Plane floor plan structured detection method and device and computer equipment

By extracting and fusing floor plan feature maps of different scales and generating identification boxes, the problem of insufficient speed and accuracy in floor plan detection technology is solved, and more efficient and accurate target detection is achieved, suitable for floor plan recognition in complex scenarios.

CN119006847AActive Publication Date: 2024-11-22SHENZHEN UNIV +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411490098.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-24
Publication Date
2024-11-22
Estimated Expiration
2044-10-24

AI Technical Summary

Technical Problem

The existing floor plan detection technology is insufficient in real-time and fine inspection, and cannot accurately detect the position and shape of room elements, especially in complex situations, and is not handled well.

Method used

By obtaining floor plan, extracting floor plan feature maps of different scales, generating floor plan feature sequences, selecting preset number feature vectors, generating identification boxes to mark target positions and categories, and using feature interaction within scale and cross-scale feature fusion processing, enhancing The scale invariance and rotation invariance of the algorithm improve detection accuracy and robustness.

Benefits of technology

It improves the recognition efficiency and accuracy of floor plan, can more accurately locate and identify targets, meet real-time detection requirements, reduce error rates, and improve automation levels, and is especially suitable for real estate planning and interior design fields.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006847A_ABST
    Figure CN119006847A_ABST
Patent Text Reader

Abstract

The invention relates to a structured detection method and device for a planar floor plan and computer equipment. The method comprises the following steps: acquiring a planar house type image, wherein the planar house type image comprises a plurality of to-be-identified objects; house type feature maps of different scales are extracted from the plane house type map, a house type feature sequence is generated based on the house type feature maps of different scales, and the house type feature sequence comprises a plurality of house type feature vectors; for each to-be-identified object, selecting a preset number of house type feature vectors corresponding to the to-be-identified object from the house type feature sequence; based on the preset number of house type feature vectors of the to-be-recognized objects, multiple recognition frames are generated in the plane house type image, and the recognition frames are used for marking the target positions of the to-be-recognized objects in the plane house type image and the target categories of the to-be-recognized objects. By adopting the method, the requirement of real-time detection of the house type image can be met, meanwhile, high detection precision is ensured, and the method has important significance for improving the working efficiency and quality.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device and computer equipment for structured detection of floor plan drawings. Background Art

[0002] The main problem that current floor plan detection technology solves is to automatically identify and extract rooms and room elements in floor plans, such as walls, doors, windows, etc. This technology can help the real estate industry, interior design field, and architectural planning field to automatically generate floor plans, save labor costs, and improve work efficiency.

[0003] However, the current floor plan detection technology still has some shortcomings. For example, in some scenarios where real-time floor plan detection is required, it cannot meet higher speed and efficiency requirements. In some scenarios where detailed floor plan detection is required, it cannot accurately detect the position and shape of each element in the room, and cannot process images in various complex situations. Summary of the invention

[0004] Based on this, it is necessary to provide a floor plan structured detection method, device, computer equipment, computer-readable storage medium and computer program product that can improve the recognition efficiency and recognition accuracy of floor plan in order to address the above technical problems.

[0005] In a first aspect, the present application provides a method for structured detection of a floor plan, comprising:

[0006] Acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified;

[0007] Extracting apartment feature graphs of different scales from the floor plan, and generating an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors;

[0008] For each object to be identified, selecting a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0009] Based on a preset number of apartment feature vectors of each of the multiple objects to be identified, multiple identification boxes are generated in the floor plan, and the identification boxes are used to mark the target positions of the objects to be identified and the target categories of the objects to be identified in the floor plan.

[0010] In one embodiment, generating a housing type feature sequence based on the housing type feature graphs of different scales includes:

[0011] For each of the multiple apartment feature maps of different scales, the targeted apartment feature map is subjected to intra-scale feature interaction to obtain a target feature map; and the multiple target feature maps of different scales are subjected to cross-scale feature fusion processing to obtain an apartment feature sequence.

[0012] In one embodiment, performing in-scale feature interaction on the targeted apartment feature map to obtain a target feature map includes:

[0013] The targeted apartment feature map is converted into a one-dimensional vector sequence; feature vectors at different positions in the one-dimensional vector sequence are subjected to feature interaction to obtain interactive feature vectors at different positions; the interactive feature vectors at different positions are subjected to nonlinear transformation to obtain nonlinear feature vectors at different positions, and the nonlinear feature vectors at different positions constitute a target vector sequence; the target vector sequence is restored to obtain a target feature map.

[0014] In one embodiment, converting the targeted apartment feature map into a one-dimensional vector sequence includes:

[0015] The characteristic vector of each position in the targeted apartment characteristic map is stretched to obtain a plurality of first one-dimensional vectors; and the plurality of first one-dimensional vectors are arranged according to the priority order corresponding to each of the plurality of first one-dimensional vectors to obtain a one-dimensional vector sequence.

[0016] In one embodiment, the cross-scale feature fusion processing of target feature maps of multiple different scales to obtain a house type feature sequence includes:

[0017] A plurality of target feature maps of different scales are sampled and aligned to obtain a plurality of aligned feature maps with the same resolution; the plurality of aligned feature maps are fused through multi-layer convolution operations to obtain a fused feature map; a feature vector at each position in the fused feature map is stretched to obtain a plurality of second one-dimensional vectors; the plurality of second one-dimensional vectors are arranged according to their respective corresponding priority orders to obtain a housing feature sequence.

[0018] In one of the embodiments, for each of the multiple objects to be identified, based on a preset number of floor plan feature vectors of the object to be identified, a confidence score of the identification box corresponding to the object to be identified in the floor plan is determined, and the confidence score is used to characterize the probability that the target position of the identification box corresponding to the object to be identified in the floor plan coincides with the actual position of the object to be identified in the floor plan; and the confidence scores corresponding to each of the multiple identification boxes are displayed in the floor plan.

[0019] In a second aspect, the present application further provides a structured detection device for a floor plan, comprising:

[0020] An acquisition module, used to acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified;

[0021] A first generating module is used to extract apartment feature graphs of different scales from the floor plan, and generate an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors;

[0022] A selection module, for selecting, for each object to be identified, a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0023] The second generating module is used to generate a plurality of identification boxes in the floor plan diagram based on a preset number of apartment feature vectors of each of the plurality of objects to be identified, wherein the identification boxes are used to mark the target positions of the objects to be identified in the floor plan diagram and the target categories of the objects to be identified.

[0024] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0025] Acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified;

[0026] Extracting apartment feature graphs of different scales from the floor plan, and generating an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors;

[0027] For each object to be identified, selecting a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0028] Based on a preset number of apartment feature vectors of each of the multiple objects to be identified, multiple identification boxes are generated in the floor plan, and the identification boxes are used to mark the target positions of the objects to be identified and the target categories of the objects to be identified in the floor plan.

[0029] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0030] Acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified;

[0031] Extracting apartment feature graphs of different scales from the floor plan, and generating an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors;

[0032] For each object to be identified, selecting a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0033] Based on a preset number of apartment feature vectors of each of the multiple objects to be identified, multiple identification boxes are generated in the floor plan, and the identification boxes are used to mark the target positions of the objects to be identified and the target categories of the objects to be identified in the floor plan.

[0034] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0035] Acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified;

[0036] Extracting apartment feature graphs of different scales from the floor plan, and generating an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors;

[0037] For each object to be identified, selecting a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0038] Based on a preset number of apartment feature vectors of each of the multiple objects to be identified, multiple identification boxes are generated in the floor plan, and the identification boxes are used to mark the target positions of the objects to be identified and the target categories of the objects to be identified in the floor plan.

[0039] The above-mentioned floor plan structured detection method, device, computer equipment, computer-readable storage medium and computer program product obtain a floor plan, which includes multiple objects to be identified; ensure that all objects to be identified can be included, and provide original data support for subsequent feature extraction and target detection.

[0040] The apartment feature maps of different scales are extracted from the floor plan, and the apartment feature sequences are generated based on the apartment feature maps of different scales. The apartment feature sequences include multiple apartment feature vectors, which effectively solves the challenges brought by scale changes and rotations and improves the scale invariance and rotation invariance of the algorithm. In addition, the generated apartment feature sequences can better capture the detailed information of the room and its elements, which helps to improve the accuracy and robustness of detection, especially when dealing with complex structures and layouts.

[0041] For each object to be identified, a preset number of apartment feature vectors corresponding to the object to be identified are selected from the apartment feature sequence; the recognition ability of specific objects is enhanced. It helps to reduce the interference of irrelevant features and improve the accuracy and efficiency of target detection, especially in the face of occlusion and noise problems, and can more accurately locate and identify targets.

[0042] Based on a preset number of apartment feature vectors of multiple objects to be identified, multiple identification boxes are generated in the floor plan. The identification boxes are used to mark the target position and target category of the objects to be identified in the floor plan. Through this process, not only can the positions of the objects to be identified be accurately located, but also these objects can be correctly classified. This method can meet the requirements of real-time detection of apartment plans while ensuring high accuracy of detection, which is of great significance for improving work efficiency and quality. Especially when applied to fields such as real estate planning and interior design, it can significantly improve the level of automation, reduce manual intervention, and reduce the error rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A diagram of an application environment of a structured detection method for a floor plan in an embodiment;

[0045] Figure 2 1 is a schematic diagram of a flow chart of a structured detection method for a floor plan in one embodiment;

[0046] Figure 3 A network structure diagram of a structured detection method for a floor plan in an embodiment;

[0047] Figure 4 A schematic diagram of a flow chart of a structured detection method for a floor plan in another embodiment;

[0048] Figure 5 It is a structural block diagram of a device for detecting structured floor plan diagrams in one embodiment;

[0049] Figure 6 It is a structural block diagram of a structured detection device for a floor plan in another embodiment;

[0050] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] The structured detection method for floor plan diagrams provided in the embodiments of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 generates a structured detection request for the floor plan, and sends the structured detection request for the floor plan to the server 104, so that the server 104 generates multiple identification boxes in the floor plan based on a preset number of floor plan feature vectors of each of the multiple objects to be identified. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0053] In an exemplary embodiment, Figure 2 As shown, a method for detecting a structured floor plan is provided, and the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 202 to 208. Among them:

[0054] Step 202, obtaining a floor plan, where the floor plan includes a plurality of objects to be identified.

[0055] Among them, the floor plan refers to the spatial layout of the interior of the house, which usually includes but is not limited to elements such as rooms, corridors, doors, windows, etc. This type of diagram is usually used to show the floor plan of a residential or commercial space, including the location, size, shape of each room, and the location of openings such as doors and windows.

[0056] Objects to be identified refer to specific elements that need to be detected and identified in the floor plan. These elements mainly include but are not limited to: rooms (such as bedrooms, living rooms, kitchens, bathrooms, etc.), room elements (such as walls, doors, windows, etc.), and other structures (such as balconies, stairs, corridors, etc.).

[0057] Specifically, the original floor plan data is first collected, and the collected original floor plan data is subjected to image preprocessing. It is understandable that the specific image preprocessing can be set according to actual needs, and this application does not make specific limitations. Optionally, the image preprocessing method for the original floor plan data includes: the first step: converting the original floor plan data into a unified data format. For example, JPEG (Joint Photographic Experts Group), PNG (Portable Network Graphics), PDF (Portable Document Format). The second step: adjusting the data after the format is unified to a unified size. For example, 1024x1024 pixels. The third step: denoising, sharpening and other processing are performed on the data after the size is unified to improve the image quality, reduce the impact of noise on detection, and finally obtain the floor plan.

[0058] Step 204: extracting apartment feature graphs of different scales from the floor plan, and generating an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors.

[0059] Among them, the apartment feature maps of different scales refer to multi-resolution feature maps extracted from the floor plan. These feature maps capture different details and abstract information of the image at different levels. Optionally, the apartment feature maps of different scales include low-level feature maps, middle-level feature maps, and high-level feature maps. Among them, the low-level feature map contains more detailed information and is suitable for detecting smaller objects, such as doors, windows, etc. The middle-level feature map contains medium-level abstract information and is suitable for detecting medium-sized objects, such as furniture, etc. The high-level feature map contains more abstract information and is suitable for detecting larger objects, such as rooms, corridors, etc.

[0060] The apartment feature sequence refers to a series of feature vectors generated by processing apartment feature maps of different scales. These feature vectors are arranged in a certain order to form a sequence.

[0061] The apartment feature vector refers to a unit in the apartment feature sequence, and each vector contains multi-scale feature information at a specific location.

[0062] Specifically, the backbone network (such as SwinTransformer, sliding window transformer) is first used to extract features from the input floor plan. The backbone network outputs feature maps of different resolutions at different stages, including: outputting low-level feature maps, which contain more detailed information and are suitable for detecting small objects (such as doors and windows). Outputting mid-level feature maps, which contain medium-level abstract information and are suitable for detecting medium-sized objects (such as furniture). Outputting high-level feature maps, which contain high-level abstract information and are suitable for detecting large objects (such as rooms and corridors).

[0063] Then, feature maps of the same scale interact with each other to enhance the expressiveness of local information. For example, the feature maps are processed using a Transformer Encoder structure, including a multi-head self-attention mechanism (MHSA) and a feed-forward neural network (FFN).

[0064] Next, the feature maps of different scales after each interaction are fused to generate a multi-scale feature map containing richer information. For example, by combining feature maps from different layers, it is possible to consider information at different levels of abstraction at the same time. The feature maps of lower layers may contain more detailed information, while the feature maps of higher layers carry higher levels of abstract information. By fusing this information, the accuracy of detection can be improved.

[0065] Finally, the feature vectors of each position in the fused multi-scale feature map (i.e., the apartment type feature vectors) are arranged in a certain order to form an ordered sequence (i.e., the apartment type feature sequence).

[0066] Step 206: for each object to be identified, select a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence.

[0067] The preset number refers to a fixed number set in advance during the feature selection process, which represents the number of feature vectors selected from the house type feature sequence for each object to be identified. It can be understood that this number is determined based on actual needs, aiming to balance the richness of feature representation and computational efficiency.

[0068] Specifically, taking one of the multiple objects to be identified as an example, according to the object to be identified, a preset number of apartment feature vectors at different positions corresponding to the object to be identified are selected from the apartment feature sequence. For example, if the object to be identified is a kitchen, the preset number is pre-set to 100, then 100 apartment feature vectors at different positions corresponding to the kitchen are selected from the apartment feature sequence.

[0069] Step 208, based on a preset number of apartment feature vectors of each of the multiple objects to be identified, a plurality of identification boxes are generated in the floor plan, and the identification boxes are used to mark the target positions and target categories of the objects to be identified in the floor plan.

[0070] The recognition box refers to the rectangular box that marks the object to be recognized in the floor plan. The purpose of this rectangular box is to accurately locate and surround the object to be recognized, such as rooms, corridors, doors and windows, etc. The recognition box is usually represented by four coordinate values, namely the horizontal and vertical coordinates of the upper left corner and the horizontal and vertical coordinates of the lower right corner, that is, (x1, y1, x2, y2).

[0071] The target position refers to the specific position of the object to be identified in the floor plan. The target position can be determined by the identification frame. Specifically, the target position can be represented by the center point of the identification frame or the four vertices of the identification frame, which can be set according to actual needs.

[0072] The target category refers to the type or category of the object to be identified. In a floor plan, common target categories include rooms, corridors, doors and windows, etc. Each recognition box not only marks the location of the object, but also the category of the object. Optionally, the target category is represented by a label.

[0073] Specifically, the decoder first receives the initial object query selected from the feature sequence output by the encoder. The decoder gradually adjusts and optimizes the position and confidence score of the object query through a multi-layer attention mechanism and a feed-forward network (FFN). Next, in each iteration, the auxiliary prediction head in the decoder generates a temporary bounding box and confidence score, providing additional supervision signals to help the decoder learn and optimize object queries faster. Finally, after multiple iterations of optimization, the bounding box (i.e., recognition box) and confidence score generated by the decoder are output as the final detection result. For example, when the confidence score meets the preset confidence score condition, the iteration is stopped, and the generated recognition box and the corresponding confidence score are the final detection result output. And these recognition boxes are used to mark the target position and target category of the object to be identified in the floor plan.

[0074] In one embodiment, for each of the multiple apartment feature maps of different scales, the targeted apartment feature map is subjected to intra-scale feature interaction to obtain a target feature map; and the multiple target feature maps of different scales are subjected to cross-scale feature fusion processing to obtain an apartment feature sequence.

[0075] Among them, the target feature map refers to the enhanced feature map obtained by processing the feature map of a certain scale during the intra-scale feature interaction (AIFI) process. These enhanced feature maps contain more local information expression capabilities and can better capture the details and structures in the image.

[0076] The house type feature sequence refers to a feature vector sequence containing multi-scale information generated by cross-scale feature fusion processing. This sequence integrates the information in the target feature maps of different scales to form an ordered vector sequence containing rich multi-scale features.

[0077] Specifically, the feature map of each scale is first stretched into a one-dimensional vector for feature interaction processing. Then, the multi-head self-attention mechanism is used to capture the relationship between different positions in the feature map and enhance the expression ability of local information. Next, the feature vector is further processed through the feedforward network to enhance the expression ability of the feature. The processed feature vector is readjusted back to a two-dimensional form to obtain multiple enhanced feature maps of different scales, namely the target feature map. Then, the information between the target feature maps of different scales is fused, and the feature maps from different levels are combined to generate a multi-scale feature map containing richer information. Finally, the fused multi-scale feature map is stretched into a one-dimensional vector sequence to form a house type feature sequence.

[0078] Through intra-scale feature interaction and cross-scale feature fusion, the model can reduce unnecessary calculations and lower computing costs while maintaining high accuracy. In addition, the optimized feature processing and fusion strategies enable the model to achieve efficient target detection with limited computing resources.

[0079] In one of the embodiments, the targeted apartment feature map is converted into a one-dimensional vector sequence; feature vectors at different positions in the one-dimensional vector sequence are subjected to feature interaction to obtain interactive feature vectors at different positions; the interactive feature vectors at different positions are subjected to nonlinear transformation to obtain nonlinear feature vectors at different positions, and the nonlinear feature vectors at different positions constitute a target vector sequence; the target vector sequence is restored to obtain a target feature map.

[0080] Among them, the one-dimensional vector sequence refers to stretching the two-dimensional feature map into a one-dimensional vector sequence. This is done to facilitate feature interaction processing, because one-dimensional vectors are easier to perform matrix operations and attention mechanism calculations.

[0081] The interactive feature vector refers to the feature vector obtained after processing through feature interaction (such as multi-head self-attention mechanism) in a one-dimensional vector sequence. These feature vectors contain more information about the relationship between different positions and enhance the expression ability of local information.

[0082] Nonlinear feature vectors are feature vectors that have been processed by nonlinear transformations (such as feedforward networks). These feature vectors further enhance the expressive power of features, allowing the model to better capture complex patterns and relationships.

[0083] The target vector sequence is a sequence of nonlinear feature vectors at different positions. After feature interaction and nonlinear transformation, these feature vectors already contain rich information and can better represent the local and global information in the feature map.

[0084] Specifically, first, the two-dimensional feature map, i.e., the apartment feature map, is in the shape of H×W×C, where H is the height, W is the width, and C is the number of channels. It is expanded in rows or columns and converted into a one-dimensional vector sequence in the shape of (H×W)×C. Then, the interactive feature vectors in the one-dimensional vector sequence are nonlinearly transformed through multiple layers of fully connected layers and activation functions (such as ReLU) to enhance the expressive power of the features, and the nonlinear feature vector is output, which is still in the shape of (H×W)×C. Next, all nonlinear feature vectors are arranged in order to form a target vector sequence. Finally, the one-dimensional target vector sequence is readjusted back to a two-dimensional form, restored to a feature map of H×W×C, and finally the target feature map is output in the shape of H×W×C.

[0085] The model can better capture information of different scales through feature interaction and nonlinear transformation, which improves the robustness and generalization ability of the model when dealing with complex scenes. It can be applied to various apartment feature maps, and can effectively extract and utilize key features through feature interaction and nonlinear transformation regardless of the size and layout of the apartment.

[0086] In one of the embodiments, the feature vector of each position in the apartment feature map is stretched to obtain multiple first one-dimensional vectors; the multiple first one-dimensional vectors are arranged according to the priority order corresponding to each of the multiple first one-dimensional vectors to obtain a one-dimensional vector sequence.

[0087] The first one-dimensional vector refers to a one-dimensional vector obtained by stretching the feature vector of each position in the apartment feature map. The feature vector of each position in the feature map is a multi-dimensional vector (usually a multi-channel vector), which is converted into a one-dimensional vector by stretching.

[0088] A one-dimensional vector sequence is a sequence formed by arranging multiple first one-dimensional vectors in a certain priority order. This sequence is an ordered collection of one-dimensional vectors, which facilitates subsequent feature interaction and nonlinear transformation processing.

[0089] Specifically, first, the feature vector (shape 1×1×C) of each position in the two-dimensional apartment feature map (shape H×W×C) is stretched into a one-dimensional vector, namely the first one-dimensional vector (shape C). Then, each first one-dimensional vector is arranged according to the priority order. It can be understood that the priority order can be determined according to factors such as the importance of the position in the feature map and the significance of the feature vector. Finally, multiple first one-dimensional vectors are arranged in order of priority to form a one-dimensional vector sequence with a shape of (H×W)×C.

[0090] Since the feature vector of each position is converted from multi-dimensional to one-dimensional, the form of one-dimensional vector sequence enables the model to process feature maps of different shapes and sizes more flexibly and adapt to input data of different scales and complexities. And further, according to the priority order of each first one-dimensional vector, an ordered one-dimensional vector sequence is obtained, which helps the model to locate and classify targets more accurately, reduce false detection and missed detection, and improve the accuracy of target detection.

[0091] In one of the embodiments, a plurality of target feature maps of different scales are sampled and aligned to obtain a plurality of aligned feature maps with the same resolution; the plurality of aligned feature maps are fused through multi-layer convolution operations to obtain a fused feature map; the feature vector at each position in the fused feature map is stretched to obtain a plurality of second one-dimensional vectors; the plurality of second one-dimensional vectors are arranged according to their respective corresponding priority orders to obtain a housing feature sequence.

[0092] The aligned feature map refers to the feature maps of different scales that are adjusted to the same resolution after the sampling alignment process. This process ensures that feature maps of different scales can be directly fused without causing information loss or mismatch due to different resolutions.

[0093] The fused feature map is a new feature map generated by fusing the information of these feature maps through a series of convolution operations on the basis of the aligned feature maps. This feature map contains comprehensive information extracted from feature maps of different scales, which helps to improve the performance of subsequent tasks (such as classification, detection, etc.).

[0094] The second one-dimensional vector refers to the one-dimensional vector obtained by stretching the feature vector at each position in the fused feature map. It can be understood that the "second" here means that this is the second vectorization operation performed in the entire process (the first is the process from the original feature map to the target feature map). These vectors can be regarded as feature representations at each spatial position, and they are organized into a sequence for subsequent processing.

[0095] Specifically, each target feature map is first sampled, and an upsampling or downsampling method is used to ensure that all feature maps reach a preset uniform resolution, thereby obtaining multiple aligned feature maps with the same resolution. Optionally, the sampling method can be selected as nearest neighbor interpolation, bilinear interpolation, or bicubic interpolation, which can be set according to actual needs.

[0096] Then the aligned feature map is used as input and sent to the multi-layer convolutional network for processing to generate a fused feature map, which integrates multi-scale information and has stronger expression ability. Optionally, the multi-layer convolutional network contains multiple types of convolutional layers, such as ordinary convolutional layers, depth-separable convolutional layers, residual blocks, etc., which can be set according to actual needs. In addition, activation functions and normalization layers are added between convolutional layers to enhance the learning and generalization capabilities of the model.

[0097] Next, traverse each spatial position of the fused feature map and obtain the feature vector at that position. Each feature vector is stretched from its original multi-dimensional form into a one-dimensional vector to obtain multiple second one-dimensional vectors, each of which represents the feature representation of a position in the fused feature map. For example, if the original feature vector is C-dimensional, the length of the stretched vector is C.

[0098] Finally, the priority order of each second one-dimensional vector is determined. The priority can be set based on different criteria, such as the importance of the spatial location, the significance of the feature, etc. All second one-dimensional vectors are sorted in order of priority to form an ordered vector sequence. Finally, the apartment feature sequence is obtained, which not only retains the spatial information, but also reflects the importance of the features of each location.

[0099] By sampling, aligning and fusing the target feature maps of different scales, feature information from different scales can be effectively integrated. This helps capture the details and overall structure of the target at different scales, which plays an important role in improving the recognition accuracy and robustness of the model. The feature vectors in the fused feature map are stretched and converted into one-dimensional vectors for further processing and utilization. This process simplifies the representation of features, reduces the complexity of subsequent processing, and maintains the validity of the features. And arrange them according to the priority order of each one-dimensional vector to form an orderly sequence of apartment features. This not only highlights the key features, but also provides a basis for subsequent feature selection, dimensionality reduction and other operations, which helps to improve the efficiency and performance of the model.

[0100] In one of the embodiments, for each of a plurality of objects to be identified, a confidence score of an identification box corresponding to the object to be identified in the floor plan is determined based on a preset number of floor plan feature vectors of the object to be identified, the confidence score being used to characterize the probability that a target position of the identification box corresponding to the object to be identified in the floor plan coincides with an actual position of the object to be identified in the floor plan; and the confidence scores corresponding to each of the plurality of identification boxes are displayed in the floor plan.

[0101] The confidence score refers to the probability that the recognition box predicted by the model coincides with the actual target location. It is an indicator for evaluating the accuracy of the recognition box, usually a value between 0 and 1. The larger the value, the higher the model's confidence in the recognition box. The actual location refers to the actual location of the object to be identified in the floor plan. This is the precise location information obtained through annotated data or real measurements.

[0102] Specifically, each object to be identified first corresponds to a preset number of apartment feature vectors, and the feature vectors are processed using a decoder and an auxiliary prediction head (such as a multi-layer attention mechanism and a feedforward network). Generate an identification box and a confidence score, and the confidence score is used to characterize the probability of overlap between the identification box and the actual position. Then, each identification box is marked in the floor plan, and the confidence score corresponding to each identification box is displayed near the identification box. Finally, the visualization result of the identification box with the confidence score in the floor plan is output.

[0103] Optionally, a method for calculating the confidence score of each recognition box includes:

[0104] Step 1: Use the trained encoder to extract feature sequences from the floor plan.

[0105] Step 2: For each object to be identified, select a preset number of apartment feature vectors from the feature sequence.

[0106] Step 3: Use the trained decoder and auxiliary prediction head to generate recognition boxes and confidence scores.

[0107] Step 4: Set a confidence threshold (for example, 0.5) to filter out the recognition boxes whose confidence scores are lower than the threshold.

[0108] Step 5: Display the remaining recognition boxes and their corresponding confidence scores on the floor plan.

[0109] By calculating the confidence score of each recognition box, the model can evaluate the degree of overlap between the recognition box and the actual position. Recognition boxes with high confidence scores are more likely to be accurate, thereby improving the accuracy of detection. And by selecting the feature vector most relevant to the object to be identified, the model can more accurately locate and identify the target, and maintain high detection performance even in complex backgrounds or occlusions.

[0110] In one embodiment, a plurality of objects to be identified in a floor plan are identified by a floor plan structured detection model to generate a plurality of identification boxes. The floor plan is composed of a backbone network, a hybrid encoder, and a transformer decoder with an auxiliary prediction head.

[0111] In one embodiment, reference Figure 3 ,The training method of the floor plan structured detection model includes:

[0112] Feature extraction is performed on the floor plan to obtain feature maps output by each of the multiple processing layers. Optionally, a SwinTransformer is used as the backbone network to extract multi-scale features from the input floor plan. The last three stages of the backbone network (S3, S4, S5) generate feature maps of different scales, which correspond to different levels of image information, such as low-level feature maps (S3) capturing small structures (such as doors and windows), and high-level feature maps (S5) capturing larger structures (such as room outlines).

[0113] According to the true label box of the floor plan and the feature maps output by each of the multiple processing layers, an image feature sequence is generated, and multiple image features are selected from the image feature sequence as the initial query object. Optionally, the feature maps of the same scale are first interacted through the AIFI (Attention-based Intra-scale Feature Interaction) module to enhance the expression of local information. Through the multi-head self-attention mechanism MHSA (Multi-Head Self-Attention) or the deformable attention mechanism (Deformable Attention) and the feed-forward network FFN (Feed-Forward Network), the feature map is stretched into a one-dimensional vector for interactive processing and then restored to a two-dimensional form. Then, through the cross-scale feature fusion CCFM (Cross-scale Feature Fusion Module) module, feature maps of different scales are fused to obtain more comprehensive image information. By combining feature maps from different levels, the model can simultaneously consider information at different levels of abstraction, thereby improving the accuracy of detection. Then, through the IoU (Intersection over Union) perception query selection mechanism, a part of the most representative features is selected from the multi-scale feature map output by the encoder. These features are features with a high IoU with the ground truth boxes. Specifically, IoU=A\B. Among them, IoU refers to an indicator used to measure the degree of overlap between two bounding boxes (or areas), A is the overlapping area of ​​the two bounding boxes, and B is the union area of ​​the two bounding boxes. It can be understood that the value range of IoU is between 0 and 1. The closer the value is to 1, the higher the overlap of the two bounding boxes, and the more accurate the detection result of the model. In the target detection task in this application, IoU is often used to evaluate the degree of match between the predicted box generated by the model and the real label box. Stretch the selected features into a one-dimensional vector to form a sequence. The specific method is to arrange the feature vectors (usually multi-channel) at each position in the feature map in a certain order to form a sequence, which can be used as input for subsequent processing. Finally, from the generated image feature sequence, a fixed number of image features are selected as initial query objects (ObjectQueries) through the IoU-aware query selection mechanism. These query objects are preliminary estimates of the target objects and will be further optimized in the decoder.

[0114] According to the initial query object, a first detection box and a first confidence score are generated, and the first confidence score is used to characterize the proximity between the first detection box and the true label box. Optionally, in the encoder stage, a fixed number of features are selected from the feature sequence output by the encoder as the initial query objects (ObjectQueries) through the IoU-aware query selection mechanism. These initial query objects are preliminary estimates of potential target objects. The decoder generates a prediction box (first detection box) and a confidence score (first confidence score) based on the current initial query object.

[0115] According to the first confidence score, the initial query object is updated to obtain an updated query object. Optionally, in each iteration, the decoder generates a new bounding box and confidence score based on the current query object. These newly generated bounding boxes and confidence scores will be fed back to the decoder, and the decoder will adjust the position and confidence of the query object based on this information to make it closer to the real target object. Specifically, the decoder updates the position and confidence of the query object based on the first confidence score. If the confidence score of a query object is high, it means that the position and shape of the query object are already close to the real label box, and the decoder will fine-tune the query object; if the confidence score is low, the decoder will significantly adjust the position and shape of the query object to better match the real label box.

[0116] According to the updated query object, a second detection box and a second confidence score are generated. If the second confidence score is greater than the confidence threshold, the second detection box is used as the final detection box. Optionally, after multiple iterations of optimization, the query object generated by the decoder is already very close to the real target object. At this time, the decoder generates a new detection box (second detection box) and a new confidence score (second confidence score) based on the updated query object. The generated second confidence score is used to characterize the closeness of the second detection box to the real label box. If the second confidence score is greater than the preset confidence threshold, the detection box is considered to be reliable and can be output as the final detection box. After multiple iterations of optimization, the bounding box (second detection box) and confidence score (second confidence score) generated by the decoder will be output as the final detection result. These results represent the model's best estimate of the target object in the image.

[0117] In an exemplary embodiment, Figure 4 As shown, it includes steps 402 to 410. Among them:

[0118] Step 402, obtaining a floor plan, the floor plan including a plurality of objects to be identified;

[0119] Step 404, extracting apartment feature maps of different scales from the floor plan, stretching the feature vector of each position in the apartment feature map to obtain multiple first one-dimensional vectors; arranging the multiple first one-dimensional vectors according to the priority order corresponding to each of the multiple first one-dimensional vectors to obtain a one-dimensional vector sequence; performing feature interaction on the feature vectors at different positions in the one-dimensional vector sequence to obtain interactive feature vectors at different positions; performing nonlinear transformation on the interactive feature vectors at different positions to obtain nonlinear feature vectors at different positions, and the nonlinear feature vectors at different positions constitute a target vector sequence; restoring the target vector sequence to obtain a target feature map; performing sampling alignment on multiple target feature maps of different scales to obtain multiple aligned feature maps with the same resolution; fusing the multiple aligned feature maps through multi-layer convolution operations to obtain a fused feature map; stretching the feature vector of each position in the fused feature map to obtain multiple second one-dimensional vectors; arranging the multiple second one-dimensional vectors according to the priority order corresponding to each of the multiple second one-dimensional vectors to obtain an apartment feature sequence, wherein the apartment feature sequence includes multiple apartment feature vectors;

[0120] Step 406, for each object to be identified, selecting a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0121] Step 408, based on a preset number of apartment feature vectors of each of the multiple objects to be identified, generate multiple identification boxes in the floor plan, the identification boxes are used to mark the target positions and target categories of the objects to be identified in the floor plan;

[0122] Step 410, for each of the multiple objects to be identified, based on a preset number of apartment feature vectors of the object to be identified, determine the confidence score of the identification box corresponding to the object to be identified in the floor plan, the confidence score is used to characterize the probability that the target position of the identification box corresponding to the object to be identified in the floor plan coincides with the actual position of the object to be identified in the floor plan; display the confidence scores corresponding to each of the multiple identification boxes in the floor plan.

[0123] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0124] Based on the same inventive concept, the embodiment of the present application also provides a plan floor plan structured detection device for implementing the plan floor plan structured detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more plan floor plan structured detection device embodiments provided below can refer to the limitations of the plan floor plan structured detection method above, and will not be repeated here.

[0125] In an exemplary embodiment, Figure 5 As shown, a floor plan structured detection device 500 is provided, comprising: an acquisition module 502, a first generation module 504, a selection module 506 and a second generation module 508, wherein:

[0126] An acquisition module 502 is used to acquire a floor plan, where the floor plan includes a plurality of objects to be identified;

[0127] A first generating module 504 is used to extract apartment feature graphs of different scales from the floor plan, and generate an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors;

[0128] A selection module 506 is used to select, for each object to be identified, a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence;

[0129] The second generating module 508 is used to generate multiple identification boxes in the floor plan based on a preset number of apartment feature vectors of each of the multiple objects to be identified, and the identification boxes are used to mark the target positions and target categories of the objects to be identified in the floor plan.

[0130] In one embodiment, the first generating module 504 is used to perform intra-scale feature interaction on each of the apartment feature maps of multiple different scales to obtain a target feature map; and perform cross-scale feature fusion processing on the target feature maps of multiple different scales to obtain an apartment feature sequence.

[0131] In one embodiment, the first generation module 504 is used to convert the targeted apartment feature map into a one-dimensional vector sequence; perform feature interaction on feature vectors at different positions in the one-dimensional vector sequence to obtain interactive feature vectors at different positions; perform nonlinear transformation on the interactive feature vectors at different positions to obtain nonlinear feature vectors at different positions, and the nonlinear feature vectors at different positions constitute a target vector sequence; and restore the target vector sequence to obtain a target feature map.

[0132] In one embodiment, the first generation module 504 is used to stretch the feature vector of each position in the apartment feature map to obtain multiple first one-dimensional vectors; and arrange the multiple first one-dimensional vectors according to the priority order corresponding to each of the multiple first one-dimensional vectors to obtain a one-dimensional vector sequence.

[0133] In one embodiment, the first generation module 504 is used to sample and align multiple target feature maps of different scales to obtain multiple aligned feature maps with the same resolution; fuse the multiple aligned feature maps through multi-layer convolution operations to obtain a fused feature map; stretch the feature vector of each position in the fused feature map to obtain multiple second one-dimensional vectors; arrange the multiple second one-dimensional vectors according to the priority order corresponding to each of the multiple second one-dimensional vectors to obtain a house type feature sequence.

[0134] In one of the embodiments, the floor plan structured detection device also includes a determination module 510, which is used to determine, for each of a plurality of objects to be identified, a confidence score of an identification box corresponding to the object to be identified in the floor plan based on a preset number of floor plan feature vectors of the object to be identified, the confidence score being used to characterize the probability that a target position of the identification box corresponding to the object to be identified in the floor plan coincides with an actual position of the object to be identified in the floor plan; and the confidence scores corresponding to each of the plurality of identification boxes are displayed in the floor plan.

[0135] In another embodiment, if Figure 6 As shown, Figure 6The structural block diagram of the structured detection device of the floor plan in another embodiment includes: an acquisition module 502, a first generation module 504, a selection module 506 and a second generation module 508. The floor plan structured detection device 500 also includes a determination module 510, which is used to determine the confidence score of the identification box corresponding to the object to be identified in the floor plan for each of the multiple objects to be identified based on a preset number of floor plan feature vectors of the object to be identified, the confidence score is used to characterize the probability that the target position of the identification box corresponding to the object to be identified in the floor plan coincides with the actual position of the object to be identified in the floor plan; and the confidence scores corresponding to the multiple identification boxes are displayed in the floor plan.

[0136] Each module in the above-mentioned floor plan structured detection device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0137] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data related to the structural detection of the floor plan. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for structural detection of a floor plan is implemented.

[0138] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0139] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0140] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0141] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0142] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0143] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0144] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0145] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for detecting a structured floor plan, characterized in that: The method comprises: Acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified; Extracting apartment feature graphs of different scales from the floor plan, and generating an apartment feature sequence based on the apartment feature graphs of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors; For each object to be identified, selecting a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence; Based on a preset number of apartment feature vectors of each of the multiple objects to be identified, multiple identification boxes are generated in the floor plan, and the identification boxes are used to mark the target positions of the objects to be identified and the target categories of the objects to be identified in the floor plan.

2. The method according to claim 1, characterized in that The generating of the apartment type feature sequence based on the apartment type feature graphs of different scales includes: For each of the multiple apartment feature maps of different scales, perform intra-scale feature interaction on the apartment feature map to obtain a target feature map; The target feature maps of multiple scales are subjected to cross-scale feature fusion processing to obtain the apartment feature sequence.

3. The method according to claim 2, characterized in that The step of performing in-scale feature interaction on the targeted apartment feature map to obtain a target feature map includes: Convert the targeted apartment feature map into a one-dimensional vector sequence; Performing feature interaction on feature vectors at different positions in the one-dimensional vector sequence to obtain interactive feature vectors at different positions; Performing nonlinear transformation on the interacted feature vectors at different positions to obtain nonlinear feature vectors at different positions, wherein the nonlinear feature vectors at different positions constitute a target vector sequence; The target vector sequence is restored to obtain a target feature map.

4. The method according to claim 3, characterized in that The step of converting the targeted apartment type feature map into a one-dimensional vector sequence includes: Stretching the feature vector of each position in the targeted apartment feature map to obtain a plurality of first one-dimensional vectors; The multiple first one-dimensional vectors are arranged according to the priority order corresponding to each of the multiple first one-dimensional vectors to obtain a one-dimensional vector sequence.

5. The method according to claim 2, characterized in that: The cross-scale feature fusion processing of target feature maps of multiple different scales is performed to obtain a house type feature sequence, including: Perform sampling and alignment processing on multiple target feature maps of different scales to obtain multiple aligned feature maps with the same resolution; Through multi-layer convolution operations, a plurality of the aligned feature maps are fused to obtain a fused feature map; Stretching the feature vector at each position in the fused feature map to obtain a plurality of second one-dimensional vectors; The multiple second one-dimensional vectors are arranged according to the priority orders corresponding to the multiple second one-dimensional vectors to obtain a house type feature sequence.

6. The method according to claim 1, characterized in that The method further comprises: For each of the multiple objects to be identified, based on a preset number of apartment feature vectors of the object to be identified, determine a confidence score of an identification box corresponding to the object to be identified in the floor plan, wherein the confidence score is used to represent a probability that a target position of the identification box corresponding to the object to be identified in the floor plan coincides with an actual position of the object to be identified in the floor plan; The confidence scores corresponding to the multiple identification boxes are displayed in the floor plan.

7. A structured detection device for a floor plan, characterized in that: The device comprises: An acquisition module, used to acquire a floor plan, wherein the floor plan includes a plurality of objects to be identified; A first generating module is used to extract apartment feature diagrams of different scales from the floor plan, and generate an apartment feature sequence based on the apartment feature diagrams of different scales, wherein the apartment feature sequence includes a plurality of apartment feature vectors; A selection module, for selecting, for each object to be identified, a preset number of apartment feature vectors corresponding to the object to be identified from the apartment feature sequence; The second generating module is used to generate a plurality of identification boxes in the floor plan diagram based on a preset number of apartment feature vectors of each of the plurality of objects to be identified, wherein the identification boxes are used to mark the target positions of the objects to be identified in the floor plan diagram and the target categories of the objects to be identified.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Citation Information

Patent Citations

  • House type image recognition method and device based on target detection and computer equipment

    CN111310755A

  • Home decoration design method and device, electronic device and storage medium

    CN111680421A

  • Pedestrian detection method and device and electronic equipment

    CN118072350A

  • Lightweight road defect detection method based on dynamic deformable attention mechanism

    CN118279562A

  • Kitchen garbage target detection method based on active learning selection strategy

    CN118429625A