key frames are determined; and deleting the non-key frames and the repeated frames, and generating a compressed video to complete compression of the source video. Compared with the prior art, the content of each frame can be analyzed in detail; key information in the video can be identified, non-key frames and repeated frames can be deleted, and redundant information in the video can be remarkably reduced; therefore, the video compression efficiency can be improved, and the video compression quality can be optimized.

The invention discloses a health state prediction method based on multi-vital sign time sequence data. The method comprises the following steps: acquiring vital sign time sequence data of a plurality of different modalities; the multiple pieces of vital sign time sequence data are input into a slice and block frequency domain time sequence processing module to be preprocessed, an enhanced feature matrix is obtained, and preprocessing comprises the steps that time slice division is conducted on each piece of vital sign time sequence data, multiple pieces of fragment data are obtained, feature extraction is conducted on each piece of fragment data, and time domain fragment features are obtained; performing frequency domain conversion on each time domain fragment feature to obtain a frequency domain fragment feature, and converting the frequency domain fragment feature back to a time domain to obtain an enhanced feature matrix; the enhanced feature matrix is input into a Transform module, a health state prediction result is obtained, and the Transform module comprises an encoder module, a decoder module and an output layer module. According to the method, the adaptability and generalization of the model in coping with multi-modal and dynamic characteristics can be improved.

The invention relates to the technical field of remote sensing, and discloses an unfavorable geology detection method, system and device based on multi-source aerial remote sensing data fusion, and the method comprises the steps: obtaining laser radar data, hyperspectral image data, CCD image data, infrared image data and integrated navigation data of a target area, and carrying out the preprocessing; carrying out space matching processing on the preprocessed laser radar data, the preprocessed hyperspectral image data, the preprocessed CCD image data and the preprocessed infrared image data; fusing the laser radar data, the hyperspectral image data, the CCD image data and the infrared image data after application space matching processing for geological detection to obtain different geological detection data; and obtaining an unfavorable geological detection result of the target area in combination with different geological detection data. According to the method, the multi-source data fusion processing precision can be improved, multi-source heterogeneous information is effectively integrated, comprehensive and efficient detection of various unfavorable geologies is realized, the misjudgment and missed judgment risks are reduced, and the credibility of unfavorable geology detection results is improved.

The invention discloses a power emergency unmanned aerial vehicle power transmission line defect detection method, which comprises the following steps: receiving a power transmission line image acquired by a power emergency unmanned aerial vehicle, and preprocessing the image to accurately identify and distinguish defects and backgrounds; carrying out feature extraction on the image, and obtaining statistical property features through texture analysis, a gray-level co-occurrence matrix, a gray-level difference method, discrete wavelet transform and fast Fourier transform methods; fusing and selecting the extracted features to form an optimized feature set; using the feature set to train a super-pixel segmentation model based on deep learning for power transmission line defect detection; and adopting an average threshold method to carry out confidence analysis on the deep learning model after super-pixel segmentation. Through multi-domain feature extraction optimization, high-precision defect detection and robust confidence analysis, the accuracy and efficiency of power transmission line defect detection are significantly improved.