A method and device for delivering and accepting decoration materials, and a storage medium

Abstract

The application provides a decoration material delivery and acceptance method, equipment and storage medium, the method comprises the steps of: training a style transfer network and a classifier; receiving a plurality of frames of original image data actually collected for decoration materials of a target building; collecting a plurality of frames of model image data for the same building materials as the plurality of frames of original image data in a target building model; inputting the original image data into the style transfer network to convert the original image data into target image data in the modeling style of a building information modeling platform; inputting the target image data and the model image data into the classifier to calculate the probability that the content of the target image data is the same as that of the model image data; calculating the coverage rate of the model image data in the building model; and generating a delivery and acceptance result for the decoration materials of the target building according to the probability and the coverage rate. The embodiment can automatically perform preliminary delivery and acceptance, reduce the subjective influence of auditors, unify the scale of delivery and acceptance, and improve the efficiency of delivery and acceptance of decoration materials.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application belongs to the technical field of computer vision, and particularly relates to a delivery and acceptance method of decoration materials, equipment and a storage medium. BACKGROUND

[0002] When a construction unit completes decoration of a building, multiple frames of image data of the building are shot, the multiple frames of image data are uploaded to the cloud, and a delivery and acceptance unit performs acceptance of decoration materials. An auditing personnel of the delivery and acceptance unit mainly opens design drawings of the building, and determines whether the construction unit has constructed according to requirements by comparing contents in the image data.

[0003] This delivery and acceptance mode mainly depends on the auditing personnel, and there is a certain requirement for experience of the auditing personnel, so that there is a certain threshold for the delivery and acceptance, and moreover, there is a certain subjectivity of the auditing personnel in the acceptance, and a scale of the delivery and acceptance fluctuates, so that efficiency of the delivery and acceptance of the decoration materials is relatively low. SUMMARY

[0004] Therefore, the embodiment of the present application provides a delivery and acceptance method of decoration materials, equipment and a storage medium, so as to improve the efficiency of the delivery and acceptance of the decoration materials.

[0005] A first aspect of the embodiment of the present application provides a delivery and acceptance method of decoration materials, applied to a building information model platform, the building information model platform storing a target building model modeling a target building; the method comprises:

[0006] training a style transfer network and a classifier; the style transfer network is used for converting image data into a modeling style of the building information model platform; the classifier is used for identifying whether image data contents in the same modeling style of the building information model platform are same;

[0007] when performing a delivery and acceptance operation on decoration materials of the target building, receiving multiple frames of original image data actually collected from the decoration materials of the target building;

[0008] collecting, in the target building model, multiple frames of model image data of the same building materials as the multiple frames of original image data;

[0009] inputting the original image data into the style transfer network to convert the original image data into target image data in the modeling style of the building information model platform;

[0010] if the target image data and the model image data describe the same building materials, inputting the target image data and the model image data into the classifier to calculate a probability that the target image data and the model image data have the same content;

[0011] calculate coverage of the model image data in the building model;

[0012] generate a delivery acceptance result of the decoration material of the target building according to the probability and the coverage.

[0013] A second aspect of the embodiment of the application provides a decoration material delivery acceptance device, applied to a building information model platform, the building information model platform storing a target building model modeling a target building; the device comprises:

[0014] a model training module, configured to train a style transfer network and a classifier; the style transfer network is configured to convert image data into a modeling style of the building information model platform; the classifier is configured to identify whether image data contents in the same modeling style of the building information model platform are the same;

[0015] an image data receiving module, configured to receive a plurality of frames of original image data actually collected from the decoration material of the target building when performing a delivery acceptance operation on the decoration material of the target building;

[0016] an image data collecting module, configured to collect, in the target building model, a plurality of frames of model image data describing the same building material as the plurality of frames of original image data;

[0017] an image data transferring module, configured to input the original image data into the style transfer network to convert the original image data into target image data in the modeling style of the building information model platform;

[0018] a probability calculating module, configured to, if the target image data describes the same building material as the model image data, input the target image data and the model image data into the classifier to calculate a probability that the target image data and the model image data have the same content;

[0019] a coverage calculating module, configured to calculate coverage of the model image data in the building model;

[0020] a delivery acceptance module, configured to generate a delivery acceptance result of the decoration material of the target building according to the probability and the coverage.

[0021] A third aspect of the embodiment of the application provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the decoration material delivery acceptance method of the first aspect.

[0022] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the decoration material delivery and acceptance method in the first aspect.

[0023] The fifth aspect of the embodiment of the present application provides a computer program product, which, when running on a computer, causes the computer to execute the decoration material delivery and acceptance method in the first aspect.

[0024] In the embodiment, the style transfer network and the classifier are trained; the style transfer network is used to convert the image data into the modeling style of the building information modeling platform; the classifier is used to identify whether the image data contents in the same modeling style of the building information modeling platform are the same; when performing the delivery and acceptance operation on the decoration materials of the target building, the multiple frames of original image data actually collected from the decoration materials of the target building are received; the multiple frames of model image data describing the same building materials as the multiple frames of original image data are collected in the target building model; the original image data is input into the style transfer network to be converted into the target image data in the modeling style of the building information modeling platform; if the target image data and the model image data describe the same building materials, the target image data and the model image data are input into the classifier to calculate the probability that the target image data and the model image data have the same content; the coverage of the model image data in the building model is calculated; and the delivery and acceptance result of the decoration materials of the target building is generated according to the probability and the coverage. The embodiment provides a standardized process for the preliminary delivery and acceptance of the decoration of the building, reduces the experience requirement for the auditors, reduces the threshold of the delivery and acceptance, and the automatic preliminary delivery and acceptance can reduce the subjective influence of the auditors, unify the scale of the delivery and acceptance, and thus improve the efficiency of the delivery and acceptance of the decoration materials. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0026] Figure 1 is a schematic diagram of a decoration material delivery and acceptance method provided by the embodiment of the present application;

[0027] Figure 2 is a schematic diagram of a style transfer network and a classifier provided by the embodiment of the present application;

[0028] Figure 3is a schematic view of a decoration material delivery and acceptance device provided by an embodiment of the present application.

[0029] Figure 4 is a schematic view of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0030] In the following description, for the purpose of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0031] The technical solutions of the present application are described below through specific embodiments.

[0032] Referring to Figure 1 , a schematic view of a decoration material delivery and acceptance method provided by an embodiment of the present application is shown, which is applied to a BIM (Building Information Management, building information model) platform, as shown in Figure 1 , which can specifically include the following steps:

[0033] Step 101, training a style transfer network and a classifier.

[0034] In an offline environment, a style transfer network and a classifier can be constructed based on deep learning, and appropriate samples are collected to train the style transfer network and the classifier separately or jointly, wherein the style transfer network is used to convert image data into a modeling style of a building information model platform, and the classifier is used to identify whether the image data content under the same modeling style of the building information model platform is the same.

[0035] In an embodiment of the present application, step 101 can include the following steps:

[0036] Step 1011, collecting first sample image data and second sample image data.

[0037] In the present embodiment, the first sample image data and the second sample image data can be collected as samples to jointly train the style transfer network and the classifier.

[0038] Wherein, the building information model platform stores a sample building model for three-dimensional modeling of a sample building, the first sample image data is image data collected from the sample building model under certain vision, and the second sample image data is image data actually collected for decoration materials of the sample building.

[0039] Step 1012, load the first generative adversarial network, the second generative adversarial network, the style transfer network and the classifier.

[0040] In the embodiment, the first generative adversarial network and the second generative adversarial network can be loaded, wherein the first generative adversarial network and the second generative adversarial network are both generative adversarial networks (GAN). The GAN is composed of a generator and a discriminator. The generator receives a random noise as input and generates a simulated sample. The goal of the generator is to deceive the discriminator so that it cannot distinguish between the generated sample and the real sample. The discriminator receives a sample (which can be a real sample or a generated sample) as input and outputs the probability that the sample is a real sample. The goal of the discriminator is to accurately distinguish between real samples and generated samples.

[0041] During the training process, the generator and the discriminator are in mutual opposition: the generator continuously improves the quality of the generated sample in order to deceive the discriminator; the discriminator continuously improves its discrimination ability in order to better distinguish between real samples and generated samples. This process eventually reaches a balance, and the generator can generate high-quality samples, and the judgment ability of the discriminator reaches the best.

[0042] In the embodiment, the first generative adversarial network has a first generator and a first discriminator, and the second generative adversarial network has a second generator and a second discriminator; the first generator, the second generator and the style transfer network have the same structure, and the second discriminator and the classifier have the same structure.

[0043] The first generator, the second generator and the style transfer network can all be CNN (Convolutional Neural Networks), such as SANet (Style Attention Network), etc., and the first discriminator can be a binary classification model (especially a classification head). The second discriminator and the classifier can both be binary classification models with certain feature extraction capabilities.

[0044] Step 1013, under the condition of maintaining the first generator unchanged, training the first discriminator according to the first sample image data and the second sample image data, so that the first discriminator is used to classify whether the image data is the modeling style of the building information modeling platform.

[0045] In the embodiment, the second generative adversarial network can be ignored, and the first generative adversarial network can be trained separately. The process of training the first generative adversarial network separately can be divided into two alternating training stages.

[0046] The first training stage is to perform supervised training on the first discriminator with the first sample image data and the second sample image data as samples under the condition that the first generator remains unchanged, so that the first discriminator has the ability to classify whether the image data is in the modeling style of the building information modeling platform.

[0047] In a specific implementation, the second sample image data can be input into the first generator, and the first generator converts the second sample image data into third sample image data in the modeling style of the building information modeling platform.

[0048] The first sample pair is constructed, which includes the first sample image data and a first label representing true (Ture), and the third sample image data and a second label representing false (False).

[0049] The first sample image data and the third sample image data in the first sample pair are respectively input into the first discriminator using a random or the like to generate first classification information; wherein the first classification information indicates whether it is in the modeling style of the building information modeling platform.

[0050] Under the condition that the parameters of the first generator remain unchanged, the parameters of the first discriminator are updated according to the loss value of the first classification information, the first label and the second label on the binary classification, at this time, the first discriminator will be punished for misclassifying the true sample as a false sample or misclassifying the false sample as a true sample, and the first discriminator updates its parameters through back propagation of the loss value.

[0051] Step 1014, if the first discriminator is trained, then under the condition that the first discriminator remains unchanged, the first generator is trained according to the second sample image data, so that the first generator is used to convert image data into the modeling style of the building information modeling platform.

[0052] If the first training stage (i.e., the first discriminator is trained) is completed, the second training stage is entered, and the first generator is supervised trained with the second sample image data as samples under the condition that the first discriminator remains unchanged, so that the first generator has the ability to convert image data into the modeling style of the building information modeling platform.

[0053] In a specific implementation, the second sample image data can be input into the first generator, and the first generator converts the second sample image data into fourth sample image data in the modeling style of the building information modeling platform.

[0054] The fourth sample image data is labeled with a third label (Ture) representing true.

[0055] The fourth sample image data is input into the first discriminator to generate second classification information; wherein the second classification information indicates whether it is in the modeling style of the building information modeling platform.

[0056] Under the condition of maintaining the parameters of the first discriminator unchanged, the parameters of the first generator are updated according to the loss value in classification between the second classification information and the third label, at this time, the first discriminator and the first generator are back propagated based on the loss value to obtain the gradient, and the parameters of the first generator are updated based on the gradient without updating the parameters of the first discriminator.

[0057] Generally, the first training stage and the second training stage are alternately performed, that is, the first discriminator is trained for one or more cycles, the generator is trained for one or more cycles, and the process is repeated until the training of the first generative adversarial network is completed.

[0058] If the training of the first generator is completed, the parameters of the first generator can be initialized to the style transfer network and the second generator, respectively.

[0059] If the training of the first generator is completed, the parameters of the first generator can be initialized to the style transfer network and the second generator, respectively, that is, when the style transfer network is initialized, the style transfer network inherits the parameters of the first generator, and when the second generator is initialized, the second generator inherits the parameters of the first generator.

[0060] Step 1016, under the condition of maintaining the second generator unchanged, training the second discriminator according to the first sample image data and the second sample image data, so that the second discriminator is used to identify whether the image data content under the modeling style of the same building information model platform is the same.

[0061] When the first generative adversarial network is trained, the first generative adversarial network can be ignored, and the second generative adversarial network is trained alone. In the process of training the second generative adversarial network alone, the main task is to train the second discriminator.

[0062] In this process, the second generator is maintained unchanged, and the second discriminator is supervised trained with the first sample image data and the second sample image data as samples, so that the second discriminator has the ability to identify whether the image data content under the modeling style of the same building information model platform is the same.

[0063] In a specific implementation, the second sample image data is input into the first generator, and the first generator converts the second sample image data into fifth sample image data under the modeling style of the building information model platform.

[0064] The second sample pair is constructed, and the second sample pair includes a first image pair, a fourth label representing true (Ture), a second image pair, and a fifth label representing false (False); wherein the first image pair includes first sample image data and fifth sample image data with the same content; and the second image pair includes first sample image data and fifth sample image data with different content.

[0065] The first image pair and the second image pair in the second sample pair are respectively input into the second discriminator using a random or the like manner to generate third classification information; wherein the third classification information represents a probability of the same content.

[0066] Under the condition of maintaining the parameters of the second generator unchanged, the parameters of the second discriminator are updated according to the loss value on the binary classification between the third classification information, the fourth label, and the fifth label, at this time, the second discriminator will be punished because of misclassifying the true sample as the false sample or misclassifying the false sample as the true sample, and the second discriminator updates its parameters through back propagation of the loss value.

[0067] Step 1017, if the second discriminator is completed, the parameters of the second discriminator are initialized to the classifier.

[0068] If the second discriminator is completed, the parameters of the second discriminator can be initialized to the classifier, that is, when the classifier is initialized, the classifier inherits the parameters of the second discriminator.

[0069] In the embodiment, the style transfer network and the classifier are trained based on the first generative adversarial network and the second generative adversarial network, not only the training scheme is mature, the samples are reused, the annotation amount is reduced, and the development cost is effectively reduced, but also the joint training makes the style transfer network and the classifier have relevance, and the cascade use of the style transfer network and the classifier can improve the auditing accuracy of the delivery and acceptance of the decoration materials.

[0070] In the embodiment, since the structures of the first generator, the second generator, and the style transfer network are the same, and the structures of the second discriminator and the classifier are the same, the applications of the first generator, the second generator, and the style transfer network are basically similar, and the applications of the second discriminator and the classifier are basically similar, so the description is relatively simple, and the relevant parts can be referred to the part of the style transfer network and the classifier in steps 104 and 105, which will not be described in detail in the embodiment.

[0071] Step 102, when performing the delivery and acceptance operation on the decoration materials of the target building, a plurality of frames of original image data actually collected from the decoration materials of the target building are received.

[0072] When the construction unit completes the decoration of the target building, the construction personnel can use mobile terminals such as mobile phones and tablet computers in actual environments such as indoor and outdoor environments to collect multiple frames of original image data of the target building (mainly the decoration materials). The construction personnel operates on the BIM platform, triggers the handover and acceptance operation on the decoration materials of the target building, at this time, the multiple frames of original image data can be uploaded to the BIM platform for review.

[0073] Step 103, collecting multiple frames of model image data of the same building material as the multiple frames of original image data in the target building model.

[0074] In the building information model platform, a target building model is stored for three-dimensional modeling of the target building. In this embodiment, multiple frames of model image data are collected in the target building model with reference to the multiple frames of original image data. The so-called correspondence means that each frame of original image data and each frame of model image data describes the same building material.

[0075] In an embodiment of the present application, step 103 can include the following steps:

[0076] Step 1031, for each frame of original image data, performing a semantic segmentation operation on the original image data to obtain first semantic information representing the building material.

[0077] In this embodiment, a semantic segmentation model such as U-Net (U-shaped network), FCN (Fully Convolutional Network), DeepLab, etc. can be constructed in advance, and image data (including real image data and modeling rendered image data) containing various building materials (types) are used as samples to supervise the training of the semantic segmentation model.

[0078] Each frame of original image data is input into the semantic segmentation model, and the semantic segmentation model performs a semantic segmentation operation on each frame of original image data to generate first semantic information representing the building material (type) for each pixel point in each frame of original image data.

[0079] Step 1032, converting the first semantic information into a structured first semantic sequence.

[0080] In this embodiment, the first semantic information on each frame of original image data can be converted into a structured first semantic sequence according to a preset encoding manner, which is convenient for analysis and comparison.

[0081] For example, the original image data is divided into multiple first image blocks according to a preset segmentation manner, such as a rectangular region with a specified width and height.

[0082] In each first image block, a first proportion of a quantity of each first semantic information in the first image block in a quantity of all first semantic information is counted.

[0083] In the same first image block, the first proportions of the various first semantic information are compared, and the first image block is configured with the first semantic information with the highest first proportion.

[0084] The first semantic information configured for each first image block is arranged according to a preset arrangement manner (such as from left to right, from top to bottom, etc.), and a structured first semantic sequence is obtained.

[0085] Step 1033, controlling the lens to capture model image data at multiple positions and multiple angles in the target building model.

[0086] In this embodiment, a walkable area can be defined in the target building model, the height of the walkable area is set according to the habit of human body shooting, multiple shooting points (i.e. horizontal positions) are captured on the walkable area on average, the lens is positioned at the corresponding height (i.e. vertical position) in the shooting point, and the lens is controlled to capture model image data at multiple angles.

[0087] Step 1034, identifying second semantic information representing building materials in each frame of model image data.

[0088] In this embodiment, each frame of model image data is input into a semantic segmentation model, the semantic segmentation model performs a semantic segmentation operation on each frame of model image data, and generates second semantic information representing building materials (types) for each pixel point on each frame of model image data.

[0089] Step 1035, converting the second semantic information into a structured second semantic sequence.

[0090] In this embodiment, the second semantic information on each frame of model image data can be converted into a structured second semantic sequence in the same encoding manner, which is convenient for analysis and comparison.

[0091] Exemplarily, the model image data is divided into multiple second image blocks according to a preset division manner, such as each second image block being a rectangular area with a specified width and height, etc.

[0092] In each second image block, a second proportion of a quantity of each second semantic information in the second image block in a quantity of all second semantic information is counted.

[0093] In the same second image block, the second proportions of the various second semantic information are compared, and the second image block is configured with the second semantic information with the highest second proportion.

[0094] The second semantic information configured for each second image block is arranged according to a preset arrangement mode (such as from left to right, from top to bottom, etc.), and a structured second semantic sequence is obtained.

[0095] In the present example, the first semantic information is converted into a structured first semantic sequence and the second semantic information is converted into a structured second semantic sequence using the mode of block voting, which can reflect the general content of the original image data and the model image data, and the operation is simple and the resource consumption is low.

[0096] Step 1036, the distance between the first semantic sequence and the second semantic sequence is calculated.

[0097] In the present embodiment, the distance between the first semantic sequence and the second semantic sequence can be calculated, such as the Hamming distance, to measure the similarity between the first semantic sequence and the second semantic sequence, thereby representing the similarity between the original image data and the model image data.

[0098] Step 1037, if the distance is less than or equal to a preset similarity threshold, it is determined that the original image data and the model image data describe the same building material.

[0099] The distance between the first semantic sequence and the second semantic sequence is compared with the preset similarity threshold. If the distance between the first semantic sequence and the second semantic sequence is less than or equal to the preset similarity threshold, it indicates that the similarity between the original image data and the model image data is high, and it can be determined that the original image data and the model image data describe the same building material.

[0100] The original image data and the model image data belong to different styles, and there are certain differences in content and texture between them. If SIFT (Scale Invariant Feature Transform) or other operators are used to directly calculate the similarity between the original image data and the model image data, the similarity has high volatility, and it is difficult to select a reasonable threshold.

[0101] In the present embodiment, the target building belongs to a specific semantic environment, and the probability of collision between different positions of the decoration materials in the semantics is low. Therefore, the similarity between the original image data and the model image data is measured in the semantics of the decoration materials, which has good generalization, ensures the accuracy of the similarity, reasonably selects the similarity threshold, and thus better adapts to the comparison task of original image data and model image data of different styles.

[0102] Step 104, the original image data is input into the style transfer network to be converted into target image data in the modeling style of the building information model platform.

[0103] In the embodiment, the original image data is input into the style transfer network, and the style transfer network converts the original image data into target image data in the modeling style of the building information modeling platform, so as to compare whether the contents are the same in the same style with the model image data.

[0104] Since the modeling capability of the building information modeling platform is predictable, the style transfer network has good transfer capability, can transfer the original image data in the real style to the virtual style (i.e., the modeling style of the building information modeling platform), can effectively reduce the information deviation during the transfer, and thus guarantee the accuracy of comparing whether the contents are the same in the same style with the model image data.

[0105] On the contrary, the decoration capability of the construction unit in the real world is not predictable, and if the model image data in the virtual style (i.e., the modeling style of the building information modeling platform) is transferred to the real style, the information deviation during the transfer is large.

[0106] In a structure of the style transfer network, as shown in Figure 2 The style transfer network includes a first convolution block ConvBlock_1, a second convolution block ConvBlock_2, a third convolution block ConvBlock_3, and a fully connected block FCBlock. The first convolution block ConvBlock_1 has two convolution layers, the second convolution block ConvBlock_2 has two convolution layers, the third convolution block ConvBlock_3 has four convolution layers, and the fully connected block FCBlock has two fully connected layers.

[0107] Further, the convolution kernels of the convolution layers in the first convolution block ConvBlock_1, the second convolution block ConvBlock_2, and the third convolution block ConvBlock_3 are all 3×3, and ReLU (Rectified Linear Unit) is used as the activation function.

[0108] In this structure, the original image data is input into the first convolution block ConvBlock_1, and the encoding operation is performed through the two convolution layers to obtain the first image style feature. When the size of the original image data is 224×224×3, the size of the first image style feature is 224×224×64.

[0109] The maximum pooling operation (Max Pooling) is performed on the first image style feature to obtain the second image style feature. The filter of the maximum pooling operation is 2×2, the step is 2, and the size of the second image style feature is 112×112×64.

[0110] The second image style feature is input into the second convolution block ConvBlock_2, and an encoding operation is performed through two convolution layers to obtain a third image style feature. The size of the third image style feature is 112x112x128.

[0111] A maximum pooling operation (Max Pooling) is performed on the third image style feature to obtain a fourth image style feature. The size of the fourth image style feature is 56x56x128.

[0112] The fourth image style feature is input into the third convolution block ConvBlock_3, and a decoding operation is performed through four convolution layers to obtain a fifth image style feature. The size of the fifth image style feature is 56x56x256.

[0113] A maximum pooling operation (Max Pooling) is performed on the fifth image style feature to obtain a sixth image style feature. The size of the sixth image style feature is 28x28x256.

[0114] The sixth image style feature is input into the full connection block FCBlock, and a decoding operation is performed through two full connection layers to obtain a seventh image style feature.

[0115] An activation operation is performed on the seventh image style feature using functions such as Softmax to obtain target image data in the modeling style of the building information modeling platform.

[0116] In step 105, if the target image data and the model image data describe the same building material, the target image data and the model image data are input into the classifier to calculate the probability that the target image data and the model image data have the same content.

[0117] In this embodiment, the corresponding target image data and model image data are aligned according to the original image data, that is, for the original image data and the model image data describing the same building material, the target image data after style transfer of the original image data and the model image data are combined to form an image pair, so that the target image data and the model image data in the image pair describe the same building material.

[0118] At this time, the target image data and the model image data are jointly input into the classifier, and the classifier calculates the probability that the target image data and the model image data have the same content.

[0119] Wherein, the content of the target image data and the model image data being the same can mean that the target image data and the model image data are the same in the texture of the decoration material, the target image data and the model image data are the same in the color of the rusted material, and the like.

[0120] In a structure of a classifier, as shown in Figure 2 The classifier includes a first branch structure Branch_1, a second branch structure Branch_2, a backbone structure Backbone, and a head structure Head. The first branch structure Branch_1 has four convolutional layers, the second branch structure Branch_2 has four convolutional layers, the backbone structure Backbone has three convolutional layers, and the head structure Head has three fully connected layers.

[0121] Further, the convolutional kernels of the convolutional layers in the first branch structure Branch_1, the second branch structure Branch_2, and the backbone structure Backbone are all 3x3, and ReLU is used as the activation function.

[0122] In this structure, the target image data is input into the first branch structure Branch_1, and the features are extracted through four convolutional layers, denoted as the first image content features.

[0123] The model image data is input into the second branch structure Branch_2, and the features are extracted through four convolutional layers, denoted as the second image content features.

[0124] The first image content features and the second image content features are concatenated as new features, denoted as the third image content features.

[0125] The third image content features are input into the backbone structure Backbone, and the features are further extracted through three convolutional layers, denoted as the fourth image content features.

[0126] The fourth image content features are input into the head structure Head to perform a binary classification operation, and the fourth image content features are mapped to the probability of the target image data having the same content as the model image data through three fully connected layers.

[0127] Step 106, calculate the coverage rate of the model image data in the building model.

[0128] In this embodiment, since there are differences in angles when the construction personnel collects the model image data, there are differences in the decoration materials of the model image data. Therefore, multiple frames of model image data can be traversed to calculate the coverage rate of each frame of model image data in the building model in the form of projection, expansion area, etc., so as to facilitate the auditing of the decoration materials.

[0129] In specific implementation, the building model can be vertically projected onto a horizontal plane to obtain an overall projection area, and for each frame of model image data, the structure representing the model image data in the target building model is vertically projected onto the horizontal plane to obtain a local projection area.

[0130] For each frame of model image data, a ratio between an area of the local projection region and an area of the overall projection region is calculated to obtain an area proportion.

[0131] A degree of dispersion (DOD) of the multiple local projection regions in the overall projection region is calculated, such as a log-likelihood ratio.

[0132] For each frame of model image data, the area proportion is multiplied by a sum value between a preset dispersion coefficient (such as 0.5) and the degree of dispersion to obtain a coverage rate of the model image data in the building model.

[0133] Therefore, the coverage rate can be expressed as Coverage = SP x (DP + DOD), where Coverage is the coverage rate, SP is the area proportion, DP is the dispersion coefficient, and DOD is the degree of dispersion.

[0134] In step 107, a delivery acceptance result of the decoration material of the target building is generated according to the probability and the coverage rate.

[0135] In this embodiment, the decoration material of the target building is preliminarily audited according to the probability and the coverage rate of the same content between the multiple frames of target image data and the multiple frames of model image data, and a delivery acceptance result is generated and provided to an auditing unit as a reference for delivery acceptance.

[0136] In a specific implementation, for the multiple frames of model image data, the probability and the coverage rate are substituted into the following formula to calculate a confidence degree of the decoration material of the building meeting the design requirements:

[0137]

[0138] where Score is the confidence degree, n is the number of model image data, Coverage i is the coverage rate corresponding to the i-th frame of model image data, ΔCoverage is a preset compensation coefficient (in the case of mainly using the area of the vertical projection, the compensation coefficient can be used to compensate the expanded area), and P i is the probability corresponding to the i-th frame of model image data.

[0139] The confidence degree is compared with a preset confidence threshold.

[0140] If the confidence degree is greater than or equal to the preset confidence threshold, it is determined that the delivery acceptance result of the decoration material of the building is successful.

[0141] If the confidence degree is less than the preset confidence threshold, it is determined that the delivery acceptance result of the decoration material of the building is failed.

[0142] In addition, in addition to the acceptance success or the acceptance failure, other information such as the confidence, marking the structure related to the original image data in the target building model, describing the decoration materials in the original image data using the LLM (Large Language Model), and the like can be included in the delivery acceptance result, and the present embodiment is not limited thereto.

[0143] In the present embodiment, the style transfer network and the classifier are trained; the style transfer network is used to convert the image data into the modeling style of the building information modeling platform; the classifier is used to identify whether the image data content under the same modeling style of the building information modeling platform is the same; when performing the delivery acceptance operation on the decoration materials of the target building, a plurality of frames of original image data actually collected on the decoration materials of the target building are received; a plurality of frames of model image data describing the same building materials as the plurality of frames of original image data are collected in the target building model; the original image data is input into the style transfer network to be converted into target image data under the modeling style of the building information modeling platform; if the target image data and the model image data describe the same building materials, the target image data and the model image data are input into the classifier to calculate the probability that the target image data and the model image data have the same content; the coverage rate of the model image data in the building model is calculated; and the delivery acceptance result of the decoration materials of the target building is generated according to the probability and the coverage rate. The present embodiment provides a standardized process for preliminary delivery acceptance of the decoration of the building, reduces the experience requirement for the auditors, reduces the threshold of the delivery acceptance, and moreover, the automatic preliminary delivery acceptance can reduce the subjective influence of the auditors, unifies the scale of the delivery acceptance, thereby improving the efficiency of the delivery acceptance of the decoration materials.

[0144] It should be noted that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0145] Referring to Figure 3 , a schematic diagram of a decoration material delivery acceptance device provided by an embodiment of the present application is shown, which is applied to a building information modeling platform, and the building information modeling platform stores a target building model for modeling a target building; as Figure 3 indicated, the device can specifically include the following modules:

[0146] The model training module 301 is configured to train a style transfer network and a classifier; the style transfer network is used to convert image data into the modeling style of the building information modeling platform; and the classifier is used to identify whether the image data content under the same modeling style of the building information modeling platform is the same;

[0147] The image data receiving module 302 is configured to receive a plurality of frames of original image data actually collected from the decoration materials of the target building when performing delivery acceptance operation on the decoration materials of the target building.

[0148] The image data collecting module 303 is configured to collect a plurality of frames of model image data describing the same building materials as the original image data in the target building model.

[0149] The image data migrating module 304 is configured to input the original image data into the style migration network to convert the original image data into target image data in the modeling style of the building information modeling platform.

[0150] The probability calculating module 305 is configured to input the target image data and the model image data into the classifier to calculate a probability that the target image data and the model image data describe the same building materials if the target image data and the model image data describe the same building materials.

[0151] The coverage calculating module 306 is configured to calculate a coverage of the model image data in the building model.

[0152] The delivery acceptance module 307 is configured to generate a delivery acceptance result of the decoration materials of the target building according to the probability and the coverage.

[0153] Optionally, the model training module 301 comprises:

[0154] The sample collecting module is configured to collect first sample image data and second sample image data; the first sample image data is image data collected from the sample building model, and the second sample image data is image data actually collected from the decoration materials of the sample building.

[0155] The model loading module is configured to load a first generative adversarial network, a second generative adversarial network, a style migration network and a classifier; the first generative adversarial network comprises a first generator and a first discriminator, and the second generative adversarial network comprises a second generator and a second discriminator; the first generator, the second generator and the style migration network have the same structure, and the second discriminator and the classifier have the same structure.

[0156] The first discriminator training module is configured to train the first discriminator according to the first sample image data and the second sample image data while maintaining the first generator unchanged, so that the first discriminator is used to classify whether the image data is in the modeling style of the building information modeling platform.

[0157] the first generator is trained based on the second sample image data under the condition that the first discriminator is maintained unchanged, so that the first generator is used for converting image data into a modeling style of the building information modeling platform;

[0158] the first initialization module is configured to initialize parameters of the first generator to the style transfer network and the second generator respectively if the training of the first generator is completed;

[0159] the second discriminator training module is configured to train the second discriminator based on the first sample image data and the second sample image data under the condition that the second generator is maintained unchanged, so that the second discriminator is used for identifying whether image data contents in the same modeling style of the building information modeling platform are the same;

[0160] the second initialization module is configured to initialize parameters of the second discriminator into the classifier if the training of the second discriminator is completed.

[0161] Optionally, the first discriminator training module is further configured to:

[0162] input the second sample image data into the first generator to convert the second sample image data into third sample image data in the modeling style of the building information modeling platform;

[0163] construct a first sample pair; the first sample pair includes the first sample image data and a first label representing true, and the third sample image data and a second label representing false;

[0164] input the first sample image data and the third sample image data in the first sample pair into the first discriminator respectively to generate first classification information; the first classification information represents whether it is the modeling style of the building information modeling platform;

[0165] update parameters of the first discriminator based on a loss value between the first classification information, the first label and the second label under the condition that the parameters of the first generator are maintained unchanged;

[0166] the generator training module is further configured to:

[0167] input the second sample image data into the first generator to convert the second sample image data into fourth sample image data in the modeling style of the building information modeling platform;

[0168] label the fourth sample image data with a third label representing true;

[0169] inputting the fourth sample image data into the first discriminator to generate second classification information; the second classification information indicates whether it is a modeling style of the building information model platform;

[0170] updating parameters of the first generator according to a loss value between the second classification information and the third label while maintaining parameters of the first discriminator unchanged;

[0171] The second discriminator training module is further configured to:

[0172] inputting the second sample image data into the first generator to convert the second sample image data into fifth sample image data in a modeling style of the building information model platform;

[0173] constructing a second sample pair; the second sample pair includes a first image pair and a fourth label indicating true, and a second image pair and a fifth label indicating false; the first image pair includes the first sample image data and the fifth sample image data with same content; the second image pair includes the first sample image data and the fifth sample image data with different content;

[0174] inputting the first image pair and the second image pair in the second sample pair into the second discriminator to generate third classification information; the third classification information indicates a probability of same content;

[0175] updating parameters of the second discriminator according to a loss value between the third classification information, the fourth label and the fifth label while maintaining parameters of the second generator unchanged.

[0176] Optionally, the image data acquisition module 303 includes:

[0177] a first semantic segmentation module configured to perform semantic segmentation operation on the original image data to obtain first semantic information indicating building materials;

[0178] a first semantic sequence conversion module configured to convert the first semantic information into a structured first semantic sequence;

[0179] a model acquisition module configured to control a lens to acquire model image data at multiple positions and from multiple angles in the target building model;

[0180] a second semantic segmentation module configured to identify second semantic information indicating building materials in the model image data;

[0181] a second semantic sequence conversion module configured to convert the second semantic information into a structured second semantic sequence;

[0182] a semantic distance calculation module, configured to calculate a distance between the first semantic sequence and the second semantic sequence;

[0183] a description determination module, configured to determine that the original image data and the model image data describe the same building material if the distance is less than or equal to a preset similarity threshold.

[0184] Optionally, the first semantic sequence conversion module is further configured to:

[0185] divide the original image data into a plurality of first image blocks according to a preset division manner;

[0186] calculate a first proportion of each first semantic information in the first image blocks;

[0187] configure the first semantic information with the highest first proportion to the first image blocks;

[0188] arrange the first semantic information configured to each first image block according to a preset arrangement manner to obtain a structured first semantic sequence;

[0189] the second semantic sequence conversion module is further configured to:

[0190] divide the model image data into a plurality of second image blocks according to a preset division manner;

[0191] calculate a second proportion of each second semantic information in the second image blocks;

[0192] configure the second semantic information with the highest second proportion to the second image blocks;

[0193] arrange the second semantic information configured to each second image block according to a preset arrangement manner to obtain a structured second semantic sequence.

[0194] Optionally, the style transfer network comprises a first convolutional block, a second convolutional block, a third convolutional block and a fully connected block; the first convolutional block has two convolutional layers, the second convolutional block has two convolutional layers, the third convolutional block has four convolutional layers, and the fully connected block has two fully connected layers;

[0195] the image data transfer module 304 is further configured to:

[0196] input the original image data into the first convolutional block to perform an encoding operation to obtain first image style features;

[0197] perform a max-pooling operation on the first image style features to obtain second image style features;

[0198] The second image style feature is input into the second convolution block to perform an encoding operation, to obtain a third image style feature;

[0199] A maximum pooling operation is performed on the third image style feature to obtain a fourth image style feature;

[0200] The fourth image style feature is input into the third convolution block to perform a decoding operation, to obtain a fifth image style feature;

[0201] A maximum pooling operation is performed on the fifth image style feature to obtain a sixth image style feature;

[0202] The sixth image style feature is input into the fully connected block to perform a decoding operation, to obtain a seventh image style feature;

[0203] An activation operation is performed on the seventh image style feature to obtain target image data in the modeling style of the building information modeling platform.

[0204] Optionally, the classifier comprises a first branch structure, a second branch structure, a trunk structure, and a classification head structure; the first branch structure has four convolution layers, the second branch structure has four convolution layers, the trunk structure has three convolution layers, and the classification head structure has three fully connected layers;

[0205] The probability calculation module 305 is further configured to:

[0206] The target image data is input into the first branch structure to extract a first image content feature;

[0207] The model image data is input into the second branch structure to extract a second image content feature;

[0208] The first image content feature and the second image content feature are spliced into a third image content feature;

[0209] The third image content feature is input into the trunk structure to extract a fourth image content feature;

[0210] The fourth image content feature is input into the classification head structure to be mapped into a probability that the target image data and the model image data have the same content.

[0211] Optionally, the coverage calculation module 306 is further configured to:

[0212] The building model is projected vertically to a horizontal plane to obtain an overall projection area;

[0213] For each frame of the model image data, a structure in the target building model representing the model image data is projected vertically to a horizontal plane to obtain a local projection area;

[0214] For each frame of the model image data, a ratio between an area of the local projection area and an area of the overall projection area is calculated to obtain an area ratio;

[0215] A dispersion degree of a plurality of the local projection areas in the overall projection area is calculated;

[0216] For each frame of the model image data, the area ratio is multiplied by a sum value between a preset dispersion coefficient and the dispersion degree to obtain a coverage of the model image data in the building model;

[0217] The delivery acceptance module 307 is further configured to:

[0218] For a plurality of frames of the model image data, the probability and the coverage are substituted into the following formula to calculate a confidence degree of the decoration materials of the building meeting design requirements:

[0219]

[0220] wherein, Score is the confidence degree, n is the number of the model image data, Coverage i is the coverage corresponding to the i-th frame of the model image data, ΔCoverage is a preset compensation coefficient, P i is the probability corresponding to the i-th frame of the model image data;

[0221] If the confidence degree is greater than or equal to a preset confidence threshold, it is determined that the delivery acceptance result of the decoration materials of the building is successful;

[0222] If the confidence degree is less than the preset confidence threshold, it is determined that the delivery acceptance result of the decoration materials of the building is failed.

[0223] The embodiment of the present application provides a kind of decoration material delivery acceptance device, application this device, it can realize each step in each method embodiment.

[0224] For device embodiment, since it is basically similar to method embodiment, so it is described more simply, relevant place refers to the description of method embodiment part.

[0225] Referring to Figure 4 , a schematic diagram of a terminal device provided by an embodiment of the present application is shown. As Figure 4As shown, the terminal device 400 in the embodiments of the present application includes a processor 410, a memory 420, and a computer program 421 stored in the memory 420 and executable on the processor 410. The processor 410 implements the steps in the above-described decoration material delivery and acceptance method embodiments when executing the computer program 421. Alternatively, the processor 410 implements the functions of the modules / units in the above-described device embodiments when executing the computer program 421.

[0226] For example, the computer program 421 can be divided into one or more modules / units stored in the memory 420 and executed by the processor 410 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which can be used to describe the execution process of the computer program 421 in the terminal device 400.

[0227] The terminal device 400 can include, but is not limited to, the processor 410 and the memory 420. Those skilled in the art can understand that the terminal device 400 can include more or less components, or combine certain components, or different components, for example, the terminal device 400 can also include an input / output device, a network access device, a bus, etc. Figure 4 The terminal device 400 is only an example and does not constitute a limitation on the terminal device 400, and can include more or less components than the illustration, or combine certain components, or different components, for example, the terminal device 400 can also include an input / output device, a network access device, a bus, etc.

[0228] The processor 410 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0229] The memory 420 can be an internal storage unit of the terminal device 400, for example, a hard disk or a memory of the terminal device 400. The memory 420 can also be an external storage device of the terminal device 400, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like equipped on the terminal device 400. Further, the memory 420 can also include both the internal storage unit and the external storage device of the terminal device 400. The memory 420 is used to store the computer program 421 and other programs and data required by the terminal device 400. The memory 420 can also be used to temporarily store data that has been output or will be output.

[0230] The embodiment of the present application further discloses a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the decoration material delivery and acceptance method as described in the foregoing various embodiments.

[0231] The embodiment of the present application further discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the decoration material delivery and acceptance method as described in the foregoing various embodiments.

[0232] The embodiment of the present application further discloses a computer program product, which, when running on a computer, causes the computer to execute the decoration material delivery and acceptance method as described in the foregoing various embodiments.

[0233] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them. Although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method of delivering acceptance of a finishing material, characterized by, The method is applied to a building information model platform, the building information model platform stores a target building model modeling a target building; the method comprises: training a style transfer network and a classifier; the style transfer network is used for converting image data into a modeling style of the building information model platform; the classifier is used for identifying whether image data under the same modeling style of the building information model platform is the same in content; when performing a delivery acceptance operation on decoration materials of the target building, receiving a plurality of frames of original image data actually collected from the decoration materials of the target building; collecting a plurality of frames of model image data from the target building model, the model image data describing the same building materials as the plurality of frames of original image data; inputting the original image data into the style transfer network to convert the original image data into target image data under the modeling style of the building information model platform; if the target image data and the model image data describe the same building materials, inputting the target image data and the model image data into the classifier to calculate a probability that the target image data and the model image data are the same in content; calculating a coverage rate of the model image data in the building model; generating a delivery acceptance result of the decoration materials of the target building according to the probability and the coverage rate.

2. The method of claim 1, wherein, The building information model platform stores a sample building model modeling a sample building; The training of the style transfer network and the classifier comprises: collecting first sample image data and second sample image data; the first sample image data is image data collected from the sample building model, and the second sample image data is image data actually collected from decoration materials of the sample building; loading a first generative adversarial network, a second generative adversarial network, a style transfer network and a classifier; the first generative adversarial network has a first generator and a first discriminator, and the second generative adversarial network has a second generator and a second discriminator; the first generator, the second generator and the style transfer network have the same structure, and the second discriminator and the classifier have the same structure; under the condition of maintaining the first generator unchanged, training the first discriminator according to the first sample image data and the second sample image data, so that the first discriminator is used for classifying whether image data is a modeling style of the building information model platform; if the first discriminator is trained, under the condition of maintaining the first discriminator unchanged, training the first generator according to the second sample image data, so that the first generator is used for converting image data into a modeling style of the building information model platform; if the first generator is trained, initializing parameters of the first generator to the style transfer network and the second generator, respectively; under the condition of maintaining the second generator unchanged, training the second discriminator according to the first sample image data and the second sample image data, so that the second discriminator is used for identifying whether image data under the same modeling style of the building information model platform is the same in content; If the second discriminator is completed, the parameters of the second discriminator are initialized into the classifier.

3. The method of claim 2, wherein, the training the first discriminator according to the first sample image data and the second sample image data under the condition of maintaining the first generator unchanged, so that the first discriminator is used for classifying whether image data is the modeling style of the building information modeling platform, comprises: inputting the second sample image data into the first generator to convert into third sample image data under the modeling style of the building information modeling platform; constructing a first sample pair; the first sample pair comprises the first sample image data and a first label representing true, the third sample image data and a second label representing false; inputting the first sample image data and the third sample image data in the first sample pair into the first discriminator respectively to generate first classification information; the first classification information represents whether it is the modeling style of the building information modeling platform; updating the parameters of the first discriminator according to the loss value between the first classification information, the first label and the second label under the condition of maintaining the parameters of the first generator unchanged; the training the first generator according to the second sample image data under the condition of maintaining the first discriminator unchanged, so that the first generator is used for converting image data into the modeling style of the building information modeling platform, comprises: inputting the second sample image data into the first generator to convert into fourth sample image data under the modeling style of the building information modeling platform; labeling the fourth sample image data with a third label representing true; inputting the fourth sample image data into the first discriminator to generate second classification information; the second classification information represents whether it is the modeling style of the building information modeling platform; updating the parameters of the first generator according to the loss value between the second classification information and the third label under the condition of maintaining the parameters of the first discriminator unchanged; the training the second discriminator according to the first sample image data and the second sample image data under the condition of maintaining the second generator unchanged, so that the second discriminator is used for identifying whether image data contents under the same modeling style of the building information modeling platform are the same, comprises: inputting the second sample image data into the first generator to convert into fifth sample image data under the modeling style of the building information modeling platform; constructing a second sample pair; the second sample pair comprises a first image pair and a fourth label representing true, a second image pair and a fifth label representing false; the first image pair comprises the first sample image data and the fifth sample image data with the same content; the second image pair comprises the first sample image data and the fifth sample image data with different contents; inputting the first image pair and the second image pair in the second sample pair into the second discriminator respectively to generate third classification information; the third classification information represents the probability of the same content; The parameters of the second discriminator are updated according to a loss value between the third classification information, the fourth label and the fifth label while the parameters of the second generator remain unchanged.

4. The method of claim 1, wherein, The collecting of the plurality of model image data in the target building model and the plurality of original image data comprises: performing a semantic segmentation operation on the original image data to obtain first semantic information representing building materials; converting the first semantic information into a structured first semantic sequence; controlling the lens to collect model image data at a plurality of positions and angles in the target building model; identifying second semantic information representing building materials in each frame of the model image data; converting the second semantic information into a structured second semantic sequence; calculating the distance between the first semantic sequence and the second semantic sequence; if the distance is less than or equal to a preset similarity threshold, determining that the original image data and the model image data describe the same building materials.

5. The method of claim 4, wherein the converting of the first semantic information into a structured first semantic sequence comprises: segmenting the original image data into a plurality of first image blocks according to a preset segmentation manner; counting a first proportion of each first semantic information in the first image blocks; configuring the first semantic information with the highest first proportion to the first image blocks; arranging the first semantic information configured to each first image block according to a preset arrangement manner to obtain a structured first semantic sequence; the converting of the second semantic information into a structured second semantic sequence comprises: segmenting the model image data into a plurality of second image blocks according to a preset segmentation manner; counting a second proportion of each second semantic information in the second image blocks; configuring the second semantic information with the highest second proportion to the second image blocks; arranging the second semantic information configured to each second image block according to a preset arrangement manner to obtain a structured second semantic sequence.

6. The method according to any one of claims 1-5, characterized in that, The style transfer network comprises a first convolutional block, a second convolutional block, a third convolutional block and a fully connected block; the first convolutional block has two convolutional layers, the second convolutional block has two convolutional layers, the third convolutional block has four convolutional layers, and the fully connected block has two fully connected layers; the inputting of the original image data into the style transfer network to convert the original image data into target image data in the modeling style of the building information modeling platform comprises: inputting the original image data into the first convolutional block to perform an encoding operation to obtain first image style features; performing a max-pooling operation on the first image style features to obtain second image style features; inputting the second image style features into the second convolutional block to perform an encoding operation to obtain third image style features; performing a max-pooling operation on the third image style features to obtain fourth image style features; inputting the fourth image style features into the third convolutional block to perform a decoding operation to obtain fifth image style features; perform a max-pooling operation on the fifth image style feature to obtain a sixth image style feature; input the sixth image style feature into the fully connected block to perform a decoding operation to obtain a seventh image style feature; perform an activation operation on the seventh image style feature to obtain target image data in a modeling style of the building information modeling platform.

7. The method of claim 6, wherein, The classifier includes a first branch structure, a second branch structure, a trunk structure, and a classification head structure; the first branch structure has four convolutional layers, the second branch structure has four convolutional layers, the trunk structure has three convolutional layers, and the classification head structure has three fully connected layers; The calculation of the probability that the target image data and the model image data have the same content includes: inputting the target image data into the first branch structure to extract first image content features; inputting the model image data into the second branch structure to extract second image content features; concatenating the first image content features and the second image content features into third image content features; inputting the third image content features into the trunk structure to extract fourth image content features; inputting the fourth image content features into the classification head structure to map the probability that the target image data and the model image data have the same content.

8. The method of any one of claims 1-5, wherein The calculation of the coverage of the model image data in the building model includes: vertically projecting the building model to a horizontal plane to obtain an overall projection area; for each frame of the model image data, vertically projecting a structure in the target building model representing the model image data to a horizontal plane to obtain a local projection area; for each frame of the model image data, calculating a ratio between an area of the local projection area and an area of the overall projection area to obtain an area ratio; calculating a dispersion degree of a plurality of local projection areas in the overall projection area; for each frame of the model image data, multiplying the area ratio by a sum of a preset dispersion coefficient and the dispersion degree to obtain the coverage of the model image data in the building model. The generation of the delivery acceptance result for the decoration materials of the target building based on the probability and the coverage includes: for a plurality of frames of the model image data, substituting the probability and the coverage into the following formula to calculate a confidence degree that the decoration materials of the building meet design requirements: wherein Score is the confidence, n is the number of the model image data, Coverage i is the coverage corresponding to the model image data of the i-th frame, ΔCoverage is a preset compensation coefficient, P i is the probability corresponding to the model image data of the i-th frame; if the confidence degree is greater than or equal to a preset confidence threshold, determining that the delivery acceptance result for the decoration materials of the building is successful; if the confidence degree is less than the preset confidence threshold, determining that the delivery acceptance result for the decoration materials of the building is unsuccessful.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the delivery acceptance method for the decoration materials according to any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the delivery acceptance method for the decoration materials according to any one of claims 1-8.

Images