Drawing-to-configuration AI generation method and system based on CV large model

Abstract

The invention relates to the field of electrical technology, and discloses a drawing-to-configuration AI generation method and system based on a CV large model, and the method comprises the steps: constructing an original sample set based on a single-cabinet drawing, and screening out a complex primitive sample set from the original sample set; secondly, labeling basic primitives and non-detachable combined primitives in each sample graph of the original sample set and complex primitives in each sample graph of the complex primitive sample set; and training the CV visual large model based on the labeled original sample set and the labeled complex primitive sample set to obtain a basic primitive recognition model and a complex primitive recognition model. And finally, the complex primitive recognition model and the basic primitive recognition model are sequentially adopted to recognize the target system diagram, the type and coordinates of each primitive and the topological relation between the primitives are obtained, and a configuration diagram in a vector format is generated. When the method and the system disclosed by the invention are applied in a complex circuit design scene, the accuracy and the reliability of electrical drawing identification can be remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of electrical technology, and in particular relates to an AI generation method and system from drawings to configurations based on a CV large model. Background Art

[0002] The automatic review and conversion of electrical drawings has a wide range of application value in the field of electrical engineering. For example, in the power SCADA system, the primary system diagram can be automatically converted into an interactive monitoring interface to achieve real-time monitoring and control of the power system operation status; in the industrial digital twin system, by identifying the P&ID (process and instrument flow chart) drawings, the three-dimensional configuration model of the equipment can be quickly generated; in the electrical intelligent operation and maintenance system, combined with the data collected by the Internet of Things (IoT), the equipment operation status, such as temperature, load and other key information, can be displayed in real time in the SVG format graphical interface.

[0003] However, the traditional drawing review and conversion process mainly relies on manual operation, which is cumbersome and inefficient, and is prone to errors or inconsistencies due to human factors, which seriously restricts engineering efficiency and system reliability. With the development of artificial intelligence (AI) recognition technology, computer vision (CV)-based methods can quickly and accurately identify various elements in electrical drawings, significantly improve the efficiency of drawing review and conversion, and reduce human errors.

[0004] At present, the technology of automatic drawing recognition is still in the development stage and has certain limitations. The main problems include the lack of large-scale, high-quality computer vision pre-training models for the field of electrical components. When processing complex electrical schematics, the recognition results generated by AI may be erroneous or logically unreasonable. The root cause of this problem is that the performance of AI models is highly dependent on the quality and quantity of training data. In the field of electrical drawing recognition and schematic drawing, if the training data coverage is insufficient, the samples are unevenly distributed or the annotations are inaccurate, it will cause AI systems to make recognition errors in actual applications, limiting their reliability and promotion and application.

[0005] Therefore, how to improve the accuracy and reliability of electrical drawing recognition, especially for complex circuit design scenarios, is still an important technical problem that needs to be solved urgently. Summary of the invention

[0006] In an embodiment of the present invention, a method and system for AI generation from drawings to configurations based on a CV large model are provided to solve the problem that when the prior art processes complex electrical schematics, the recognition results may be erroneous or logically unreasonable.

[0007] In order to solve the above technical problems, the embodiments of the present invention disclose the following technical solutions: One aspect of the present invention provides an AI generation method from drawings to configurations based on a large CV model, which is applied to the primary system diagram of an electrical system and includes: Construct an original sample set based on the single cabinet diagrams disassembled from multiple primary system diagrams, and each sample is a single cabinet diagram or a combined cabinet diagram formed by combining multiple single cabinet diagrams; Select samples containing preset complex graphic elements from the original sample set to form a complex graphic element sample set; Use a preset image annotation tool to annotate the preset basic graphic elements and non-disassemblable combined graphic elements in each sample diagram of the original sample set, and, annotate the complex graphic elements in each sample diagram of the complex graphic element sample set; Train a preset large CV vision model based on the annotated original sample set and complex graphic element sample set respectively to obtain a basic graphic element recognition model and a complex graphic element recognition model; Successively use the complex graphic element recognition model and the basic graphic element recognition model to recognize the target system diagram, and obtain the type of each graphic element in the target system diagram; Obtain the coordinates of each graphic element and the topological relationship between the graphic elements, and generate a configuration diagram in vector format.

[0008] Optionally, the constructing an original sample set based on the single cabinet diagrams disassembled from multiple primary system diagrams includes: Use the Cohen-Sutherland line clipping algorithm to disassemble each primary system diagram into multiple subsystem diagrams, and, respectively disassemble each subsystem diagram into multiple single cabinet diagrams; Use the sliding window combination algorithm to combine two or more adjacent single cabinet diagrams to form different combined cabinet diagrams; Perform rotation processing on each single cabinet diagram and combined cabinet diagram respectively to obtain a rotated diagram; Construct an original sample set based on all single cabinet diagrams, combined cabinet diagrams and rotated diagrams.

[0009] Optionally, before performing the step of using the Cohen-Sutherland line clipping algorithm to disassemble each primary system diagram into multiple subsystem diagrams, and, respectively disassembling each subsystem diagram into multiple single cabinet diagrams, the method further includes: Judge whether the primary system diagram is a hand-drawn diagram or a scanned diagram, If so, perform denoising processing on the primary system diagram and convert it into a grayscale diagram, and perform global histogram equalization processing on the grayscale diagram.

[0010] Optionally, the using a preset image annotation tool to annotate the preset basic graphic elements and non-disassemblable combined graphic elements in each sample diagram of the original sample set, and, annotating the complex graphic elements in each sample diagram of the complex graphic element sample set includes: For the original sample set, use the labelme image annotation tool to annotate the categories and bounding boxes of a variety of preset basic graphic elements and a variety of non - detachable combined graphic elements; For the complex graphic element sample set, use the labelme image annotation tool to annotate the categories and bounding boxes of the preset complex graphic elements; Judge whether there are graphic element categories with the number of samples less than the preset training volume in the original sample set after annotation, If so, add samples of the graphic element category in the original sample set until the number of samples of the graphic element category exceeds the preset training volume.

[0011] Optionally, the step of training the preset CV vision large model based on the annotated original sample set and complex graphic element sample set respectively to obtain a basic graphic element recognition model and a complex graphic element recognition model includes: Use the annotated original sample set to train the YOLOv12 model to obtain a basic graphic element recognition model for identifying basic graphic elements and non - detachable combined graphic elements in the system diagram; Use the annotated complex graphic element sample set to train the YOLOv12 model to obtain a complex graphic element recognition model for identifying complex graphic elements in the system diagram.

[0012] Optionally, the step of sequentially using the complex graphic element recognition model and the basic graphic element recognition model to identify the target system diagram to obtain the type of each graphic element in the target system diagram includes: Decompose the target system diagram to be processed into multiple subsystem diagrams; For each subsystem diagram, obtain the type of each graphic element by the following method: Based on the complex graphic element recognition model, judge whether the subsystem diagram contains complex graphic elements, If the subsystem diagram contains complex graphic elements, identify the type and bounding box of the complex graphic elements, and perform masking processing to cover the content in the bounding box on the subsystem diagram; After using the basic graphic element recognition model to identify the basic graphic elements and / or non - detachable combined graphic elements in the masked subsystem diagram, remove the mask on the subsystem diagram; If the subsystem diagram does not contain complex graphic elements, use the basic graphic element recognition model to identify the basic graphic elements and / or non - detachable combined graphic elements in the subsystem diagram.

[0013] Optionally, the step of sequentially using the complex graphic element recognition model and the basic graphic element recognition model to identify the target system diagram to obtain the type of each graphic element in the target system diagram further includes: For each subsystem diagram, it is determined whether there are non - detachable combined graphic elements. The non - detachable combined graphic elements include current transformer types, fuse types, disconnector types, and capacitor types. Each type of non - detachable combined graphic element includes multiple sub - type combined graphic elements. If so, a preset algorithm is used to determine the sub - type combined graphic elements corresponding to the non - detachable combined graphic elements.

[0014] Optionally, the step of using a preset algorithm to determine the sub - type combined graphic elements corresponding to the non - detachable combined graphic elements includes: For the non - detachable combined graphic elements of the current transformer type, the Hough circle transform algorithm is used to count the number of coils in the graphic element, and the sub - type combined graphic element to which it belongs is determined according to the number of coils. For the non - detachable combined graphic elements of the fuse type, the edge detection algorithm and the contour detection algorithm are used to count the number of quadrilaterals in the graphic element. The graphic element with the number of quadrilaterals equal to 4 is determined as a varistor, and the graphic element with the number of quadrilaterals equal to 2 is determined as a fuse. For the non - detachable combined graphic elements of the disconnector type, the edge detection algorithm and the contour detection algorithm are used to count the number of quadrilaterals in the graphic element. The graphic element with the number of quadrilaterals not equal to 0 is determined as a fuse - type disconnector, and the graphic element with the number of quadrilaterals equal to 0 is determined as a disconnector. For the non - detachable combined graphic elements of the capacitor type, the edge detection algorithm is used to count the number of capacitors in the graphic element. The graphic element with the number of capacitors greater than 1 is determined as a capacitor bank, and the graphic element with the number of capacitors equal to 1 is determined as a capacitor.

[0015] Optionally, after performing the step of sequentially using the complex graphic element recognition model and the basic graphic element recognition model to recognize the target system diagram and obtaining the type of each graphic element in the target system diagram, the method further includes: For each subsystem diagram, the following steps are performed: It is determined whether the recognition result contains all the sub - graphic elements that constitute any one of the preset detachable combined graphic elements. The detachable combined graphic element is composed of two connected sub - graphic elements, and the sub - graphic elements belong to the basic graphic elements. If so, the distance between the bounding boxes of any two of the sub - graphic elements is calculated. For two sub - graphic elements with a distance less than the preset distance threshold, the path algorithm is used to determine whether they are connected. If they are connected, the corresponding detachable combined graphic element is used to replace the two sub - graphic elements.

[0016] Optionally, the step of obtaining the coordinates of each graphic element and the topological relationship between graphic elements and generating a configuration diagram in vector format includes: For each subsystem diagram, the following steps are performed: Obtain the contour lines and electrical connections of all graphic elements in the subsystem diagram. Extract the connection relationships between different graphic elements based on the graphic element contour lines and electrical connections, and generate a path sequence array, where the path sequence array contains the categories and bounding box coordinates of the graphic elements in each path. Construct a configuration diagram in vector format based on the path sequence array.

[0017] Optionally, obtaining the contour lines and electrical connections of all graphic elements in the subsystem diagram includes: Convert the subsystem diagram into a grayscale image G1; Apply Gaussian filtering to G1 to remove noise points in the image, obtaining G2; Perform global histogram equalization on G2 to enhance the image contrast, and perform binarization processing and black-and-white inversion processing, obtaining G3; Use the Canny edge detection algorithm to detect the contour lines and connection lines of all graphic elements in G3, and use the morphological dilation method to connect the broken lines in the detection results, obtaining G4.

[0018] Optionally, extracting the connection relationships between different graphic elements based on the graphic element contour lines and electrical connections, and generating a path sequence array, includes: Perform masking processing on the bounding boxes of the graphic elements in G4 so that the mask graphics represent the corresponding graphic elements, obtaining G5; Obtain all paths in G5 based on the region connectivity function, where the paths are the paths connecting different graphic elements; Output the categories and bounding box coordinates of each graphic element in all paths, generating a path sequence array for the corresponding subsystem diagram.

[0019] Optionally, constructing a configuration diagram in vector format based on the path sequence array includes: Establish a standard symbol library based on the information data of all graphic elements; Represent the graphic elements in the path sequence array as points, and represent the connection relationships between the graphic elements as lines; Generate a configuration diagram in vector format based on the standard symbol library, and use the hierarchical layout algorithm to optimize the positions of the graphic elements and the arrangement of the connection lines.

[0020] Optionally, the method further includes: Adopt Figure 1 A consistency check algorithm to check whether the configuration diagram conforms to the electrical connection logic.

[0021] Another aspect of the present invention discloses an AI generation system for drawing to configuration based on a CV large model, which is applied to the AI generation method for drawing to configuration based on a CV large model provided in the foregoing aspect.

[0022] An AI generation method and system for drawing to configuration based on a CV large model disclosed in an embodiment of the present invention. First, an original sample set is constructed based on single cabinet diagrams disassembled from multiple primary system diagrams, and samples containing preset complex graphic elements are selected from the original sample set to form a complex graphic element sample set. Then, a preset image annotation tool is used to annotate the preset basic graphic elements and non-disassemblable combined graphic elements in each sample diagram of the original sample set, and to annotate the complex graphic elements in each sample diagram of the complex graphic element sample set. The preset CV vision large model is trained respectively based on the annotated original sample set and complex graphic element sample set to obtain a basic graphic element recognition model and a complex graphic element recognition model. Finally, the complex graphic element recognition model and the basic graphic element recognition model are used in sequence to recognize the target system diagram, obtain the type of each graphic element in the target system diagram, and obtain the coordinates of each graphic element and the topological relationship between the graphic elements, and generate a configuration diagram in vector format. Applying the method and system disclosed in this application in the complex circuit design scenario can significantly improve the accuracy and reliability of electrical drawing recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 FIG. is a schematic flowchart of an AI generation method for drawing to configuration based on a CV large model provided by an embodiment of the present invention; Figure 2 FIG. is a schematic flowchart for implementing step S100 provided by an embodiment of the present invention; Figure 1 in; Figure 3 FIG. is a schematic flowchart for implementing step S300 provided by an embodiment of the present invention; Figure 1 in; Figure 4 FIG. is a schematic flowchart for implementing step S400 provided by an embodiment of the present invention; Figure 1 in; Figure 5 FIG. is a schematic flowchart for implementing step S500 provided by an embodiment of the present invention; Figure 1 in; Figure 6 FIG. is a schematic flowchart of the processing process of a subsystem diagram provided by an embodiment of the present invention; Figure 7 FIG. is a schematic flowchart for implementing step S600 provided by an embodiment of the present invention; Figure 1 in; Figure 8 FIG. is a schematic flowchart for implementing step S601 provided by an embodiment of the present invention; Figure 7 in; Figure 9 FIG. is a schematic flowchart for implementing step S602 provided by an embodiment of the present invention; Figure 7 in; Figure 10A schematic flowchart of step S603 provided by an embodiment of the present invention Figure 7 for implementing the method. Specific implementation manners

[0024] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0025] Figure 1 A schematic flowchart of an AI generation method from drawings to configuration based on a CV large model provided by an embodiment of the present invention, which is used to generate a configuration vector diagram according to the primary system diagram of an electrical system. As Figure 1 shown, the method includes the following steps: Step S100: Construct an original sample set based on the single cabinet diagrams disassembled from multiple primary system diagrams.

[0026] In an embodiment disclosed by the present invention, as Figure 2 shown, step S100 can be implemented by the following sub-steps: Step S101: Use the Cohen-Sutherland line clipping algorithm to disassemble each primary system diagram into multiple subsystem diagrams, and respectively disassemble each subsystem diagram into multiple single cabinet diagrams.

[0027] Use the Cohen-Sutherland line clipping algorithm to disassemble the primary system diagram. For example, a certain primary system diagram can be disassembled into 6 subsystem diagrams by using the Cohen-Sutherland algorithm. The Cohen-Sutherland algorithm is a two-dimensional graphic clipping technology based on a rectangular viewport. By comparing each line segment in the diagram with a predefined window, it can quickly determine whether the line segment is completely inside the window, completely outside, or intersects the window boundary by using region coding. With the help of this algorithm, the wires, borders, connection relationships, etc. in a complex system diagram can be logically segmented, so as to identify sub-regions with relative closure and structural independence.

[0028] After disassembling the primary system diagram into subsystem diagrams, use the Cohen-Sutherland algorithm again to disassemble each subsystem diagram into different single cabinet diagrams. For example, a certain subsystem diagram can be disassembled into 6 single cabinet diagrams.

[0029] Since the source of the primary system diagram may be a scanned diagram or a hand-drawn diagram, in an embodiment disclosed by the present invention, before executing step S101, it is necessary to determine whether the primary system diagram is a hand-drawn diagram or a scanned diagram.

[0030] If the primary system diagram to be disassembled is a hand-drawn diagram or a scanned diagram, it needs to be preprocessed first and then structurally disassembled. The steps are as follows: First, Gaussian filtering is used to remove the noise points in the image. The calculation formula is as follows:

[0031] where represents the original image, * represents the convolution operation, represents the two-dimensional Gaussian kernel function with a standard deviation of and , where x and y are the relative coordinate values corresponding to the pixels, represents the denoised image.

[0032] Then, the obtained image is converted into a grayscale image, and the global histogram equalization processing is performed on the grayscale image using the cumulative distribution function CDF and the grayscale transformation function to increase the image brightness and contrast.

[0033] The CDF formula is: , and the grayscale transformation function:

[0034] where j is the gray level of the transformed image, j = 0, 1, 2, …, L - 1; L is the number of gray levels; , is the number of pixels with the original gray level k, n is the total number of original image pixels; int represents the rounding operation.

[0035] After completing the above preprocessing steps, the processed image is divided into several subsystem diagrams using the Cohen-Sutherland line clipping algorithm, and each subsystem diagram is disassembled into multiple single cabinet diagrams, and each single cabinet Figure 1 generally corresponds to an independent electrical cabinet or equipment area.

[0036] Step S102: Use the sliding window combination algorithm to combine two or more adjacent single cabinet diagrams to form different combined cabinet diagrams.

[0037] To expand the sample quantity and increase the combination diversity, the sliding window combination algorithm (i.e., the cannon shooting method) is used to combine two or more adjacent single cabinet diagrams to form combined cabinet diagrams. The combination strategy usually adopts the window sliding method with a step size of 1. For example, single cabinet Figure 1 and 2 form 12, Figures 1 to 3 form 123, and so on, generating various combination forms including 12, 123, 1234, 2345, 3456, etc. This combination method helps to improve the model's understanding ability of complex electrical structures in electrical graphic recognition training.

[0038] Step S103: Rotate each single cabinet diagram and combined cabinet diagram respectively to obtain the rotated diagrams.

[0039] Considering that the components may have different directions due to spatial layout and other reasons when drawing, all single cabinet drawings and combined cabinet drawings are rotated. Commonly used rotation angles include integer angles such as 90 degrees, 180 degrees, and 270 degrees. Small angle rotations such as ±10 degrees can also be added as needed. Through rotation transformation, the deviation of drawings under different shooting, scanning or design states can be effectively simulated, thereby improving the recognition effect of the model in real scenes.

[0040] Step S104: construct an original sample set based on all single cabinet images, combined cabinet images and rotation images.

[0041] All single cabinet images, combination cabinet images and their rotation images are summarized to construct the original sample set.

[0042] Step S200: Filter out samples containing preset complex primitives from the original sample set to form a complex primitive sample set.

[0043] Complex primitives are graphics with nested structures, complex connections, and high recognition difficulty, such as power supplies, etc. Samples containing one or more complex primitives are selected from the original sample set, and these samples are used to form a complex primitive sample set.

[0044] Step S300: using a preset image annotation tool to annotate preset basic primitives and inseparable combination primitives in each sample image of the original sample set, and annotate complex primitives in each sample image of the complex primitive sample set.

[0045] In one embodiment disclosed in the present invention, Figure 3 As shown, the following sub-steps are used to implement step S300: Step S301: for the original sample set, the labelme image annotation tool is used to annotate the categories and bounding boxes of a plurality of preset basic primitives and a plurality of inseparable combination primitives.

[0046] Basic elements are the most basic and smallest electrical symbols in electrical drawings, corresponding to electrical equipment or components with a single function. Such elements have clear standard graphic representations on drawings, such as circuit breakers, disconnectors, etc.

[0047] Inseparable composite graphics are composite graphics that are expressed as a whole in the drawing but contain multiple basic functional units. The internal structure of this type of composite graphics has been standardized or integrated into a fixed structure, and the logic cannot be further disassembled.

[0048] In a specific embodiment disclosed in the present invention, the non-detachable combination graphic elements at least include a current transformer class, a fuse class, a disconnector class and a capacitor class, wherein each class includes a plurality of sub-class combination graphic elements.

[0049] For example, the non-detachable combined graphic elements of the current transformer type include subclass combined graphic elements such as three-phase current transformers, four-phase current transformers, five-phase current transformers, etc. The number of coils included in the graphic elements of different subclasses is different, and the three-phase current transformer includes 3 coils.

[0050] The non-detachable combined graphic elements of the fuse type include subclass combined graphic elements such as varistors and fuses. The non-detachable combined graphic elements of the disconnect switch type include subclass combined graphic elements such as fuse disconnect switches and disconnect switches. The subclass combined graphic elements of the capacitor type are capacitors and capacitance.

[0051] In the original sample set, for each sample diagram, the open-source image annotation tool Labelme is used to annotate the preset basic graphic elements and non-detachable combined graphic elements. The user frames the boundary regions of each graphic element according to the drawing content and assigns the corresponding category label to it. The system will output the annotation information file of each graphic element, record its category and bounding box coordinates, and form standardized training data.

[0052] In this step, only the basic graphic elements and non-detachable combined graphic elements are annotated, and the recognition of the subclass combined graphic elements in the non-detachable combined graphic elements will be completed in the subsequent embodiments.

[0053] Step S302: For the complex graphic element sample set, use the labelme image annotation tool to annotate the category and bounding box of the preset complex graphic elements.

[0054] In the complex graphic element sample set, for each complex graphic element included in each sample diagram, the Labelme tool is also used for annotation to determine its physical boundary and complete classification and positioning. After the annotation is completed, a structured annotation information file is also generated.

[0055] Step S303: Determine whether there are graphic element categories in the original sample set with the number of samples less than the preset training volume.

[0056] All samples in the original sample set are counted to determine the number of samples of each graphic element category and compare it with the lower limit of the preset training data. The preset training volume is a threshold set according to the basic requirements of model training for the sample volume. If it is found that the number of occurrences of some graphic element categories in the sample set is small and may not be sufficient to support effective training, they will be determined as low-sample categories.

[0057] For example, record the number of each graphic element category in the annotated sample set as , set the target category number threshold and , where represents the minimum value of the mean of all graphic element category numbers in the annotated sample set, Represents the minimum value of the number of all graphic element categories in the labeled sample set. When the labeled quantity of a certain graphic element category does not exceed it is determined that the number of its samples is less than the preset training quantity.

[0058] If there is a graphic element category with the number of samples less than the preset training quantity, step S304 is executed; if there is no graphic element category with the number of samples less than the preset training quantity, step S400 is executed.

[0059] Step S304: Add samples of the graphic element category to the original sample set until the number of samples of the graphic element category exceeds the preset training quantity.

[0060] To address the problem of sample imbalance, data augmentation and sample supplementation operations will be performed on the categories with insufficient graphic element quantities. The methods adopted can be: screening and adding more graphic element samples containing this category in the original drawing set, or generating new training samples by performing image enhancement (such as rotation, mirroring, noise addition) on the existing images.

[0061] In this way, gradually increase the number of samples of this graphic element category to above the preset training quantity threshold to ensure category coverage and recognition accuracy during the model training stage.

[0062] When the average value of the labeled quantities of all component categories exceeds and the labeled quantity of each component category exceeds step S400 is executed.

[0063] Step S400: Train the preset CV vision large model based on the labeled original sample set and the complex graphic element sample set respectively to obtain a basic graphic element recognition model and a complex graphic element recognition model.

[0064] Train the preset computer vision CV large model to obtain deep learning models suitable for the recognition of different graphic element types respectively.

[0065] In an embodiment disclosed in the present invention, as Figure 4 shown, the following sub-steps can be adopted to implement step S400: Step S401: Train the YOLOv12 model with the labeled original sample set to obtain a basic graphic element recognition model.

[0066] The YOLOv12 model is trained using the original labeled sample set to obtain a basic primitive recognition model. The original sample set includes a large number of image samples labeled with basic primitives and non-detachable combined primitives. During the training process, the YOLOv12 model gradually improves its recognition ability for these primitives by learning features such as the bounding boxes, class labels, and position distributions of the primitives. The trained model has the ability to automatically detect and label basic primitives and non-detachable combined primitives in the full image or sub-images, and is applicable to subsequent electrical diagram structure analysis and automated recognition processes.

[0067] Step S402: The YOLOv12 model is trained using the labeled complex primitive sample set to obtain a complex primitive recognition model.

[0068] The YOLOv12 model is separately trained using the labeled complex primitive sample set to obtain a complex primitive recognition model. The images in the complex primitive sample set mainly contain complex electrical components with relatively complex structures and diverse graphical expression forms. By separately training on these complex primitives, the model can more accurately extract the key features in complex graphics, improving the recognition accuracy and robustness. After training, this model will be used to locate and identify various complex primitives in the system diagram to improve the comprehensiveness and accuracy of the overall system recognition.

[0069] During the model training process, the already labeled original sample set and complex primitive sample set are respectively divided into training set, validation set, and test set according to a certain proportion. Among them, the training set is used for the parameter learning of the model and is the main data source for repeated iterations during the model training process, accounting for 80% of the total data volume. The validation set is used to evaluate the performance of the model on the data not involved in training in real time during the training process, for adjusting hyperparameters, preventing overfitting, etc., accounting for 15% of the total data volume. The test set does not participate in the training at all and is only used to objectively evaluate the final effect of the model after the training is completed, accounting for 5% of the total data volume.

[0070] To ensure that the model can recognize all types of primitives, when dividing the data set, it is also necessary to ensure that there are sufficient sample numbers for each primitive category in the three subsets to avoid some categories being "ignored" or misjudged by the model due to insufficient samples. This process can be achieved through methods such as stratified sampling to ensure that each type of sample has a representative distribution in the training, validation, and test sets.

[0071] After completing the data division, the YOLOv12 object detection model is used to train the training set. The model iteratively learns the feature information of the primitives and uses the validation set for intermediate effect evaluation and parameter adjustment. Finally, the model effect is verified on the test set to evaluate the performance of the model in the actual drawing automatic recognition scenario, such as key indicators such as recognition accuracy, recall rate, and positioning error.

[0072] The accuracy threshold can be set in advance When the average precision of the model at an IoU threshold of 0.5 exceeds it is determined that the model has completed training.

[0073] Step S500: Identify the target system diagram using the complex primitive recognition model and the basic primitive recognition model in sequence, and obtain the type of each primitive in the target system diagram.

[0074] In this step, the trained complex primitive recognition model and the basic primitive recognition model are called in sequence to achieve hierarchical recognition of multiple types of primitives.

[0075] In an embodiment disclosed by the present invention, as Figure 5 shown, the following sub-steps can be used to implement step S500: Step S501: Decompose the target system diagram to be processed into multiple subsystem diagrams.

[0076] Before performing primitive recognition, it is first necessary to perform structural decomposition processing on the target system diagram. The Cohen-Sutherland line clipping algorithm can be used to perform spatial clipping on the line network structure in the system diagram to obtain multiple subsystem diagrams.

[0077] For each subsystem diagram, the following method is used to obtain the type of each primitive: Step S502: Based on the complex primitive recognition model, determine whether the subsystem diagram contains complex primitives.

[0078] After each subsystem diagram is extracted, the trained complex primitive recognition model is first called to analyze it to determine whether it contains preset complex primitives.

[0079] If the subsystem diagram contains complex primitives, execute step S503.

[0080] Step S503: Identify the type and bounding box of the complex primitive, and perform masking processing to cover the content in the bounding box on the subsystem diagram.

[0081] Once the complex primitive recognition model detects a complex primitive, its type and precise position (bounding box) will be recorded. Subsequently, image masking processing is performed on the area covered by the bounding box on the original subsystem diagram. The masking method is to fill the area with a uniform color (such as black) to prevent its content from interfering with the judgment of the subsequent basic primitive recognition model.

[0082] Step S504: After using the basic primitive recognition model to identify the basic primitives and / or non-detachable combined primitives in the masked subsystem diagram, remove the mask on the subsystem diagram.

[0083] Apply the basic primitive recognition model to the masked image. The model will detect all objects in the remaining area that conform to the characteristics of the basic primitive template, and output the type and position of each primitive. After all basic primitives are recognized, restore the content of the previously covered area to restore all primitive recognition results in the entire subsystem diagram, including complex primitives and basic primitives.

[0084] If the subsystem diagram does not contain complex primitives, execute step S505.

[0085] Step S505: Use the basic primitive recognition model to recognize basic primitives and / or non-separable combined primitives in the subsystem diagram.

[0086] If it is determined in S502 that no complex primitives are detected in the subsystem diagram, directly skip the covering and restoring process, and directly call the basic primitive recognition model to recognize the subsystem diagram. In this case, there is no interference area in the image, and the recognition model can directly locate and recognize all basic primitives and non-separable combined primitives in the complete image. The recognition results of the basic primitives will be output in the form of class labels and bounding box coordinates.

[0087] In an embodiment disclosed in the present invention, after using the basic primitive recognition mode to recognize the subsystem diagram and obtaining the recognition results, for the recognition results of each subsystem diagram, it is determined whether there are non-separable combined primitives.

[0088] Non-separable combined primitives include current transformer types, fuse types, disconnector types, and capacitor types. Each type of non-separable combined primitive includes multiple sub-class combined primitives. For specific classifications, refer to the descriptions in the foregoing embodiments.

[0089] If there are non-separable combined primitives in the subsystem diagram, use a preset algorithm to determine the sub-class combined primitives corresponding to the non-separable combined primitives.

[0090] In an embodiment disclosed in the present invention, the following method can be used to determine the sub-class combined primitives to which the non-separable combined primitives belong: (1) For non-separable combined primitives of the current transformer type, use the Hough circle transform algorithm to count the number of coils in the primitive, and determine the corresponding sub-class combined primitive according to the number of coils. Among them, if the number of coils is 3, determine the primitive as a three-phase current transformer; if the number of coils is 4, determine the primitive as a four-phase current transformer, and so on.

[0091] (2) For non-separable combined primitives of the fuse type, use the edge detection algorithm and the contour detection algorithm to count the number of quadrilaterals in the primitive. Determine the primitive with 4 quadrilaterals as a varistor, and determine the primitive with 2 quadrilaterals as a fuse.

[0092] (3) For non-detachable combined graphic elements of disconnect switches, use edge detection algorithms and contour detection algorithms to count the number of quadrilaterals in the graphic elements. Determine the graphic elements with a non-zero number of quadrilaterals as fuse disconnect switches, and determine the graphic elements with a zero number of quadrilaterals as disconnect switches.

[0093] (4) For non-detachable combined graphic elements of capacitors, use edge detection algorithms to count the number of capacitors in the graphic elements. Determine the graphic elements with a capacitor number greater than 1 as capacitors, and determine the graphic elements with a capacitor number equal to 1 as a single capacitor.

[0094] In an embodiment disclosed by the present invention, after completing step S500, for each subsystem diagram, as Figure 6 shown, the following steps need to be executed: Step S061: Determine whether the recognition result contains all the sub-graphic elements that constitute any one of the preset detachable combined graphic elements.

[0095] Among them, the detachable combined graphic element is composed of two connected sub-graphic elements, and the sub-graphic element belongs to the basic graphic element.

[0096] Search whether there are sub-graphic elements that meet a certain preset combination rule in the subsystem diagram. For example, a detachable combined graphic element may be composed of a "circuit breaker" graphic element and a "current transformer" graphic element. If these two basic graphic elements exist simultaneously in the system detection result in this diagram, it is considered that there is a potential detachable combined graphic element.

[0097] If the recognition result contains all the sub-graphic elements that constitute any one of the preset detachable combined graphic elements, then execute step S062; if the recognition result does not contain all the sub-graphic elements that constitute any one of the preset detachable combined graphic elements, then execute step S600.

[0098] Step S062: Calculate the distance between the bounding boxes of any two sub-graphic elements.

[0099] When two sub-graphic elements that may form a detachable combined graphic element are found, analyze their spatial proximity relationship, and perform geometric measurement on the bounding boxes of the two graphic elements. For example, Manhattan distance or Euclidean distance can be used to measure their closeness, so as to determine whether they have the possibility of combination.

[0100] Step S063: Use a path algorithm to determine whether two sub-graphic elements with a distance less than the preset distance threshold are connected.

[0101] If the distance between the bounding boxes is less than a pre-set combined distance threshold (e.g., 30 pixels), they will not be immediately merged but further verified through a path connectivity algorithm. For example, a region connectivity algorithm (such as D8 adjacency relationship) or an image path detection algorithm (such as pixel connection tracing) can be used to determine whether there is an electrical connection or path between two primitive elements in the image. Only when there is indeed a connection line or a morphological connected region between the two in the image can it be determined that they form a valid connection relationship.

[0102] If they are connected to each other, execute step S064: Replace the two sub-primitive elements with corresponding detachable combined primitive elements.

[0103] After confirming that the two sub-primitive elements meet the triple criteria of "combination rule + distance threshold + connectivity", they are merged into an overall detachable combined primitive element. At the data level, the original records of the two basic primitive elements will be replaced with new records of the detachable combined primitive element, including information such as the coordinates of the merged bounding box and the primitive element type label.

[0104] Step S600: Obtain the coordinates of each primitive element and the topological relationship between the primitive elements, and generate a configuration diagram in vector format.

[0105] In an embodiment disclosed by the present invention, for each subsystem diagram, as Figure 7 shown, the following steps are all executed: Step S601: Obtain the contour lines and electrical connections of all primitive elements in the subsystem diagram.

[0106] This step is used for the precise extraction of image edges and connections, ensuring that all electrical components and their connection relationships can be converted into structured image feature data through edge recognition.

[0107] In an embodiment disclosed by the present invention, as Figure 8 shown, the following sub-steps can be adopted to implement step S601: Step S6011: Convert the subsystem diagram into a grayscale image G1.

[0108] To reduce the processing complexity, first convert the color subsystem diagram into a grayscale image G1, so that the image processing only needs to focus on the pixel brightness information, removing color interference and providing a unified input format for subsequent operations such as filtering and histogram equalization.

[0109] The conversion formula is as follows:

[0110] where R, G, and B respectively represent the numerical values of the red, green, and blue primary color channels of the image, and L represents the grayscale value of the converted grayscale image.

[0111] Step S6012: Apply Gaussian filtering to G1 to remove the noise points in the image, obtaining G2.

[0112] Use a Gaussian kernel to smooth the image, removing the noise points in the original image and those introduced by the disassembly of the original image. The resulting image is denoted as G2.

[0113] Step S6013: Perform global histogram equalization on G2 to enhance the image contrast, and then perform binarization and black-and-white inversion processing to obtain G3.

[0114] First, enhance the overall brightness distribution contrast of the image through global histogram equalization to make the primitive contours more obvious. Then, perform binarization to segment the image into foreground (lines) and background (blank). Finally, perform black-and-white inversion so that the electrical lines are white and the background is black, thus better meeting the input requirements of the edge detection algorithm in the subsequent steps.

[0115] Step S6014: Use the Canny edge detection algorithm to detect the contour lines and connection lines of all primitives in G3, and use the morphological dilation method to connect the broken lines in the detection results, obtaining G4.

[0116] Use the Canny algorithm to perform edge detection on G3 to detect the significant edges and connection lines in the image. To repair the edge breaks caused by image quality or complex structures, further apply morphological dilation to the detection results. Use a structural element (such as a 3×3 square kernel) to expand and connect the edge lines, generating the G4 image with more complete edge connectivity.

[0117] Step S602: Extract the connection relationships between different primitives based on the primitive contour lines and electrical connections, and generate a path sequence array.

[0118] In this step, through image structure analysis, obtain the physical connection relationships between the primitives in the system diagram, that is, the connection topology information, and form a structured array for generating the configuration diagram. Among them, the path sequence array contains the categories and bounding box coordinates of the primitives in each path.

[0119] In an embodiment disclosed in the present invention, as Figure 9 shown, the following sub-steps can be used to implement Step S602: Step S6021: Perform masking processing on the bounding boxes of the primitives in G4 so that the masking graphics represent the corresponding primitives, obtaining G5.

[0120] Since many primitives themselves are not necessarily closed, to facilitate subsequent analysis, distinguish the primitives from the connection lines as nodes in the image, and perform masking on each identified primitive bounding box, that is, fill a specific gray value (such as white) in the primitive bounding box area.

[0121] Step S6022: Obtain all the paths in G5 based on the region connectivity function.

[0122] Among them, a path is a route connecting different primitive elements.

[0123] Use the region connectivity function (e.g., D8 distance) to determine whether there is a path extending from the bounding box region of one primitive element to that of another. Each path is regarded as an actual electrical connection line segment in the system. Obtain the number of all connected regions in image G5 and the label of the connected region to which each pixel belongs.

[0124] The D8 distance calculation formula is as follows:

[0125] Among them, (x, y) are the coordinates of pixel p, and (u, v) are the coordinates of pixel q. If and only if is less than or equal to 2, it is considered that point q and point p are adjacent. If there is a sequence of pixel points between point p and point q such that any two points are adjacent, it is considered that there is a path between point p and point q.

[0126] Step S6023: Output the category and bounding box coordinates of each primitive element in all the paths, and generate a path sequence array for the corresponding subsystem diagram.

[0127] After identifying all the connected paths, each path contains all the component categories and component bounding box coordinates on that path. Record them in the form of a structured array, and use this array as the path sequence structure of the subsystem diagram.

[0128] Step S603: Construct a configuration diagram in vector format according to the path sequence array.

[0129] In an embodiment disclosed in the present invention, as Figure 10 shown, the following sub-steps can be used to implement Step S603: Step S6031: Establish a standard symbol library according to the information data of all primitive elements.

[0130] Convert all component information into the internal data model of the configuration tool, and call the DOM operation of SVG to generate a standard component symbol library. Each template contains a standard shape (such as SVG path, vector geometric figure), etc., to ensure that the primitive element styles are consistent and standardized during drawing.

[0131] Step S6032: Represent the primitive elements in the path sequence array as points, and represent the connection relationship between the primitive elements as lines.

[0132] In a vector coordinate system, the center of each primitive bounding box in the path sequence array is regarded as a node, and the connection lines with other primitives are represented as lines. By traversing the path sequence, a node-edge topology graph of the electrical network structure is constructed, providing a topological basis for subsequent drawing and layout.

[0133] Step S6033: Generate a configuration diagram in vector format based on the standard symbol library, and optimize the positions of primitives and the arrangement of connection lines using a hierarchical layout algorithm.

[0134] According to the topology graph structure and the standard symbol library information, primitive symbols are placed into the coordinate system in sequence, and the connection lines between them are drawn. To improve the clarity and logic of the image, a hierarchical layout algorithm is introduced to adjust the order of primitive positions according to the signal flow or current direction and arrange the connection lines with minimal crossings. The final output format can be the configuration diagram data in a vector graphics file (such as SVG).

[0135] In an embodiment disclosed by the present invention, a Figure 1 consistency check algorithm is used to verify whether the configuration diagram conforms to the electrical connection logic.

[0136] After generating the vector format configuration diagram (SVG file), the SVG file can be synchronously uploaded to the processing page of the configuration tool on the cloud platform. This page is a visualization environment that supports interactive graphic editing, and users can perform operations such as dragging, connecting lines, and configuring primitive attributes in it to further modify the graphic structure or improve the primitive logic.

[0137] After uploading the SVG file, a Figure 1 consistency check algorithm is called to analyze the configuration diagram. This algorithm mainly judges the connection relationships between various primitives in the diagram one by one according to the connection rules in electrical engineering to verify whether they conform to logical correctness. For example, it judges issues such as whether the power output is connected to the load end, whether the signal is closed-loop, whether the connection line is broken, and whether the primitive is isolated. At the same time, this algorithm also supports pattern matching of standard electrical connection modes and analysis of the integrity of the connection path.

[0138] If the check algorithm determines that there are inconsistencies or errors in the connection logic of the current graphic, the problem area will be displayed in a highlighted or prompt manner. At this time, the user can use functions such as dragging, connecting lines, deleting, or the property panel in the graphic interaction interface of the configuration tool to manually modify and adjust the error part until it passes the Figure 1 consistency check.

[0139] After the graphic logic is corrected, the user can bind the SVG diagram to a specific engineering project. One is to store the attribute information of each primitive in the component database, including its device type, parameters, port numbers, numbers, etc.; the other is to configure and integrate the SVG primitives with the project monitoring interface as part of the runtime monitoring interface.

[0140] Another embodiment of the present invention discloses an AI generation system for converting drawings to configurations based on a CV large model, which is applied to the method for converting drawings to configurations based on a CV large model disclosed in the foregoing embodiment.

[0141] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also regarded as the protection scope of the present invention.

Claims

1. An AI generation method from drawings to configuration based on a large CV model, applied to the primary system diagram of an electrical system, characterized in that Including: Constructing an original sample set based on single cabinet diagrams disassembled from multiple primary system diagrams, where each sample is a single cabinet diagram or a combined cabinet diagram formed by combining multiple single cabinet diagrams; Screening out samples containing preset complex graphic elements from the original sample set to form a complex graphic element sample set; Using a preset image annotation tool to annotate the preset basic graphic elements and non-disassemblable combined graphic elements in each sample diagram of the original sample set, and, annotating the complex graphic elements in each sample diagram of the complex graphic element sample set; Respectively training a preset CV vision large model based on the annotated original sample set and complex graphic element sample set to obtain a basic graphic element recognition model and a complex graphic element recognition model; Sequentially using the complex graphic element recognition model and the basic graphic element recognition model to recognize the target system diagram to obtain the type of each graphic element in the target system diagram; Obtaining the coordinates of each graphic element and the topological relationship between graphic elements, and generating a configuration diagram in vector format.

2. The generation method according to claim 1, wherein The constructing an original sample set based on single cabinet diagrams disassembled from multiple primary system diagrams includes: Using the Cohen-Sutherland line clipping algorithm to disassemble each primary system diagram into multiple subsystem diagrams, and, respectively disassembling each subsystem diagram into multiple single cabinet diagrams; Using a sliding window combination algorithm to combine two or more adjacent single cabinet diagrams to form different combined cabinet diagrams; Performing rotation processing on each single cabinet diagram and combined cabinet diagram respectively to obtain a rotated diagram; Constructing an original sample set based on all single cabinet diagrams, combined cabinet diagrams and rotated diagrams.

3. The generation method according to claim 2, characterized in that Before performing the step of using the Cohen-Sutherland line clipping algorithm to disassemble each primary system diagram into multiple subsystem diagrams, and, respectively disassembling each subsystem diagram into multiple single cabinet diagrams, the method further includes: Judging whether the primary system diagram is a hand-drawn diagram or a scanned diagram, If so, performing denoising processing on the primary system diagram and converting it into a grayscale diagram, and performing global histogram equalization processing on the grayscale diagram.

4. The generation method according to claim 1, wherein The using a preset image annotation tool to annotate the preset basic graphic elements and non-disassemblable combined graphic elements in each sample diagram of the original sample set, and, annotating the complex graphic elements in each sample diagram of the complex graphic element sample set includes: For the original sample set, using the labelme image annotation tool to annotate the categories and bounding boxes of a preset variety of basic graphic elements and a variety of non-disassemblable combined graphic elements; For the complex graphic element sample set, using the labelme image annotation tool to annotate the categories and bounding boxes of the preset complex graphic elements; Judging whether there are graphic element categories in the original sample set with the number of samples less than the preset training amount after annotation, If so, adding samples of the graphic element category to the original sample set until the number of samples of the graphic element category exceeds the preset training amount.

5. The generation method according to claim 1, characterized in that The respectively training a preset CV vision large model based on the annotated original sample set and complex graphic element sample set to obtain a basic graphic element recognition model and a complex graphic element recognition model includes: Using the annotated original sample set to train the YOLOv12 model to obtain a basic graphic element recognition model for recognizing basic graphic elements and non-disassemblable combined graphic elements in the system diagram; Train the YOLOv12 model using the labeled complex primitive sample set to obtain a complex primitive recognition model for identifying complex primitives in the system diagram.

6. The generation method according to claim 1, wherein The method of sequentially using the complex primitive recognition model and the basic primitive recognition model to identify the target system diagram to obtain the type of each primitive in the target system diagram includes: Decompose the target system diagram to be processed into multiple subsystem diagrams; For each subsystem diagram, obtain the type of each primitive in the following manner: Based on the complex primitive recognition model, determine whether the subsystem diagram contains complex primitives. If the subsystem diagram contains complex primitives, identify the type and bounding box of the complex primitives, and perform masking processing to cover the content within the bounding box on the subsystem diagram. After using the basic primitive recognition model to identify the basic primitives and / or non-disassemblable combined primitives in the masked subsystem diagram, remove the mask on the subsystem diagram. If the subsystem diagram does not contain complex primitives, use the basic primitive recognition model to identify the basic primitives and / or non-disassemblable combined primitives in the subsystem diagram.

7. The generation method according to claim 6, wherein The method of sequentially using the complex primitive recognition model and the basic primitive recognition model to identify the target system diagram to obtain the type of each primitive in the target system diagram further includes: For each subsystem diagram, determine whether there are non-disassemblable combined primitives. The non-disassemblable combined primitives include current transformer types, fuse types, disconnector types, and capacitor types, and each type of non-disassemblable combined primitive includes multiple sub-type combined primitives. If so, use a preset algorithm to determine the sub-type combined primitive corresponding to the non-disassemblable combined primitive.

8. The generation method according to claim 7, wherein The step of using a preset algorithm to determine the sub-type combined primitive corresponding to the non-disassemblable combined primitive includes: For the non-disassemblable combined primitive of the current transformer type, use the Hough circle transform algorithm to count the number of coils in the primitive, and determine the corresponding sub-type combined primitive according to the number of coils. For the non-disassemblable combined primitive of the fuse type, use the edge detection algorithm and the contour detection algorithm to count the number of quadrilaterals in the primitive. Determine the varistor for the primitive with the number of quadrilaterals being 4, and determine the fuse for the primitive with the number of quadrilaterals being 2. For the non-disassemblable combined primitive of the disconnector type, use the edge detection algorithm and the contour detection algorithm to count the number of quadrilaterals in the primitive. Determine the fuse-disconnector for the primitive with the number of quadrilaterals not being 0, and determine the disconnector for the primitive with the number of quadrilaterals being 0. For the non-disassemblable combined primitive of the capacitor type, use the edge detection algorithm to count the number of capacitors in the primitive. Determine the capacitor bank for the primitive with the number of capacitors being greater than 1, and determine the capacitor for the primitive with the number of capacitors being equal to 1.

9. The generation method according to claim 8, wherein After performing the step of sequentially using the complex primitive recognition model and the basic primitive recognition model to identify the target system diagram to obtain the type of each primitive in the target system diagram, the method further includes: For each subsystem diagram, perform the following steps: Determine whether the recognition result contains all the sub-primitives that form any one of the preset disassemblable combined primitives. The disassemblable combined primitive is composed of two connected sub-primitives, and the sub-primitives belong to the basic primitives. If so, calculate the distance between the bounding boxes of any two of the sub-primitives. For two sub-graphic elements with a distance less than a preset distance threshold, use a path algorithm to determine whether they are connected. If they are connected, replace the two sub-graphic elements with corresponding detachable combined graphic elements.

10. The generation method according to any one of claims 6 to 9, characterized in that, The obtaining of the coordinates of each graphic element and the topological relationship between graphic elements, and generating a configuration diagram in vector format includes: For each subsystem diagram, the following steps are executed: Obtain the contour lines and electrical connections of all graphic elements in the subsystem diagram; Extract the connection relationships between different graphic elements according to the graphic element contour lines and electrical connections, and generate a path sequence array, where the path sequence array contains the category and bounding box coordinates of the graphic elements in each path; Construct a configuration diagram in vector format according to the path sequence array.

11. The generation method according to claim 10, characterized in that, The obtaining of the contour lines and electrical connections of all graphic elements in the subsystem diagram includes: Convert the subsystem diagram into a grayscale image G1; Use Gaussian filtering on G1 to remove noise points in the image, obtaining G2; Perform global histogram equalization processing on G2 to enhance the image contrast, and perform binarization processing and black-and-white inversion processing, obtaining G3; Use the Canny edge detection algorithm to detect the contour lines and connection lines of all graphic elements in G3, and use the morphological dilation method to connect the broken lines in the detection results, obtaining G4.

12. The generation method according to claim 11, wherein The extracting of the connection relationships between different graphic elements according to the graphic element contour lines and electrical connections, and generating a path sequence array includes: Perform masking processing on the bounding boxes of the graphic elements in G4 so that the masking graphics represent the corresponding graphic elements, obtaining G5; Obtain all paths in G5 based on the region connectivity function, where the paths are the paths connecting different graphic elements; Output the category and bounding box coordinates of each graphic element in all paths, generating a path sequence array for the corresponding subsystem diagram.

13. The generation method according to claim 12, characterized in that, The constructing of a configuration diagram in vector format according to the path sequence array includes: Establish a standard symbol library according to the information data of all graphic elements; Represent the graphic elements in the path sequence array as points and represent the connection relationships between graphic elements as lines; Generate a configuration diagram in vector format based on the standard symbol library, and use a hierarchical layout algorithm to optimize the positions of graphic elements and the layout of connection lines.

14. The generation method according to claim 1, wherein The method further includes: Use a graph consistency check algorithm to check whether the configuration diagram conforms to the electrical connection logic.

15. An AI generation system from drawings to configuration based on a large CV model, characterized in that, Applied to the AI generation method from drawing to configuration based on the CV large model according to any one of claims 1 to 14.

Images