459 results about "Text matching" patented technology

The invention discloses a power transmission line insulator defect detection method and system based on machine vision, and relates to the technical field of machine vision, and the method comprises the steps: combining a pre-trained DenseCLIP model with a YOLOv11 network, extracting an insulator image feature through a visual encoder of the DenseCLIP model, converting the insulator image feature into a visual embedded vector, and carrying out the detection of the defect of the insulator through the YOLOv11 network; constructing a text prompt according to a defect category, converting the text prompt into a language embedded vector, calculating the similarity between the vision embedded vector and the language embedded vector, generating a pixel-text matching score graph to guide YOLOv11 network learning, and constructing an insulator defect detection model; and constructing a mixed training set by using an artificial sample and a real sample, carrying out training by adopting a transfer learning staged training strategy, importing a to-be-detected insulator image into an insulator defect detection model, obtaining an enhanced feature graph and a pixel-text score graph after visual and language embedding vector fusion, and generating a final prediction result. According to the invention, the detection precision and real-time performance of insulator defects are improved, and the intelligent operation and maintenance efficiency of a power grid is remarkably improved.

The invention provides a multi-modal geographic positioning method based on a knowledge graph. The method comprises the following steps: graph construction: constructing a geographic knowledge graph; image positioning: calculating the entity similarity between the to-be-queried image and each sub-image so as to select an image positioning candidate sub-image, calculating the image matching similarity between the to-be-queried image and each image positioning candidate sub-image so as to determine an image positioning target sub-image, and taking the latitude and longitude coordinates of each image positioning target sub-image as a positioning result; text positioning: converting a to-be-queried text into map representation, calculating the map similarity between a text query sub-graph and each sub-graph to select a text positioning candidate sub-graph, and calculating the semantic similarity between the to-be-queried text and each text positioning candidate sub-graph, and calculating a text matching similarity corresponding to each text positioning candidate sub-graph based on the graph similarity and the semantic similarity corresponding to each text positioning candidate sub-graph so as to determine a text positioning target sub-graph, and taking latitude and longitude coordinates of each text positioning target sub-graph as a positioning result.

The invention discloses a transformer substation defect identification method and system based on multi-mode open set associative reasoning. The method comprises the steps that transformer substation oily equipment defect images are collected and marked; constructing a multi-modal data set of image-text matching; yOLOv8 is adopted to extract visual features, and a CLIP model is adopted to process text features; designing a re-parameterized visual language path aggregation network for feature fusion; training the model based on a contrast loss function; the applications are deployed after the performance is evaluated through multiple indexes; through the multi-modal feature fusion and the re-parameterization path aggregation network, the problem that the oil leakage defect is difficult to mark due to liquid flowability and form variability is effectively solved, the false detection condition under multi-device cross interference is remarkably improved, and by combining multi-scale feature learning and an attention mechanism, the oil leakage detection accuracy is improved. And the identification accuracy of the oil stain defect is greatly improved.

The invention discloses a graph representation learning and retrieval method and system based on a code warehouse graph structure, and the method comprises the steps: constructing a heterogeneous code graph related to an original code, a newly-added code or a modified code, carrying out the graph representation learning of the heterogeneous code graph, generating a node embedding vector, carrying out the self-supervised learning through an intelligent model, and carrying out the retrieval of the node embedding vector. And outputting the updated node embedding vector. The method comprises the steps of detecting whether newly-added codes or modified codes exist or not in real time, only performing heterogeneous graph modification on the newly-added or modified codes, receiving a query request from a user, generating corresponding code snippets according to the request, checking whether the target code snippets meet grammar specifications or not, and outputting a query result. According to the method, the structural relationship between codes is fully modeled, a structure perception retrieval mode is realized, the method is superior to a traditional text matching scheme, and the code semantic understanding capability is improved; and large-scale code graph online query is supported, and the applicability is wider.

The invention discloses a dance costume prop retrieval method based on causal reasoning and cross-modal matching, which comprises the following steps: in a matching task, calculating the similarity between costume prop images and text matching features by a model, and judging whether the costume prop images and the text matching features are matched or not according to a threshold value; and when the task is retrieved, calculating the similarity between the query text and the image feature set, and sorting to obtain a result. The process of calculating the features and the similarity comprises the following steps: firstly, extracting multi-scale visual features of an image by using Faster R-CNN, and extracting multi-granularity semantic features of a text by using BERT; then, the two types of features are subjected to self-attention and gating fusion respectively and then are input into a Transform layer, and final visual and text representation is obtained; then, multi-head attention calculation is carried out on the two representations, self-attention calculation is carried out after sequences are combined, and matching features are obtained; and finally, dividing the dot product of the two feature vectors by the modular length product to calculate the similarity. According to the invention, the matching accuracy of dance costume props can be improved.

The invention relates to the technical field of policy matching, and discloses an intelligent matching system based on big data and a matching method thereof. The system comprises a data acquisition module, a condition analysis module and an intelligent matching module. The data acquisition module is used for acquiring policy text, enterprise qualification and historical case data and constructing a matching basic database; the condition analysis module is used for extracting policy dominant conditions through a regular expression, mining implicit conditions from failure cases by utilizing an association rule algorithm, and calculating condition weights by adopting a random forest algorithm; the intelligent matching module integrates dominant and implicit conditions to calculate a matching coefficient, dynamically divides risk levels and generates an early warning, the matching comprehensiveness and the early warning sensitivity are improved by mining the implicit conditions, dynamically adjusting the matching model and reinforcing a learning optimization mechanism, upgrading from text matching to risk prejudgment is achieved, and the method has the advantages of being high in practicability and easy to popularize. The enterprise declaration success rate can be improved, and the repeated declaration rate is reduced.

The invention discloses an alarm information processing method and device and a storage medium, and relates to the technical field of data processing.The method comprises the steps that an unstructured alarm text is converted into an accurate keyword set through word segmentation processing, the limitation of traditional simple text matching is broken through, alarm core semantic information is effectively extracted, and redundant interference is reduced; according to the method, the mapping relation between the keyword and the alarm identifier is established and stored in the database, so that an efficient retrieval structure based on the inverted index is realized, the overhead of full-text scanning is avoided, and the retrieval speed of massive alarm data is remarkably improved; according to the method, the target alarm text can be more accurately positioned based on semantic relevance of the keywords, and the rigidity of traditional fixed condition retrieval is overcome, so that the practicability of alarm information retrieval is synchronously optimized in the aspects of efficiency and precision. The technical problem of low retrieval efficiency in a traditional retrieval mode is solved, and the technical effect of improving the retrieval speed of mass alarm data is achieved.

The invention relates to an AIS data-based port ship unloading area analysis method. The method comprises the following steps of: 1, preprocessing AIS data, namely carrying out cleaning screening and attribute supplementation on original data; 2, ship staying leg extraction: defining a staying leg by using motion features and expressing the staying leg by using a Geohash grid; step 3, port destination text matching: matching irregular destinations with standard port names; 4, bidirectional destination retrieval: analyzing an accurate destination by using a ship trajectory, and determining an affiliation port of a staying grid; and 5, grid integration and route cutting: integrating grids belonging to different ports to obtain an unloading area, and cutting a whole-cycle route of the ship. According to the method, the problems of complex unloading mode and complex route staggering around a port are solved, and reliable technical support is provided for trade monitoring of bulk energy goods such as crude oil by utilizing the complementary advantages of time-space information and destination texts in AIS data.

The invention discloses a multi-modal data processing method and device based on a large model, equipment and a medium. Performing feature extraction and association mining on the multi-modal data to obtain multi-modal data features; according to a retrieval sequence of the multiple text matching methods, sequentially using the multiple text matching methods to perform text retrieval on the multi-modal data features in a pre-constructed knowledge base to obtain a first associated sub-graph; based on the semantic similarity between the multi-modal data features and multi-modal data in a pre-constructed knowledge base and a similarity threshold value, determining a second associated sub-graph; performing standard evaluation on the plurality of candidate results in the first associated sub-graph and the second associated sub-graph, and sorting the plurality of candidate results according to an evaluation result to obtain a retrieval result; inputting the retrieval result and the multi-modal data features into a large model for information analysis to obtain prompt information and a target tool; and sending the prompt information and the target tool to the terminal, and receiving a data processing result of the terminal.

The invention relates to a weak supervision instance segmentation method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining a training image set and an image-level label, and generating a pseudo-semantic segmentation label; extracting multi-scale image features in the training image set, performing prompt contrast learning based on the multi-scale image features and the pseudo-semantic segmentation labels, and generating contrast loss and a pixel-text matching graph; splicing the multi-scale image features and the pixel-text matching graph to generate enhanced features, and training an instance positioning module based on the enhanced features to generate an instance center point and positioning loss; generating a semantic mapping graph through the segmentation network, and calculating self-matching loss of prompt guidance; calculating the total loss of the model based on the obtained loss, and performing iterative training on the instance segmentation model based on the total loss of the model to generate a target instance segmentation model; and generating a target instance segmentation result based on the to-be-processed image and the category text through the target instance segmentation model. According to the invention, the precision of instance segmentation is improved.

The invention relates to a cross-modal retrieval technology, in particular to a cross-modal image-text pedestrian retrieval method based on a generative model, which comprises the following steps: acquiring text description information and a corresponding original pedestrian image; calling a diffusion generation model to generate an intermediate image based on the text description; original image features and text features are extracted through an image encoder and a text encoder respectively, and semantic-fused text feature representation is constructed; calculating the similarity between the image feature representation and the text feature representation to obtain an image-text matching score; introducing and generating an intermediate image as a semantic bridge, and realizing multi-modal fusion among the image, the text and the intermediate image through a cross attention mechanism to obtain fused image and text features; and using the fusion features to train a recognition model, and finally outputting a pedestrian image retrieval result corresponding to the text description. According to the method, the image-text alignment precision can be improved, the robustness of a scene with incomplete text description is enhanced, and the recognition accuracy in a cross-modal retrieval task is improved.

The invention provides a maintenance guidance method and device for a household electrical appliance. The method comprises the following steps: constructing a unified vector database in advance based on text data, image data and audio data related to a household electrical appliance fault; acquiring field condition information of the household electrical appliance, wherein the field condition information comprises video information, audio information and / or text information; respectively carrying out image matching, audio matching and / or text matching on the field condition information based on the unified vector database, and outputting knowledge fragments as retrieval results; and inputting a retrieval result into the large language model, and outputting a maintenance guidance scheme of the household electrical appliance. According to the method, a retrieval and reasoning framework based on multi-modal feature direct fusion can be constructed, information dimension reduction loss is avoided by directly analyzing image, audio and text features of faults, and the comprehensive accuracy of fault diagnosis is improved to 95% or above; and through a dynamic knowledge retrieval and enhanced generation technology, the risk of'knowledge illusion 'possibly occurring in the professional field of the general large language model is effectively reduced.

The invention provides a networked automobile risk scene analysis method based on a multi-modal corpus and related equipment, and the method comprises the steps: receiving the multi-modal corpus containing various types of test data, and carrying out the format standardization and time synchronization processing of the multi-modal corpus; performing risk scene analysis on the basis of association information among different types of test data in the multi-modal corpus, identifying abnormal behaviors occurring in the test process, and generating corresponding structured event description information; and extracting key data fragments and image frames associated with the abnormal behaviors, carrying out image-text association on the key data fragments, the image frames and the event description information, and generating a test report based on a preset template. According to the method, a multi-modal corpus analysis mechanism is introduced, risk scene analysis and abnormal behavior recognition are executed, and image-text matching and report generation are automatically completed. The manual arrangement burden is reduced, and the intelligent degree of risk scene identification and the automation level of test report generation are improved.

The invention discloses an intelligent engineering quantity list matching method, which comprises the following steps of: firstly, performing deep analysis on project description to form a key feature dictionary and form a basis for subsequent accurate matching; then, adopting a two-stage process of coarse screening and fine arrangement: in the coarse screening stage, combining project codes pointing to professional classification with key features capable of quickly locking core elements, performing efficient filtering on a huge standard quota library, and quickly generating a candidate matching list which is controllable in scale and highly related; and in a fine ranking stage, turning to more detailed semantic and feature dual verification, comprehensively utilizing the deep semantic similarity of a complete item description text and an accurate comparison result of a key feature dictionary, and performing fusion calculation to obtain a final score so as to realize accurate ranking of candidate lists. In this way, the semantic gap caused by synonyms, inverted word order, non-standard expression and the like in traditional text matching can be effectively overcome, and the matching accuracy is greatly improved.

The invention discloses an unsupervised text-to-image pedestrian re-identification method and system. The method comprises the following steps: constructing an image-text matching relationship; clustering operation is carried out on the image features, and false identity tags are distributed to the pedestrian images and the corresponding text descriptions; constructing an image memory dictionary; obtaining a similarity set according to the image memory dictionary; removing pedestrian images with abnormal similarity and corresponding text descriptions according to the similarity set; respectively acquiring image features and text features of pedestrian images and text descriptions, and constructing a multi-stage ternary joint learning process; and obtaining text features to be re-recognized, and performing pedestrian image matching to obtain a pedestrian recognition result. According to the method, the intra-class compactness and inter-class separability of the samples are enhanced, and the retrieval precision and efficiency of the unsupervised text-to-image pedestrian re-recognition method are greatly improved.

The present disclosure provides a method and apparatus for generating a digital human video based on a large model, an intelligent agent, an electronic device, and a storage medium, which relate to a field of artificial intelligence technologies, in particular to technical fields such as deep learning, large models, computer vision and so on, and may be applied to scenarios such as video livestreaming, advertisement production, and e-commerce sales. The method for generating a digital human video based on a large model includes: acquiring a requirement information, where the requirement information includes an action description information for describing a specified action video segment, and the action video segment represents a specified action of a target object; processing the requirement information using a first large model to obtain a target script, where the target script includes a target speech segment text matching the action description information; and processing the target script and the action video segment using a second large model to obtain a target video for displaying a target digital human performing a speech delivery based on the target speech segment text while performing the specified action.

The invention discloses a dynamic knowledge fusion image question-answering method based on path correlation, which comprises the following steps of: S1, carrying out relationship type classification on an external knowledge base, and reserving a corresponding relationship type according to the characteristics of an image question-answering task to obtain a lightweight knowledge graph; s2, respectively extracting entities from the image and the question text, and mapping the entities to the lightweight knowledge graph nodes in the S1 to obtain a seed node set; s3, starting from the seed node set in the S2, generating candidate knowledge sub-graphs through breadth-first search; s4, performing correlation scoring on knowledge paths in the knowledge sub-graph in the S3 according to the text matching degree and the visual consistency; s5, iteratively optimizing the knowledge sub-graph through a greedy pruning strategy according to the knowledge path score in the S4; and S6, calculating a knowledge weight coefficient, realizing dynamic fusion of external knowledge, adaptively adjusting a knowledge embedding degree according to a question text, and inputting an image question and answer model to generate a final answer.

The invention discloses a text matching positioning method, system and device and a medium. The method comprises the steps of obtaining an original long text and segmenting the original long text into ordered text blocks; selecting the first N text blocks to form a first candidate set based on the common substring sorting of the to-be-matched text and each text block; intercepting the first candidate set according to head and tail characters of the to-be-matched text to obtain a second candidate set; calculating similarity scores of the text blocks in the second candidate set and the to-be-matched text through a sequence matching similarity algorithm; combining the scores with the first candidate set to determine a matching result and output a position, and judging that no matching exists when all the scores are lower than a preset minimum threshold value. According to the method, the limitation of a traditional matching mode is broken through through primary screening of common substrings, accurate interception of head and tail features and multi-layer processing of a sequence matching similarity algorithm, complex scenes such as text missing and semantic equivalent expression can be effectively handled, the accuracy and robustness of text positioning are improved, and the user experience is improved. The method can be widely applied to the fields of document management, information retrieval, academic analysis and the like.

The invention discloses a spoken language evaluation system and method based on multi-dimensional factors, and relates to the technical field of data processing, and the method comprises the steps: obtaining an audio text, a phoneme list and a phoneme time boundary list corresponding to voice data according to a pre-trained text phoneme recognition model; calculating a fluency score according to the phoneme list and the phoneme time boundary list; obtaining a semantic score of the audio text through a pre-trained semantic model; obtaining a grammar score of the audio text through a pre-trained grammar model; matching the audio text and the answer text according to a text matching method based on an editing distance to obtain a text matching score; the pronunciation score of the audio data is finally obtained through the phoneme confidence of the phoneme list; dynamically adjusting the weight of the score of each dimension according to the length of the audio text, and calculating a final spoken language evaluation score; through multi-dimensional evaluation, the spoken language ability is comprehensively evaluated according to dynamic weight adjustment of the scene, and comprehensiveness, fairness, accuracy and high efficiency of spoken language evaluation are achieved.

The invention relates to the technical field of natural language processing, in particular to a long text matching method and system based on key information and difference characteristics, and solves the problems that in the prior art, key information is dispersed due to noise interference in long text processing, and semantics of words or phrases in long text processing are more fuzzy and diversified. According to the method, a long text is preprocessed to obtain a test set sentence pair, a training set sentence pair and a verification set sentence pair, then a training model is obtained through sentence-level information entropy screening, word-level dynamic filtering, semantic difference enhancement and adaptive feature fusion, and finally after verification is conducted through the verification set sentence pair, the test set sentence pair predicts and outputs a result. The system comprises a long text matching system data preprocessing unit, a long text matching system function module training unit and a long text matching system function module result output unit. Key information is extracted for learning in long text matching, a large amount of computing power does not need to be consumed, and the matching effect is improved.

The invention discloses a secondary equipment alarm signal classification and grading method and system fused with artificial intelligence, and the method comprises the steps: collecting standard alarm signals and actual alarm signals of secondary equipment, dividing grades and types according to the types of the secondary equipment, and respectively constructing a standard table and an actual table of the alarm signals of the secondary equipment; constructing a text matching database, a text matching standard library, a text matching standard vector library and a text matching data set based on the standard table and the actual table; constructing a cascade text semantic matching model, and training the cascade text semantic matching model in combination with the text matching data set; and obtaining a to-be-processed secondary equipment alarm signal, and processing the to-be-processed secondary equipment alarm signal based on the text matching database, the trained cascade text semantic matching model and the text matching standard library to obtain a classification result of the to-be-processed secondary equipment alarm signal. According to the method, the intensity of manual intervention can be reduced, the warning signal grading and classifying efficiency is improved, and the misjudgment rate and the missed judgment rate are reduced.

The invention discloses a graph logic enhancement RAG framework for legal questions and answers. The graph logic enhancement RAG framework comprises a graph construction module and a logic enhancement reasoning module. The graph construction module constructs a heterogeneous knowledge graph comprising three entities of cases, legal provisions and historical query and five relationships of case-case and law provision-law provision, and provides a structured basis for reasoning. And the logic enhancement reasoning module is used for matching and positioning related entities by fusing semantics and texts through a dual-drive retrieval unit, mining cross-entity logic paths such as'query-case-law articles' and the like by utilizing a greedy algorithm, converting path information into natural language contexts, injecting the natural language contexts into a large language model, and generating answers containing clear legal basis. The modules form a'knowledge representation-logical reasoning-answer generation 'closed loop through knowledge graph association rules and reasoning chains. Knowledge matching efficiency, reasoning transparency and answering accuracy are improved, and the method is suitable for legal question and answer scenes.

The invention discloses a volume and noodle retrieval matching method and device and a storage medium. The method comprises the steps that a volume and noodle library is constructed; acquiring a first volume image; identifying and erasing handwritten handwriting; and paper surface information matching. According to the method, the volume information of the template is retrieved in a text matching mode, and compared with simple image similarity matching retrieval, the speed is higher, and the accuracy is higher. Moreover, according to the method, the front surface and the back surface of the test paper are respectively retrieved, and the confidence coefficient of the retrieval result is calculated according to the text word numbers of the front surface and the back surface of the retrieval result, so that the accuracy and the credibility of the retrieval result can be improved, and the method can be suitable for retrieval and matching of the test paper surface information of more subjects.

The present disclosure provides a training method and apparatus for an image-text matching model, a device and a storage medium. The method includes: acquiring a positive sample and a negative sample; where the positive sample includes text and an image, the text in the positive sample is used to describe content of the image in the positive sample; the negative sample includes text and an image, the text in the negative sample describes content that is inconsistent with content of the image in the negative sample; training the image-text matching model by using the acquired positive sample and the acquired negative sample based on a manner of contrastive learning; where the image-text matching model is used to predict, for an input image and input text, whether the input text is used to describe content of the input image.

The invention relates to the technical field of multi-modal cross-border advertisement generation, and discloses a multi-modal cross-border advertisement generation method and device, electronic equipment and a storage medium, and the method comprises the steps: greatly improving the image-text matching accuracy through introducing a cultural feature embedding and cross-modal alignment mechanism, and combining modeling and automatic screening, and improving the image-text matching efficiency. According to the method, high unification of advertisement contents in the semantic level is ensured, meanwhile, the defect of insufficient culture adaptability of a general template is effectively overcome, advertisement schemes violating regional culture rules can be accurately recognized and filtered by means of a culture feature vector library and adaptation index calculation, and the risk of misuse of culture is reduced to the minimum. The method has the beneficial effects that the audit passing rate and brand safety of the advertisement in a specific market are remarkably improved, full-link automation from content generation to final synthesis is realized, the generation efficiency is greatly improved, and the labor cost is reduced.

The invention discloses a similarity calculation method based on semantic editing fusion. The method comprises the following steps: calculating global semantic similarity of a source text and a target text by utilizing a semantic coding model; obtaining a first keyword list of the source text and a second keyword list of the target text through word segmentation processing, taking each first keyword in the first keyword list as a target word, and respectively forming a plurality of to-be-compared word pairs with a corresponding word in the second keyword list and a plurality of adjacent second keywords; based on the to-be-compared word pairs, calculating the semantic similarity between each group of target words and the corresponding words by utilizing a semantic coding model, and determining the editing distance between the source text and the target text; and determining the local semantic similarity of the source text and the target text according to the editing distance, and combining the global semantic similarity to obtain the final text matching similarity. According to the method, the problem that text matching only depends on character-level surface matching and neglects semantic association between word pairs is solved, and the text matching precision is improved.

The invention relates to a fault detection method of an industrial robot and a fault detection database construction method and device of the industrial robot, and relates to the technical field of industrial robots. The method comprises the following steps: determining the current fault duration under the same operation beat based on a plurality of pieces of operation data and time attributes of the industrial robot under the same operation beat; according to the current fault duration, correcting the time attribute of each piece of operation data from the quantized time to the corrected time; based on the time attribute of each piece of operation data and the current fault duration, determining reference sample operation data corresponding to each piece of operation data in multiple pieces of sample operation data of a pre-constructed fault detection database; and performing text matching on each piece of operation data and the corresponding reference sample operation data to obtain a fault detection result of the industrial robot. By adopting the method, the fault detection accuracy of the industrial robot can be improved.

The invention relates to the technical field of X-ray contraband detection, in particular to a small sample X-ray contraband detection method based on text matching and feature decoupling. In the training process, explicit features and implicit features of the contraband are separated through an overlapping feature interaction module, dynamic interaction of double-flow features is established through a cross attention mechanism, cross-modal similarity calculation is carried out on query vectors of the explicit features and key vectors of the implicit features, potential discrimination clues contained in implicit space are mined, and the identification accuracy of the contraband is improved. Overlapping feature decoupling and detail information supplementation are realized; extra text priori knowledge is introduced through a visual text matching module, the risk that classifier boundaries are difficult to distinguish and model fine adjustment is prone to over-fitting is relieved, inter-class discrimination and intra-class compactness of contraband features are gradually enhanced through multi-layer feature transformation, alignment of visual features and text information is achieved, and the recognition accuracy of the contraband features is improved. And the discriminative features of the types of the prohibited goods are enhanced. According to the invention, the precision of X-ray contraband detection and the detection performance of the model are improved.

The invention discloses a text matching method and device based on probability distribution, equipment and a storage medium, and relates to the technical field of natural language processing, and the method comprises the steps: obtaining semantic feature distribution of each professional knowledge text from a professional knowledge base, obtaining a knowledge probability distribution set, abstracting the intrinsic features of the text through probability distribution, and obtaining a knowledge probability distribution set; and standardized representation of the knowledge base is realized. The text input by the user is obtained, the corresponding user text probability distribution is calculated, and the fault tolerance of text matching is remarkably improved. The similarity distance between the user text probability distribution and each distribution in the knowledge probability distribution set is calculated, and the overall semantic similarity instead of local matching is measured through the distribution distance, so that the comparison process can tolerate the distribution offset. Finally, a text matching result is determined according to the minimum value of the similarity distance, the effect of stably retrieving semantic related professional knowledge under high noise is achieved, and fault tolerance and matching precision are remarkably improved.

The invention discloses a method and a device for identifying defects of power transformation equipment, which are used for solving the technical problem that the existing method for identifying the defects of the power transformation equipment limits the environment adaptability of a model, so that the identification effect is poor. The method comprises the following steps: acquiring a substation equipment image and a plurality of predefined substation equipment state text descriptions; and adopting a preset lightweight image-text matching model to carry out substation equipment defect identification according to the substation equipment image and the plurality of predefined substation equipment state text descriptions, and outputting a substation equipment defect identification result.