Automatic identification and labeling method for fire alarm system drawings

Abstract

The invention discloses an automatic identification and labeling method for a drawing of a fire alarm system, relates to the technical field of fire monitoring, and solves the technical problem that the prior art is difficult to deal with the identification difficulty caused by large difference between professional elements and symbols in the drawing and changeable positions and quantity of point location elements. Comprising the following steps: acquiring a fire alarm system drawing, and preprocessing the input fire alarm system drawing; obtaining a preset drawing type library, classifying the preprocessed drawings according to the drawing type library, and performing format conversion on the drawings according to the types of the drawings to obtain recognition drawings; calling a point location element recognition model, and inputting the recognition drawing into the point location element recognition model to obtain point location elements of the recognition drawing; automatically generating annotation information according to the point location elements and detecting the annotation information; according to the method, the recognition efficiency and accuracy of the drawing of the fire alarm system can be greatly improved, and errors and omission of manual operation are reduced.

Description

Technical Field

[0001] The invention belongs to the field of fire monitoring, and in particular is an automatic identification and marking method for drawings of a fire alarm system. Background Art

[0002] With the rapid development of the fire alarm industry, fire alarm drawings contain the structural layout information of the building, as well as the installation location and quantity of various fire alarm equipment. The recognition and annotation of drawings have become an indispensable part of this field. Accurate drawing recognition and annotation are the basis for ensuring the normal operation of the system. Traditional drawing recognition methods often rely on manual operation. Personnel engaged in drawing recognition need to undergo certain professional training in order to be familiar with the graphic representation of various types of fire alarm equipment and the meaning of relevant signs of building structures. However, even experienced professionals need to spend a lot of time and energy when faced with complex and large-scale fire alarm drawings, which is inefficient and prone to errors. The automatic recognition and annotation method of drawings can save time and human resources and improve the accuracy of recognition and annotation.

[0003] In the prior art, drawing recognition usually relies on image processing and pattern recognition technology. However, fire alarm system drawings contain a large number of professional elements and symbols, which may vary in shape, size and color, posing a great challenge to recognition. In addition, the location and number of point elements (such as detectors, alarms, etc.) in the drawings may also vary from project to project, further increasing the difficulty of recognition. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an automatic identification and marking method for fire alarm system drawings.

[0005] To achieve the above object, the present invention provides a method for automatically identifying and marking drawings of a fire alarm system, comprising:

[0006] Step 1: drawing preprocessing: obtaining a fire alarm system drawing and preprocessing the input fire alarm system drawing; the preprocessing includes: denoising, binarization and edge detection;

[0007] Step 2: Drawing type determination and format conversion: obtain a preset drawing type library, classify the pre-processed drawings according to the drawing type library, and convert the drawings into recognized drawings according to the drawing type;

[0008] Step 3: point element recognition; call the point element recognition model, input the recognition drawing into the point element recognition model, and obtain the point elements of the recognition drawing; the point element recognition model is constructed based on deep learning algorithm training; the point elements include firefighting legends such as detectors, alarm buttons, and fire alarm controllers;

[0009] Step 4: Generate annotation information; automatically generate annotation information according to point elements and detect the annotation information;

[0010] Step 5: Output the annotation information; display the generated annotation information to the user in a visual form.

[0011] Preferably, the preprocessing of the input fire alarm system drawings includes:

[0012] The drawings of the fire alarm system were adjusted and the salt and pepper noise in the drawings were removed by using median filtering. The Otsu method was used to perform binarization processing to obtain the key information in the drawings. The Canny edge detection algorithm was used to extract the edge contour of the drawings.

[0013] The present invention can make the image information in the fire alarm system drawings clearer and more accurate by preprocessing the drawings, which helps to improve the accuracy of subsequent recognition and marking.

[0014] Preferably, converting the format of the drawing according to the type of the drawing includes:

[0015] Retrieve the drawing type; convert drawings in DWG, SVG, and WMF formats to BMP format; the technology used for converting DWG to BMP is LibreDWG, and the technology used for converting SVG and WMF to BMP is Apache Batik.

[0016] The present invention converts drawings of different formats into a unified format, which facilitates the subsequent processing of the drawings using the same algorithm and processing flow, and is beneficial to improving the versatility and applicability of the method.

[0017] Preferably, the point element recognition model is constructed based on deep learning algorithm training, including:

[0018] Build a neural network model based on the PyTorch deep learning framework, build the C3Ghost module in the neural network model, and add the Transformer block and CBAM attention module to the feature fusion layer;

[0019] Obtain the identification drawings to be annotated, and use the K-means clustering method of the genetic algorithm to annotate the firefighting legends on the identification drawings; integrate the identification drawings to be annotated and the corresponding identification drawings annotated with firefighting legends into a standard data set;

[0020] The neural network model is trained using a standard data set; the trained neural network model is marked as a point element recognition model.

[0021] The present invention adopts a deep learning model to construct a point element recognition model, which can explore the model's feature extraction potential, help the model find attention areas in scenes with dense objects, and further improve the accuracy of target detection; and by training the model with a standard data set, the accuracy of model recognition can be improved.

[0022] Preferably, the automatically generating annotation information according to the point elements and detecting the annotation information includes:

[0023] Retrieve point elements, automatically generate corresponding annotation information at the point elements, and manually detect the annotation information; wherein the annotation information includes the type, position and quantity of the point elements;

[0024] Determine whether there are any errors in the annotation information. If yes, manually correct the annotation information and add the corrected recognition image to the data set; if no, continue to detect the annotation information.

[0025] The present invention automatically generates annotation information according to the identified point elements, and automatically generates annotation information according to the identified point elements, which can provide a basis for subsequently displaying the annotation information to the user.

[0026] Preferably, the generated annotation information is displayed to the user in a visual form, and an interaction method with the user is designed, including: the user can view, edit and export operations as needed.

[0027] The present invention presents the generated annotation information to the user in a visual form, and satisfies the different needs of the user through this interactive mode. The user can further process and apply the annotation information according to actual conditions, thereby improving the practicality and flexibility of the method.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The present invention can improve the recognition efficiency and accuracy of fire alarm system drawings and reduce errors and omissions in manual operations. It can be flexibly customized and optimized according to user needs to meet the actual needs of different projects. At the same time, it can also be integrated with other fire management systems to achieve information sharing and collaborative work, thereby improving the efficiency and level of the entire fire management.

[0030] (2) The present invention pre-processes the fire alarm system drawings to make the images clearer and more accurate, thereby improving the accuracy and efficiency of identification and annotation; by judging the drawing type and converting the format, the versatility and applicability of the method are improved; a deep learning model and a variety of advanced technologies are used to accurately identify point elements, thereby improving the model operation efficiency and positioning and annotation accuracy; the automatic generation of annotation information improves the annotation efficiency, and manual review and correction can optimize the algorithm; the interactive mode of annotation information output meets the different needs of users, thereby improving practicality and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 is a flow chart of the present invention;

[0033] Figure 2 A schematic diagram of the identification and annotation elements of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] See also Figure 1-Figure 2 The first embodiment of the present invention provides a method for automatically identifying and marking drawings of a fire alarm system, comprising:

[0036] Step 1: Drawing preprocessing; obtain the fire alarm system drawings and preprocess the input fire alarm system drawings; the preprocessing includes: denoising, binarization and edge detection; adjust the fire alarm system drawings and use median filtering to remove the salt and pepper noise in the drawings; use the Otsu method to highlight the drawings and perform binarization processing to obtain the key information in the drawings; use the Canny edge detection algorithm to extract the edge contour of the drawings.

[0037] Step 2: Determine the drawing type and convert the format; obtain the preset drawing type library, classify the pre-processed drawings according to the drawing type library, convert the format of the drawings according to the drawing type to obtain the identified drawings; retrieve the drawing type; convert the drawings in DWG, SVG, and WMF formats to BMP format; the technology used for converting DWG to BMP is LibreDWG, and the technology used for converting SVG and WMF to BMP is Apache Batik.

[0038] For example, suppose there are three pre-processed fire alarm system drawings in DWG, SVG, and WMF formats. Convert the DWG format to BMP format using LibreDWG, convert the SVG format to BMP format using Apache Batik, and convert the WMF format to BMP format using Apache Batik.

[0039] Step 3: Point element recognition; call the point element recognition model, input the recognition drawing into the point element recognition model, and obtain the point elements of the recognition drawing; the point element recognition model is constructed based on deep learning algorithm training; the point elements include firefighting legends such as detectors, alarm buttons, and fire alarm controllers; the point element recognition model is constructed based on deep learning algorithm training, including:

[0040] Build a neural network model based on the PyTorch deep learning framework, build the C3Ghost module in the neural network model, and add the Transformer block and CBAM attention module to the feature fusion layer;

[0041] Obtain the identification drawings to be annotated, and use the K-means clustering method of the genetic algorithm to annotate the firefighting legends on the identification drawings; integrate the identification drawings to be annotated and the corresponding identification drawings annotated with firefighting legends into a standard data set;

[0042] The neural network model is trained using a standard data set; the trained neural network model is marked as a point element recognition model.

[0043] Step 4: Generate annotation information; automatically generate annotation information based on point elements and detect the annotation information; including:

[0044] Retrieve point elements, automatically generate corresponding annotation information at the point elements, and manually detect the annotation information; wherein the annotation information includes the type, position and quantity of the point elements;

[0045] Determine whether there are any errors in the annotation information. If yes, manually correct the annotation information and add the corrected recognition image to the data set; if no, continue to detect the annotation information.

[0046] For example, in a fire alarm system drawing of a large shopping mall, after the previous steps, when it comes to the stage of generating annotation information, the point element recognition model recognizes multiple point elements.

[0047] First, the point element of the fire alarm controller was identified, and the annotation information was automatically generated according to its location coordinates (located at the center of the ceiling of the east corridor on the third floor of the mall, with coordinate values ​​of X: 15.6 meters, Y: 22.3 meters): "Type: Fire Alarm Controller; Location: Center of the Ceiling of the East Corridor on the Third Floor of the Mall; Quantity: 1". Subsequently, professionals tested this annotation information. After carefully checking the actual layout plan of the mall and installation records and other materials, it was found that the X coordinate value in the location annotation should be 15.8 meters. At this time, it was determined that there was an error in the annotation information, and the X coordinate value in the location annotation was manually corrected to 15.8 meters. Then, the corrected recognition image containing the annotation information of this smoke alarm was added to the data set for more accurate model training and optimization in the future. Reference Figure 2 The rectangle in the middle drawing represents the fire alarm controller, and the dotted box in the drawing after recognition represents the point element detected by the point element recognition model.

[0048] Secondly, when the point element recognition model recognizes a point element of a detector (located next to the entrance of the safety passage on the west side of the first floor of the mall, with coordinates of X: 8.2 meters, Y: 3.5 meters), it automatically generates annotation information as "Type: detector; Location: next to the entrance of the safety passage on the west side of the first floor of the mall; Quantity: 1". After testing by professionals, it is confirmed that the annotation information is accurate, and the point element recognition model continues to perform cyclic testing on the annotation information of other point elements according to the set time interval or new instructions to ensure the accuracy of all annotation information. Figure 2 The circle in the middle drawing represents the fire alarm controller, and the dotted box in the drawing after recognition represents the point element detected by the point element recognition model.

[0049] Third, there is a point element of the alarm button in the fire alarm system drawing, but the point element recognition model did not recognize it. After professionals checked the actual layout plan of the shopping mall and installation records and other materials, they added the annotation information of the alarm button to the recognition image and added the image to the data set for subsequent model training and optimization. Figure 2 The square in the middle drawing represents the alarm button, and the half-dot square in the drawing after recognition indicates that the point element recognition model has not detected the point element.

[0050] Step 5: Output the annotation information;

[0051] Display the generated annotation information to the user in a visual form. Design the interaction method with the user, including: the user can view, edit and export as needed.

[0052] The working principle of the present invention is as follows: obtain the fire alarm system drawings, and pre-process the input fire alarm system drawings. Obtain a preset drawing type library, classify the pre-processed drawings according to the drawing type library, and convert drawings of different formats into BMP format using corresponding technologies. Retrieve the point element recognition model, input the identification drawings into the point element recognition model, obtain the point elements of the identification drawings, and generate annotation information; automatically generate annotation information according to the point elements and detect the annotation information, and display the generated annotation information to the user in a visual form.

[0053] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for automatically identifying and marking drawings of a fire alarm system, characterized in that: The following steps are involved: Step 1: Drawing preprocessing; Obtaining a fire alarm system drawing, and preprocessing the input fire alarm system drawing; the preprocessing includes: denoising, binarization and edge detection; Step 2: Drawing type determination and format conversion: obtain a preset drawing type library, classify the pre-processed drawings according to the drawing type library, and convert the drawings into recognized drawings according to the drawing type; Step 3: point element identification; call the point element identification model, input the identification drawing into the point element identification model, and obtain the point elements of the identification drawing; wherein the point element identification model is constructed based on deep learning algorithm training; the point elements include detectors, alarm buttons, and fire alarm controllers; Step 4: Generate annotation information; automatically generate annotation information according to point elements and detect the annotation information; Step 5: Output the annotation information; display the generated annotation information to the user in a visual form.

2. The method for automatically identifying and marking drawings of a fire alarm system according to claim 1, characterized in that: The preprocessing of the input fire alarm system drawings includes: The drawings of the fire alarm system were adjusted and the salt and pepper noise in the drawings were removed by using median filtering. The Otsu method was used to perform binarization processing to obtain the key information in the drawings. The Canny edge detection algorithm was used to extract the edge contour of the drawings.

3. The method for automatically identifying and marking drawings of a fire alarm system according to claim 1, characterized in that: The format conversion of drawings according to the type of drawings includes: Retrieve the drawing type; convert drawings in DWG, SVG, and WMF formats to BMP format; the technology used for converting DWG to BMP is LibreDWG, and the technology used for converting SVG and WMF to BMP is Apache Batik.

4. The method for automatically identifying and marking drawings of a fire alarm system according to claim 1, characterized in that: The point element recognition model is constructed based on deep learning algorithm training, including: Build a neural network model based on the PyTorch deep learning framework, build the C3Ghost module in the neural network model, and add the Transformer block and CBAM attention module to the feature fusion layer; Obtain the identification drawings to be annotated, and use the K-means clustering method of the genetic algorithm to annotate the firefighting legends on the identification drawings; integrate the identification drawings to be annotated and the corresponding identification drawings annotated with firefighting legends into a standard data set; The neural network model is trained using a standard data set; the trained neural network model is marked as a point element recognition model.

5. The method for automatically identifying and marking drawings of a fire alarm system according to claim 1, characterized in that: The method of automatically generating annotation information according to point elements and detecting the annotation information includes: Retrieve point elements, automatically generate corresponding annotation information at the point elements, and manually detect the annotation information; wherein the annotation information includes the type, position and quantity of the point elements; Determine whether there are any errors in the annotation information. If yes, manually correct the annotation information and add the corrected recognition image to the data set; if no, continue to detect the annotation information.

6. The method for automatically identifying and marking drawings of a fire alarm system according to claim 1, characterized in that: The generated annotation information is displayed to the user in a visual form, and an interaction method with the user is designed, including: the user can view, edit and export the information as needed.

Images