A neoclassical architecture style identification system, construction method and identification method

Abstract

The application belongs to the technical field of building style identification, and discloses a neoclassical building style identification system, a construction method and an identification method. The method comprises the following steps: S1, constructing a whole style image classification model; S2, constructing an imitation method target detection model; S3, constructing a key component target detection model and a repetitive component target detection model; and S4, connecting the whole style image classification model, the imitation method target detection model, the key component target detection model, the repetitive component target detection model and a building component image classification model in sequence, wherein the imitation method target detection model, the key component target detection model and the repetitive component target detection model are connected in parallel, thereby forming the required building style identification system. Through the application, the problem that the existing style identification system cannot form a complete neoclassical building style identification description from the whole to the local according to a building facade picture is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of building style recognition, and more particularly to a neoclassical building style recognition system, a construction method and a recognition method. BACKGROUND

[0002] The visual recognition of building style is initially based on traditional computer vision technology, and the features of the image are extracted based on traditional image feature extraction methods, such as SIFT features and HOG features, and then the building image is classified based on traditional classification algorithms such as SVM algorithm. The traditional method has a relatively large limitation for the task, and needs to design features separately for a specific task, has poor scalability, and the actual use effect is not very good. With the development of deep learning-based computer vision technology, the visual recognition field has changed from feature engineering-based recognition to model architecture design. Based on a good model architecture, we no longer need to manually design features, but use the back propagation algorithm of the neural network to let the model automatically learn good features based on the training data. In recent years, representative convolutional neural network architectures include VGGNet, Inceptions, ResNe(X)t, DenseNet, MobileNet, and EfficientNet.

[0003] In recent years, there have been researches on building style recognition based on traditional vision or deep learning, but these methods only study the classification of overall style or the detection of local key elements, and do not form a complete recognition system for the overall to local complete architectural style description of the image building style. Simple image classification-based style recognition of historical buildings no longer meets our style description needs, and the architectural patterns and internal relationships of components contained in the building are beneficial to deepen our understanding of the style of historical buildings, so that historical buildings can be better protected. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides a neoclassical building style recognition system, a construction method and a recognition method, which solves the problem that the existing style recognition system cannot form a complete neoclassical building style recognition description from the overall to the local based on the building elevation picture.

[0005] To achieve the above-mentioned purpose, according to one aspect of the present application, a construction method of a neoclassical building style recognition system is provided, which comprises the following steps:

[0006] S1 collects building picture form pictures, identifies the overall style of each picture and forms a data set, constructs an overall style recognition model, trains the overall style image classification model using the data set, and obtains the required overall style image classification model;

[0007] S2 configures a new data set from the pictures in the data set, identifies the imitation method, key component and repetitive component of each picture in the new data set, forms an imitation method, key component and repetitive component target detection data set, and then extracts the key component and repetitive component in each picture, identifies and classifies the key component and repetitive component to form a building component image classification data set;

[0008] S3 respectively uses the imitation method, key component and repetitive component target detection data set and the building component image classification data set of step S2 for deep learning, to obtain an imitation method target detection model, a key component target detection model, a repetitive component target detection model and a building component image classification model;

[0009] S4 connects the overall style recognition model, the imitation method target detection model, the key component target detection model and the repetitive component target detection model, and the building component classification model in sequence, wherein the imitation method target detection model, the key component target detection model and the repetitive component recognition model are connected in parallel, to form the required building style recognition system.

[0010] Further preferably, in step S1, the overall style includes the following styles: ancient Greece, ancient Rome, Romanesque, Gothic, Renaissance, Baroque, Classicism and Neoclassicism.

[0011] Further preferably, in step S2, the imitation method includes Greek temple style, Roman Pantheon style, Tumbuado style and St. Peter's Cathedral style.

[0012] Further preferably, in step S3, the key component target detection model is used to identify the category of key components and their corresponding coordinates, and the category of key components includes domes, towers, floral patterns, rose windows, scrolls, colonnades and arcade corridors.

[0013] Further preferably, in step S3, the repetitive component target detection model is used to identify the category of repetitive components and their corresponding coordinates, and the repetitive components include arches, single columns, double columns and three columns.

[0014] Further preferably, the building component classification model is used to further classify the key components and repetitive components, and the categories include polygonal dome, circular dome, Gothic spire, non-spire, broken mountain flower, semicircular mountain flower, triangular mountain flower, Gothic pointed arch, Roman arch, Doric column, Ionic column, Corinthian column, Tashkent column and mixed column.

[0015] According to another aspect of the present application, there is provided a recognition system formed by the construction method described above.

[0016] According to yet another aspect of the present application, there is provided a method for recognizing the Neoclassical architecture style recognition system described above, which comprises the following steps:

[0017] S1 inputting a picture to be recognized into the system, and the overall style recognition model recognizing whether the picture to be recognized is Neoclassical, if yes, entering step S2, otherwise outputting the overall style type;

[0018] S2 the imitation method target detection model recognizing the imitation method of the picture to be recognized, and obtaining the building composition imitation method information of the input picture;

[0019] S3 the key component target detection model detecting the key component of the picture to be recognized, and obtaining the category and coordinate information of the key component, if the information of dome, tower and mountain flower exists in the detection result, obtaining the component picture by intercepting the original picture according to the coordinate information of the dome, tower and mountain flower components, inputting the component picture into the building component classification model, and obtaining the subcategory information to which the dome, tower and mountain flower belong, if not, obtaining the information of "no key component for fine classification";

[0020] S4 the repetitive component target detection model detecting the repetitive component in the picture to be recognized, and obtaining the category and coordinate information of the repetitive component, if the information of arch and single column exists in the detection, obtaining the component picture by intercepting the original picture according to the coordinate information of the arch and single column components, inputting the component picture into the building component classification model, and outputting the subcategory information to which the arch and single column belong, if not, outputting the information of "no repetitive component for fine classification".

[0021] Overall, the above technical solutions conceived by the present application have the following beneficial effects compared with the prior art:

[0022] 1. The present application is a new classical architecture style system based on image classification and target detection algorithm in deep learning, which inputs the historical building facade picture into the system, and the system uses the trained algorithm model to identify the overall style, imitation method and key elements of the building in turn, and finally generates the style description of the new classical architecture from the detail to the whole according to the model result;

[0023] 2. The recognition system provided by the present application gradually identifies the architectural style elements of the picture from the whole to the part, has high recognition accuracy, does not need to design special features for special tasks, has strong expansibility, good actual use effect and high recognition efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a structural schematic diagram of the recognition system constructed according to the preferred embodiment of the present application;

[0025] Figure 2 is a building style description schematic diagram constructed according to the preferred embodiment of the present application;

[0026] Figure 3 is a flowchart of the recognition method constructed according to the preferred embodiment of the present application;

[0027] Figure 4 is a flowchart of the recognition system construction method constructed according to the preferred embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0029] As shown in Figure 2 , we divide the new classical architecture style description into three aspects from the perspective of architecture, which are overall style description, building composition imitation method analysis and building element level description. The overall style recognition includes the recognition of eight overall styles such as ancient Greek, ancient Rome, Romanesque, Gothic, Renaissance, Baroque, classicism and neoclassicism. The building composition imitation method mainly includes Greek temple style, Roman Pantheon style, Tumbull style and St. Peter's Cathedral style. For the component level description of the building, we divide the building components into two categories, one is the key component which has important influence on the architectural style, and the other is the repetitive component which is widely used in the building. The key building components mainly include dome, tower, mountain flower, rose window, vortex, colonnade and arcade, and the repetitive building components mainly include arch, single column, double column and three columns.

[0030] Archivolt is the arc-shaped part of the building such as bridge, door and window; single column is the single column in the building, including the column used alone in the building and each column in the combined column; three columns combined together are called three columns, and two columns combined together are called double columns.

[0031] For archivolt component, according to the sharpness of the upper arc-shaped part, there are two sub-categories of Gothic archivolt and Roman archivolt; for dome component, according to the shape of the drum base, there are two sub-categories of polygonal dome and circular dome; for tower component, according to the sharpness, there are two sub-categories of Gothic tower and non-sharp tower; for tower component, according to the shape, there are three sub-categories of broken mountain flower, triangular mountain flower and semicircular mountain flower; for single column, according to the different column head, there are five sub-categories of Doric, Ionic, Tashkent, Corinthian and mixed.

[0032] As shown in Figure 1 and 3 , in order to complete the whole neoclassical style recognition task, we designed the system shown in Figure 1 . In this system, we first input the facade picture of the building we want to identify, after image preprocessing, the image is transmitted into the overall style image classification model, if the identification result shows that the building is not a neoclassical building, the picture building corresponding to the building style is directly output, and the identification process is ended; if it shows that the building is a neoclassical building, the image is sequentially transmitted into the imitation method target detection model, the key component target detection model and the repetitive component target detection model, respectively to obtain the category and coordinate information of the imitation method, the key component and the repetitive component. The style information contained in the dome, tower, mountain flower and arch is very rich, so we not only identify these components, but also continue to classify them in more detail to dig more detailed style information. Therefore, after the key component target detection model identifies the dome, tower and mountain flower, the system cuts out the three components according to the component coordinate information output by the key component target detection model and inputs them into the building component classification model to obtain the subclass information of the detected dome, tower and mountain flower components. After the repetitive component target detection model identifies the arch and single column, the system cuts out the two components according to the component coordinate information output by the repetitive component target detection model and inputs them into the building component classification model to obtain the subclass information of the detected arch and single column components.

[0033] As shown in Figure 4 , the neoclassical style recognition system mentioned in the application is realized by the following technical scheme:

[0034] Step 1: Collect building image data.

[0035] Collecting building images from existing public datasets or crawling the internet using a crawler technique, and then filtering the collected image data to remove low-quality images.

[0036] Step 2: Labeling data sets for each model based on the established system architecture

[0037] The entire system image recognition technology includes image classification and target detection. For image classification tasks, we use the image data collected in the first step to create an image classification task dataset based on the specific task classification categories, i.e., placing different categories of pictures in each category folder.

[0038] For target detection tasks, we can use open-source image labeling tools such as labelme to label images to create different task target detection datasets.

[0039] Step 3: Train the model using image classification or target detection algorithms based on the task type

[0040] For image classification and target detection tasks, we can use transfer learning to train the model, which involves first pre-training the model on a large image dataset, then fine-tuning the pre-trained model on the specific task dataset, and finally obtaining a model that performs well on the specific task dataset.

[0041] Step 4: Integrate the trained model to form Figure 1 the system shown.

[0042] The above method further includes the following specific steps:

[0043] S1 Collect various building facade pictures, create an image classification dataset based on the building style of each picture, use the dataset and deep learning technology to build and train an overall style image classification model, and obtain the required overall style image classification model;

[0044] S2 Collect building facade pictures containing building composition imitation methods, label the building composition methods in each building facade picture to create an imitation method target detection dataset, use the dataset and deep learning technology to build and train an imitation method target detection model, and obtain the required imitation method target detection model;

[0045] S3 Collect building facade pictures containing key building components, label the key building components in each building facade picture to create a key component target detection dataset, use the dataset and deep learning technology to build and train a key component target detection model, and obtain the required key building component target detection model;

[0046] S4 collects building facade pictures containing building repetitive components, labels the repetitive building components in each building facade picture, makes a repetitive component target detection dataset, uses deep learning technology to build and train a repetitive component target detection model, and obtains the required repetitive component target detection model;

[0047] S5 collects building key components and repetitive component pictures, makes a building component image classification dataset according to the sub-class of the building component in each component picture, uses the dataset and deep learning technology to build and train a building component image classification model, and obtains the required building component image classification model;

[0048] S6 connects the overall style image classification model, the imitation method target detection model, the key component target detection model, the repetitive component target detection model, and the building component image classification model in sequence, wherein the imitation method target detection model, the key component target detection model, and the repetitive component target detection model are connected in parallel, thereby forming the required building style recognition system.

[0049] Further, the building component categories include polygonal domes, circular domes, Gothic spire towers, non-spire towers, broken mountain flowers, semicircular mountain flowers, triangular mountain flowers, Gothic spandrels, Roman spandrels, Doric columns, Ionic columns, Corinthian columns, Tashkent columns, and mixed columns.

[0050] A method for recognizing using the above system, comprising the following steps:

[0051] S1 inputs the picture to be recognized into the system, and the overall style recognition model recognizes whether the picture to be recognized is Neoclassicism. If yes, it simultaneously enters steps S2, S3, and S4, otherwise, the overall style type is output;

[0052] S2 the imitation method target detection model recognizes the imitation method of the picture to be recognized, and obtains the building composition imitation method information of the input picture;

[0053] S3 the key component target detection model detects the key component of the picture to be recognized, and obtains the category and coordinate information of the key component. If the detection result contains information of dome, tower, and mountain flower, the component picture is obtained by intercepting the original picture according to the coordinate information of the dome, tower, and mountain flower components, and is input into the building component classification model to obtain the sub-class information of the dome, tower, and mountain flower. If not, the information "no key component for detailed classification" is obtained;

[0054] The repetitive component target detection model in S4 detects the repetitive component in the picture to be recognized to obtain the category and coordinate information of the repetitive component. If the information of the coupon and the single column exists in the detection result, the component picture is obtained by intercepting the original picture according to the coordinate information of the coupon and the single column, and is input into the building component classification model to obtain the sub-category information to which the coupon and the single column belong. If the information does not exist, the information of “no repetitive component for fine classification” is obtained.

[0055] S5 outputs the recognition information obtained in S2, S3 and S4.

[0056] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a neoclassical architectural style identification system, characterized in that, The method includes the following steps: S1 collects architectural images, identifies the overall style of each image and labels it to form a dataset, constructs an overall style recognition model, uses the dataset to train the overall style image classification model, and obtains the required overall style image classification model. S2 constructs a new dataset from the images in the dataset, identifies the imitation techniques, key components, and repetitive components of each image in the new dataset, and forms an object detection dataset for imitation techniques, key components, and repetitive components. Then, the key components and repetitive components in each image are extracted, and the key components and repetitive components are identified and classified to form an image classification dataset for building components. S3 utilizes the imitation method, key component, and repetitive component target detection datasets and building component image classification datasets obtained in step S2 to perform deep learning, thereby obtaining the imitation method target detection model, key component target detection model, repetitive component target detection model, and building component image classification model. S4 connects the overall style recognition model, the imitation technique target detection model, the key component target detection model, the repetitive component target detection model, and the building component classification model in series. The imitation technique target detection model, the key component target detection model, and the repetitive component recognition model are connected in parallel to form the required building style recognition system.

2. The method for constructing a neoclassical architectural style identification system as described in claim 1, characterized in that, In step S1, the overall style includes the following styles: Ancient Greek, Ancient Roman, Romanesque, Gothic, Renaissance, Baroque, Classicism, and Neoclassicism.

3. A method for constructing a neoclassical architectural style identification system as described in claim 1 or 2, characterized in that, In step S2, the imitation techniques include the Greek temple style, the Roman Pantheon style, the Tempietto style, and the St. Peter's Basilica style.

4. The method for constructing a neoclassical architectural style identification system as described in claim 1, characterized in that, In step S3, the key component target detection model is used to identify the category of key components and their corresponding coordinates. The categories of key components include domes, towers, pediments, rose windows, scrolls, colonnades, and arcades.

5. The method for constructing a neoclassical architectural style identification system as described in claim 1, characterized in that, In step S3, the repetitive component target detection model is used to identify the category of repetitive components and their corresponding coordinates. The repetitive components include vaults, single columns, double columns, and triple columns.

6. The method for constructing a neoclassical architectural style identification system as described in claim 1, characterized in that, In step S3, the building component classification model is used to further classify key components and repetitive components. The categories include polygonal domes, circular domes, Gothic spires, non-spires, broken pediments, semi-circular pediments, triangular pediments, Gothic pointed arches, Romanesque arches, Doric columns, Ionic columns, Corinthian columns, Tashkent columns, and mixed columns.

7. An identification system constructed by the construction method according to any one of claims 1-6.

8. A method for identification using the identification system of claim 7, characterized in that, The method includes the following steps: S1 Input the image to be identified into the system. The overall style recognition model identifies whether the image to be identified is neoclassical. If it is, proceed to step S2; otherwise, output the overall style type. The target detection model described in S2 identifies the imitation techniques of the image to be identified, and obtains the architectural composition imitation technique information of the input image; The key component target detection model described in S3 detects key components in the image to be identified, obtains the category and coordinate information of the key components, and if the detection results contain information about domes, towers and pediments, then the component images are extracted from the original image based on the coordinate information of the domes, towers and pediments, and input into the building component classification model to obtain the subclass information to which the domes, towers and pediments belong; If it does not exist, the message "No key components for sub-classification" is obtained; The repetitive component target detection model described in S4 detects repetitive components in the image to be identified, and obtains the category and coordinate information of the repetitive components. If there is information about vaults or single columns in the detection, the component image is extracted from the original image according to the coordinate information of the vault or single column component, and then input into the building component classification model to output the subclass information to which the vault or single column belongs. If it does not exist, output "No repeating components for subcategories".

Images