A decoration quotation automatic generation system

Abstract

This invention provides an automated renovation quotation generation system, belonging to the field of automated generation technology. It comprises six core modules: a multimodal input module, an intelligent parsing module, a scheme matching module, a process and material database, a quotation calculation module, and an output display module. These modules work collaboratively to automate the entire quotation process. It overcomes the limitations of existing systems that can only parse geometric information from CAD drawings, employing a multimodal input mode combining renovation renderings and CAD floor plans. Combined with improved YOLOv8 and U-Net deep learning algorithms, it achieves accurate identification and attribute extraction of renovation elements. Through the collaborative work of these two algorithms, it solves the problems of low accuracy in recognizing small-sized renovation elements and inaccurate boundary segmentation. Simultaneously, it extracts visual attributes such as material, color, and specifications of renovation elements, addressing the pain point of difficult understanding for customers.

Description

Technical Field

[0001] This invention belongs to the field of automation equipment technology, and specifically relates to an automated system for generating decoration quotes. Background Technology

[0002] As a core downstream supporting industry of real estate, the home renovation industry has seen continuous growth in market demand due to rising living standards and consumption upgrades. However, pain points in the industry's pricing process are becoming increasingly prominent. Currently, home renovation quotes mainly rely on manual operations by cost estimators. The core process involves cost estimators reviewing CAD drawings, manually measuring house dimensions, calculating various renovation quantities, and then combining their experience to determine renovation materials and construction techniques. Finally, a quote is generated manually. This traditional model has many unresolved problems in practical application, severely restricting industry efficiency and service quality.

[0003] First, inefficiency is a significant issue. A complete home renovation quote typically takes 2-3 hours to complete. For renovation companies, this excessively long quote cycle reduces customer responsiveness and can even lead to customer loss. Second, manual processing is prone to omissions and errors. The industry average quote error is as high as 15%-25%, which can result in losses for renovation companies due to excessively low quotes, or a loss of customer trust due to excessively high quotes. Third, most home renovation customers lack professional renovation knowledge, cannot understand complex CAD drawings, and struggle to correlate quote details with the actual renovation effect, leading to high communication costs and difficulty in building customer trust. Finally, differences in the professional experience and operating habits of different cost estimators can result in significantly different quotes for the same set of renovation drawings. The lack of unified industry standards hinders the standardized development of the industry.

[0004] In existing technologies, some companies have launched automatic quotation systems based on CAD drawings. However, their core limitation lies in their ability to only parse geometric dimensions in CAD drawings, failing to recognize visual elements such as materials, shapes, and colors in interior design renderings. This prevents them from providing a "what you see is what you get" quotation experience—customers still need to establish a connection between abstract quotation details and the final design, making it difficult to intuitively understand the reasonableness of the price. Communication costs and trust issues remain unresolved. Therefore, developing an automated system that integrates interior design renderings and CAD drawings, enabling visual element recognition, automatic quantity surveying, accurate quotation calculation, and visual display, is crucial for addressing industry pain points and driving the industry's intelligent upgrade. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an automated decoration quotation generation system. It adopts a layered architecture design, which is divided into three major layers from top to bottom: interface layer, business logic layer and data layer. It includes six core modules: multimodal input module, intelligent parsing module, scheme matching module, process and material database, quotation calculation module and output display module. Each module works together to realize the automation of the entire quotation process. Multimodal input module: As the system's input entry point, it is responsible for receiving decoration renderings and CAD floor plans uploaded by users, supporting multiple formats, and completing file preprocessing; Intelligent parsing module: The core processing module of the system, responsible for parsing the input multimodal data and extracting key information such as decoration elements, dimensions, and structure; The scheme matching module matches the optimal renovation scheme template from the preset scheme library based on the extracted renovation element information. Process and Materials Database: The core supporting module of the system, which stores various data such as materials, processes, and quotas, providing data support for quotation calculation; Quotation Calculation Module: By combining analytical information, matching schemes, and database data, it automatically calculates the project quantity and various costs, and generates a complete quotation. Output display module: Displays the quotation results to users in a visual way, supporting real-time adjustment and export in multiple formats.

[0006] Furthermore, the system data flow is as follows: users upload files through the multimodal input module, which are then preprocessed and passed to the intelligent parsing module; the intelligent parsing module extracts key information and passes it to the scheme matching module and the quotation calculation module respectively; the scheme matching module matches the optimal scheme from the preset scheme library in the process and materials database and passes the scheme information to the quotation calculation module; the quotation calculation module calls the material, process, and quota data in the process and materials database to complete the quotation calculation; finally, the quotation result is displayed to the user through the output display module, which supports user adjustment and export.

[0007] Furthermore, the core function of the multimodal input module is to receive decoration renderings and CAD floor plans uploaded by users, support multiple mainstream formats, and preprocess the uploaded files to ensure that the subsequent parsing module can efficiently and accurately extract key information. Input format support: Considering the actual application scenarios in the decoration industry, the module supports uploading the following two types of files: Interior design renderings: Supports mainstream formats such as JPG, PNG, 3D Max, SketchUp, and Kujiale for rendering 3D images, covering commonly used rendering output formats in interior design to meet the usage habits of different designers; CAD floor plan: Supports mainstream CAD formats such as DWG and DXF, covering most drawing formats used by decoration companies, ensuring complete import of drawing information.

[0008] Further, file preprocessing is required. The uploaded files need to be preprocessed, and the specific steps are as follows: Image denoising: For interior design renderings, a Gaussian filtering algorithm is used to remove image noise and reduce the interference of noise on subsequent element recognition; Resolution Enhancement: For low-resolution renderings, a bilinear interpolation algorithm is used to enhance image resolution, ensuring that details of decorative elements are clearly discernible; CAD file format conversion: For different versions of DWG and DXF files, an open-source CAD parsing library is used to convert them into the standard DXF format, eliminating format differences; After preprocessing, the module transmits the standardized renderings and CAD files to the intelligent parsing module, providing a reliable data foundation for subsequent information extraction.

[0009] Furthermore, the core of analyzing interior design renderings is to identify various interior design elements and determine their material, color, and specification attributes. Traditional YOLOv8 object detection algorithms and U-Net semantic segmentation algorithms have shortcomings in interior design element recognition: YOLOv8 has low accuracy in recognizing small-sized elements, and U-Net is not accurate enough in segmenting the boundaries of interior design elements. Moreover, both are difficult to effectively extract the attribute information of elements. To address the low accuracy of YOLOv8 in recognizing small elements, this system optimizes its feature extraction network: a lightweight attention mechanism is added to the backbone to enhance the ability to extract features from small elements; a feature fusion module is used in the neck to combine shallow and deep features, improving the accuracy of element boundary localization; the improved YOLOv8 algorithm loss function is as follows:

[0010] in, For category loss, the cross-entropy loss function is used to determine the category of decoration elements; For bounding box loss, the CIoU loss function is adopted to improve the bounding box localization accuracy; The target confidence loss is used to determine whether the detected area is a decoration element; , The weighting coefficients are set to 1.5 and 1.0 respectively to balance various losses.

[0011] Furthermore, the improved YOLOv8 algorithm process is as follows: the preprocessed image is input into the network, image features are extracted through the backbone, enhanced by the attention mechanism, feature fusion is performed through the neck part, and finally the category, bounding box coordinates and confidence score of the decoration elements are output through the head part.

[0012] Furthermore, to address the issue of inaccurate boundary segmentation in U-Net, this system adds a boundary enhancement module to the decoding part of U-Net. By capturing the edge features of the image, it improves the boundary segmentation accuracy of decorative elements. Simultaneously, Dice loss is introduced into the loss function to resolve the pixel imbalance problem of decorative elements. The improved U-Net loss function is as follows:

[0013] in, The pixel values ​​predicted by the model. The actual label pixel value, This represents the total number of pixels in the image. To avoid constants with a denominator of 0; The improved U-Net algorithm is used for semantic segmentation of decoration elements, accurately dividing the regional range of different elements. It works in conjunction with the improved YOLOv8 algorithm: YOLOv8 is responsible for identifying the element category and approximate location, while U-Net is responsible for accurately segmenting the element boundaries. The combination of the two achieves accurate identification of decoration elements.

[0014] The beneficial effects of this invention include: This invention provides an automated renovation quotation generation system. It adopts a layered architecture, divided into three main layers from top to bottom: the interface layer, the business logic layer, and the data layer. It includes six core modules: a multimodal input module, an intelligent parsing module, a scheme matching module, a process and material database, a quotation calculation module, and an output display module. These modules work collaboratively to automate the entire quotation process. It overcomes the limitations of existing systems that can only parse geometric information from CAD drawings, employing a multimodal input mode of "renovation renderings + CAD floor plans." Combined with improved YOLOv8 and U-Net deep learning algorithms, it achieves accurate identification and attribute extraction of renovation elements. Through the collaborative work of these two algorithms, it solves the problems of low accuracy in recognizing small-sized renovation elements and inaccurate boundary segmentation. Simultaneously, it extracts visual attributes such as material, color, and specifications of renovation elements, realizing the association between "visual elements - size information - quotation details," truly achieving a "what you see is what you get" quotation experience and addressing the pain point of difficult customer comprehension. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating an automated renovation quotation generation system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an automated decoration quotation generation system according to an embodiment of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to... Used to explain the present invention, and not to be construed as limiting the present invention.

[0018] like Figure 1-2 As shown, the purpose of this invention is to overcome the shortcomings of existing technologies and provide an automated renovation quotation generation system. This automated renovation quotation generation system adopts a layered architecture design, divided into three main layers from top to bottom: the interface layer, the business logic layer, and the data layer. It includes six core modules: a multimodal input module, an intelligent parsing module, a scheme matching module, a process and material database, a quotation calculation module, and an output display module. These modules work collaboratively to achieve full automation of the quotation process. The functions of each module are as follows: Multimodal input module: As the system's input entry point, it is responsible for receiving decoration renderings and CAD floor plans uploaded by users, supporting multiple formats, and completing file preprocessing; Intelligent parsing module: The core processing module of the system, responsible for parsing the input multimodal data and extracting key information such as decoration elements, dimensions, and structure; The scheme matching module matches the optimal renovation scheme template from the preset scheme library based on the extracted renovation element information. Process and Materials Database: The core supporting module of the system, which stores various data such as materials, processes, and quotas, providing data support for quotation calculation; Quotation Calculation Module: By combining analytical information, matching schemes, and database data, it automatically calculates the project quantity and various costs, and generates a complete quotation. Output display module: Displays the quotation results to users in a visual way, supporting real-time adjustment and export in multiple formats.

[0019] The system data flow is as follows: Users upload files through the multimodal input module, which are then preprocessed and passed to the intelligent parsing module; the intelligent parsing module extracts key information and passes it to the scheme matching module and the quotation calculation module respectively; the scheme matching module matches the optimal scheme from the preset scheme library of the process and materials database and passes the scheme information to the quotation calculation module; the quotation calculation module calls the material, process, and quota data in the process and materials database to complete the quotation calculation; finally, the quotation result is displayed to the user through the output display module, which supports user adjustment and export.

[0020] The core function of the multimodal input module is to receive interior design renderings and CAD floor plans uploaded by users. It supports multiple mainstream formats and preprocesses the uploaded files to ensure that the subsequent parsing module can efficiently and accurately extract key information.

[0021] Input format support: Considering the actual application scenarios in the decoration industry, the module supports uploading the following two types of files: Interior design renderings: Supports mainstream formats such as JPG, PNG, 3D Max, SketchUp, and Kujiale for rendering 3D images, covering commonly used rendering output formats in interior design to meet the usage habits of different designers; CAD floor plan: Supports mainstream CAD formats such as DWG and DXF, covering most drawing formats used by decoration companies, ensuring complete import of drawing information.

[0022] File preprocessing is necessary because user-uploaded files may contain issues such as noise, insufficient resolution, and format incompatibility, which can affect the accuracy of subsequent parsing. The specific steps are as follows: Image denoising: For interior design renderings, a Gaussian filtering algorithm is used to remove image noise and reduce the interference of noise on subsequent element recognition.

[0023] Resolution Enhancement: For low-resolution renderings, a bilinear interpolation algorithm is used to enhance image resolution, ensuring that details of decorative elements are clearly discernible.

[0024] CAD file format conversion: For different versions of DWG and DXF files, an open-source CAD parsing library (such as LibreCAD) is used to convert them into the standard DXF format, eliminating format differences and ensuring accurate extraction of subsequent dimensions and structural information; at the same time, redundant information in CAD drawings (such as redundant annotations and auxiliary lines) is removed, while retaining core content such as walls, doors and windows, and dimension annotations.

[0025] After preprocessing, the module transmits the standardized renderings and CAD files to the intelligent parsing module, providing a reliable data foundation for subsequent information extraction.

[0026] The intelligent analysis module is responsible for analyzing the pre-processed interior design renderings and CAD floor plans, extracting key information such as design elements, dimensions, and structure to provide data support for subsequent scheme matching and quotation calculation. The module employs a fusion technology of "deep learning + CAD analysis," processing the two types of input data separately to achieve accurate information extraction.

[0027] The core of interpreting interior design renderings lies in identifying various decorative elements (floors, walls, ceilings, doors, windows, furniture, lighting fixtures, sanitary ware, etc.) and determining their material, color, specifications, and other attributes. Traditional YOLOv8 object detection algorithms and U-Net semantic segmentation algorithms have shortcomings in identifying decorative elements: YOLOv8 has low accuracy in recognizing small elements (such as lighting fixtures and hardware accessories), and U-Net's boundary segmentation of decorative elements is not precise enough; furthermore, both struggle to effectively extract element attribute information. Therefore, this system improves upon both algorithms to achieve accurate identification and attribute extraction of decorative elements.

[0028] To address the low accuracy of YOLOv8 in recognizing small elements, this system optimizes its feature extraction network: a lightweight attention mechanism (SE-Net) is added to the backbone to enhance the ability to extract features from small elements; a feature fusion module is used in the neck to combine shallow and deep features, improving the accuracy of element boundary localization. The improved YOLOv8 algorithm loss function is as follows:

[0029] in, For category loss, the cross-entropy loss function is used to determine the category of decoration elements (such as walls, floors, and lighting fixtures). For bounding box loss, the CIoU loss function is adopted to improve the bounding box localization accuracy; The target confidence loss is used to determine whether the detected area is a decoration element; , The weighting coefficients are set to 1.5 and 1.0 respectively to balance various losses.

[0030] The improved YOLOv8 algorithm process is as follows: the preprocessed image is input into the network, image features are extracted through the backbone, enhanced by the attention mechanism, feature fusion is performed through the neck part, and finally the category, bounding box coordinates and confidence score of the decoration elements are output through the head part.

[0031] To address the inaccuracy of boundary segmentation in U-Net, this system adds a boundary enhancement module to the decoding part of U-Net. By capturing the edge features of the image, it improves the boundary segmentation accuracy of decorative elements. Simultaneously, Dice loss is introduced into the loss function to resolve the pixel imbalance problem of decorative elements (e.g., high pixel ratio of walls and low pixel ratio of lighting fixtures). The improved U-Net loss function is as follows:

[0032] in, The pixel values ​​predicted by the model. The actual label pixel value, This represents the total number of pixels in the image. To avoid constants with a denominator of 0 (set to 1e-5).

[0033] The improved U-Net algorithm is used for semantic segmentation of decoration elements, accurately dividing the regional range of different elements. It works in conjunction with the improved YOLOv8 algorithm: YOLOv8 is responsible for identifying the element category and approximate location, while U-Net is responsible for accurately segmenting the element boundaries. The combination of the two achieves accurate identification of decoration elements.

[0034] For interior design element attribute extraction, after identifying the category and area of ​​the interior design elements, image feature extraction algorithms are used to extract attributes such as material, color, and specifications of the elements. Color attributes: The color features of the elements are extracted using the HSV color space, and the average H (hue), S (saturation), and V (brightness) values ​​within the area are calculated to determine the element color (such as white walls and gray floors). Material properties: The texture features of the elements are extracted using texture feature extraction algorithms (such as LBP algorithm), and combined with a preset material texture library, the material of the elements is determined by similarity matching (such as latex paint wall, solid wood floor, tile floor). Specifications and attributes: Combine the dimensional information in the CAD drawings to calculate the actual dimensions of the elements (such as the length and width of doors and windows, and the specifications of the floor), and determine the element specifications (such as 800*800mm tiles, 1.2m wide doors and windows).

[0035] The core of CAD floor plan parsing is extracting structural information, dimensions, wall locations, door and window openings, and other geometric data to provide accurate dimensional data for quantity surveying. This system uses DWG parsing technology, combined with regular expressions and geometric calculations, to automatically extract information from CAD drawings. The specific process is as follows: Drawing layer analysis: CAD drawings are drawn in layers, with different layers corresponding to different architectural elements (such as wall layer, door and window layer, dimension annotation layer). The system first reads the layer information of the drawing, separates the core layer, and ignores redundant layers (such as auxiliary line layer, annotation layer). Dimensioning Extraction: Regular expressions are used to identify dimension text in drawings (such as "3000mm" and "2500mm"), and combined with the position information of dimension lines, the dimensions of various parts (such as wall length, room length and width, and door and window opening dimensions) are determined. The dimension extraction formula is: Structural information extraction: By analyzing the lines and graphics in CAD drawings, structural elements such as walls, doors and windows, beams and columns are identified, and information such as the location and thickness of walls, the location and number of doors and windows, and the size and location of beams and columns are determined. Data validation: Validate the extracted dimensions and structural information, such as checking the consistency between the room's length and width dimensions and the wall length, and the matching of door and window opening dimensions with the wall thickness, to ensure the accuracy of the data.

[0036] The intelligent analysis module ultimately outputs decoration element information (category, attribute, area range) and CAD structural dimension information, providing core data for scheme matching and quotation calculation.

[0037] The core function of the scheme matching module is to match the most similar renovation scheme template from a preset scheme library based on the renovation element information extracted by the intelligent parsing module, providing a basic framework for quotation calculation. The module uses a cosine similarity algorithm to achieve intelligent scheme matching, ensuring the rationality and applicability of the matching results.

[0038] The preset design library forms the basis for design matching, storing renovation plans of different styles and price ranges. Each plan includes a detailed materials list, workmanship specifications, and a quotation template. The library is designed according to the principles of "standardization, diversification, and updatableness," and its specific contents are as follows: Style Classification: Covering mainstream decoration styles such as modern minimalist, Nordic, Chinese, American, and light luxury, each style has multiple design templates to meet the aesthetic needs of different customers; Tier Classification: Each style is divided into three tiers: economy, comfort, and luxury. Different tiers correspond to different material standards, craftsmanship levels, and price levels to suit the budget needs of different customers. Scheme Content: Each scheme template includes room zoning (living room, bedroom, kitchen, bathroom, etc.), a list of decoration elements (such as the paint used for walls and the flooring used for floors), material details, process descriptions, quotation templates, and other information. It is also linked to the data in the process and material database to ensure the consistency of quotation calculations.

[0039] The solution library supports both manual and automatic updates: manual updates allow administrators to add new solution templates and modify existing solutions; automatic updates analyze user data and combine it with industry trends to regularly update the solution library content, ensuring the timeliness of the solutions.

[0040] The cosine similarity matching algorithm, the core of which is to calculate the similarity between the feature vector of the user's decoration elements and the feature vector of the preset scheme, is implemented using the cosine similarity algorithm. The specific steps are as follows: Feature vector construction: The decoration element information (category, material, color, specifications) extracted by intelligent parsing is quantified to construct the user's decoration feature vector. ,in The quantification value of the i-th decoration element is given (e.g., material quantification: latex paint = 1, wallpaper = 2, tile = 3; color quantification: white = 1, gray = 2, brown = 3); simultaneously, for each scheme template in the preset scheme library, a corresponding scheme feature vector is constructed. (j is the scheme number); Similarity calculation: The cosine similarity algorithm is used to calculate the user feature vector X and the feature vector of each scheme. The similarity is calculated using the following formula: Solution selection: Based on similarity scores, 3-5 of the most similar solution templates are selected and displayed to the user for selection. Users can make adjustments to the selected solution templates according to their own needs (such as changing materials or modifying processes) to further meet their actual needs.

[0041] To improve matching accuracy, this system introduces weighting coefficients on the basis of the cosine similarity algorithm: key elements that affect decoration style and pricing (such as wall material, floor material, and ceiling design) are given higher weights, while secondary elements (such as lighting style and small hardware) are given lower weights. The weighting coefficients are determined by the analytic hierarchy process to ensure that the matching results are more in line with the actual needs of users.

[0042] The materials and processes database is the core support of the system, storing data on the specifications, unit prices, and wastage rates of various decoration materials, as well as labor costs and machinery expenses for various construction processes, providing accurate and comprehensive data support for quotation calculations. The database is designed using a relational database (such as MySQL) and is divided into three sub-databases: a materials database, a processes database, and a quota database. These sub-databases are interconnected to ensure data consistency and integrity.

[0043] The materials inventory contains real-time quotes from over 200 mainstream home improvement material brands, covering commonly used building materials such as wall paint, floor tiles, hardware accessories, lighting fixtures and appliances, and bathroom fixtures. Each material comes with complete parameter information, as detailed below: Basic information: material name, brand, specifications, model, place of origin; Price information: Unit price (RMB / ㎡, RMB / meter, RMB / unit, etc.), transportation cost, installation cost; Loss information: Loss rate (determined based on material type and construction process, such as 3%-5% for ceramic tiles and 2%-3% for paint); Related information: corresponding construction process (e.g., latex paint walls correspond to the process of "wall base treatment + latex paint application"), applicable scenarios (e.g., bathroom tiles are suitable for humid environments).

[0044] The material library supports real-time updates. By connecting to the quotation interfaces of mainstream building material suppliers, it regularly updates the unit prices of materials to ensure the timeliness of quotations. At the same time, it supports administrators to manually add and modify material information to adapt to the application needs of new building materials.

[0045] The process inventory stores standard procedures, labor hours, and machinery costs for various decoration and construction processes. It supports adjustments based on region, trade, and skill level to ensure accurate and regionally applicable process cost calculations. Details are as follows: Basic process information: process name, applicable scenarios, construction process (e.g., wall latex paint process: base treatment → putty application → sanding → primer application → topcoat application); Cost information: labor hours (hours / ㎡, hours / meter), labor unit price (yuan / hour, divided by region and skill level, such as 80 yuan / hour for senior technicians and 50 yuan / hour for ordinary technicians in first-tier cities), and machinery costs (such as the usage cost of grinding machines and spraying machines). Adjust parameters: Supports adjusting the unit price of labor by region (such as Beijing, Shanghai, Guangzhou, etc.) and adjusting the working hours by type of work (such as plumber, electrician, bricklayer, carpenter) to ensure that the process cost is in line with the local market conditions.

[0046] The quota library includes a template conforming to the national standard "Specification for Measurement and Pricing of Construction Engineering Quantities" (GB50500), as well as local quota standards from various provinces and cities, providing a standardized basis for quotation calculation. Details are as follows: National standard quota: In accordance with the "Construction Engineering Quantity List Pricing Specification" (GB50500), the calculation rules for the quantity of each sub-item of the project and the cost composition (labor, materials, machinery, management fees, and profit) are clearly defined. Local quota standards: Includes local quotas from various provinces and cities (such as Beijing Construction Engineering Budget Quotas and Guangdong Construction and Decoration Engineering Comprehensive Quotas), allowing users to select the corresponding quota standards based on the project location; Fee rates: Built-in standardized management fee rates (divided by project type, such as home decoration project management fee rate of 8%-12%), profit rate (5%-8%), and tax rate (9%), which can be adjusted by users according to actual situation.

[0047] The three sub-databases of the process materials database are interconnected: the materials in the materials database correspond to the processes in the process database, and the processes in the process database correspond to the quota standards in the quota database. This ensures that relevant data can be quickly retrieved during quotation calculation, improving calculation efficiency and accuracy.

[0048] The quotation calculation module is the core functional module of the system. Combining information extracted by the intelligent analysis module, scheme templates matched by the scheme matching module, and data from the materials database, it automatically calculates the quantities and costs of various projects, generating a complete renovation quotation. The module uses preset algorithms to calculate the confidence level of renovation element recognition, quantities, comprehensive unit price, and scheme similarity, ensuring accurate and standardized quotations.

[0049] Quantity calculation is the foundation of quotation calculation. Based on the area range and CAD dimension information of the decoration elements extracted by the intelligent analysis module, and combined with the type of decoration element, the corresponding calculation formula is used to calculate the quantity of work. The core calculation formula is as follows: Wall work quantity:

[0050] Ground work volume:

[0051] Ceiling installation work quantity:

[0052] Door and window work quantity: calculated per unit. (N is the number of doors and windows), and calculate the sealing and edging work of the door and window openings by length (m).

[0053] After the quantity of work is calculated, the total amount of materials used is calculated by taking into account the loss rate in the material inventory: ,in This is the material loss rate (extracted from the material library).

[0054] The comprehensive unit price is the core of the quotation calculation, encompassing material costs, labor costs, machinery costs, management fees, and profit. The calculation formula is as follows:

[0055] Material costs : ,in This is the unit price of the material (extracted from the material library). For material transportation costs; Labor costs : ,in For man-hours (extracted from the process library). This refers to the unit price of labor. Machinery costs Based on the machinery required for the construction process, extract the machinery usage fee from the process library and calculate it according to the construction duration; Management fee : ,in Management fee rate (extracted from the quota database); profit : ,in Profit rate (extracted from the quota library).

[0056] The total price is the sum of the costs of each individual item of the project, plus taxes. The calculation formula is as follows:

[0057] in, For the quantity of sub-items of the project, Let be the comprehensive unit price of the i-th item. Let i be the quantity of work for the i-th project. Tax rate (extracted from the quota database, default 9%).

[0058] Meanwhile, the module introduces a confidence verification mechanism for the recognition of decoration elements: it calculates the confidence level of the decoration elements recognized by the intelligent parsing module. (Improving the confidence value output by the YOLOv8 algorithm), if If the error occurs, the user is prompted to manually confirm the category and attributes of the element to avoid incorrect pricing due to recognition errors, thereby further improving the accuracy of the pricing.

[0059] The core function of the output display module is to present the quotation generated by the quotation calculation module to the user in a visual way. It supports users to adjust materials and processes online, update quotations in real time, and export in multiple formats, making it convenient for users to save and share.

[0060] The quotation is presented visually in a layered manner, categorized by room (living room, bedroom, kitchen, bathroom, etc.) and construction stage (foundation engineering, main structure engineering, decoration engineering, installation engineering). Specific details are as follows: List of Items of Work: Displays the name, quantity, unit price, and total price of each room and each construction stage; Materials list: Displays the name, brand, specifications, quantity, unit price, and total price of all materials used in the project; Expense details: Shows the amount and percentage of various expenses such as labor costs, material costs, machinery costs, management fees, profits, and taxes; Renovation effect association: The quotation details are linked to the renovation renderings. When a user clicks on a quotation detail, they can locate the corresponding renovation element in the rendering, achieving a "what you see is what you get" quotation experience and helping users intuitively understand the reasonableness of the quotation.

[0061] The real-time adjustment function allows users to modify materials and processes in the quotation online, and the system updates the quotation results in real time. Material adjustment: Users can change the brand and specifications of materials, and the system will automatically call up the corresponding unit price in the material library and recalculate the material cost and the comprehensive unit price; Process adjustment: Users can modify the construction process, and the system will automatically call the corresponding labor and machinery costs from the process library and recalculate the comprehensive unit price; Parameter adjustment: Users can adjust parameters such as material loss rate, management fee rate, and profit rate, and the system updates the total quotation in real time.

[0062] Multiple export formats are supported. The system allows you to export quotations to mainstream formats such as PDF, Excel, and Word, making it convenient for users to save, print, and share. PDF format: Preserves the visual format of the quotation, making it easy for users to view and share, and suitable for submission to customers; Excel format: Displays the price details in a tabular format, making it easy for users to edit and perform data statistics; Word format: makes it easy for users to add notes and modify content, and is suitable for the document management needs of decoration companies.

[0063] This automated renovation quotation generation system achieves several innovations based on existing technology. The core innovations are as follows: This innovative multimodal fusion parsing technology overcomes the limitations of existing systems that can only analyze geometric information from CAD drawings. It employs a multimodal input mode combining "decoration renderings + CAD floor plans," and integrates improved YOLOv8 and U-Net deep learning algorithms to achieve accurate identification and attribute extraction of decoration elements. Through the collaborative work of these two algorithms, it solves the problems of low accuracy in recognizing small-sized decoration elements and inaccurate boundary segmentation. Simultaneously, it extracts visual attributes such as material, color, and specifications of decoration elements, establishing a link between "visual elements - size information - quotation details." This truly delivers a "what you see is what you get" quotation experience, addressing the pain point of customers' difficulty in understanding pricing information.

[0064] The improved deep learning algorithm innovations include targeted optimizations to YOLOv8 and U-Net algorithms: a lightweight attention mechanism and feature fusion module are added to YOLOv8 to improve the recognition accuracy of small-sized decorative elements; a boundary enhancement module and Dice loss function are added to U-Net to improve the boundary segmentation accuracy of decorative elements and the ability to handle pixel imbalance problems. The improved algorithm achieves a recognition accuracy of over 95%, significantly reducing recognition errors and providing a core guarantee for the accuracy of quotations.

[0065] A standardized preset scheme library covering multiple styles and price ranges has been established. Combined with cosine similarity algorithm and weight coefficient adjustment, intelligent matching of decoration schemes is achieved. Unlike the simple template matching of existing systems, this system's matching algorithm considers the key attribute weights of decoration elements, resulting in matching results that better match users' actual needs. Simultaneously, the scheme library supports real-time updates and manual adjustments, balancing standardization and flexibility, and solving the problem of inconsistent industry pricing standards.

[0066] A visual display method linking "quotation details - decoration effect" was designed, allowing users to intuitively view the decoration elements corresponding to the quotation, reducing the difficulty of understanding. At the same time, it supports users to adjust materials, processes and related parameters online, and update the quotation results in real time, realizing "what you see can be changed and what you change can be calculated", which improves the user interaction experience and communication efficiency, and further reduces communication costs.

[0067] Test environment: CPU is Intel Core i7-12700H, GPU is NVIDIA RTX 3060, memory is 16GB, operating system is Windows 11, programming language is Python, and database is MySQL 8.0.

[0068] Test data: 100 different apartment types and styles were selected, covering small apartments (under 60㎡), medium apartments (60-120㎡), and large apartments (over 120㎡). Each case includes complete decoration renderings, CAD floor plans, and labor quotations for system testing and performance comparison.

[0069] This test focused on three core indicators: quotation efficiency, quotation accuracy, and element recognition precision. The test results were compared with those of manual quotation and existing automated quotation systems, as detailed below: Test metric: Time (in hours) required to generate a complete quotation. Test results are shown in the table below:

[0070] Test results show that the quotation efficiency of this system is far higher than that of manual quotation and existing automatic quotation systems, shortening the quotation cycle from 2-3 hours to within half an hour, and greatly improving customer response speed.

[0071] Test metric: Quotation error rate (|System Quotation - Manual Quotation| / Manual Quotation × 100%). Test results are as follows: Manual quotation error rate: 15%-25% (industry average); Current system quotation error rate: 8%-12%; The error rate for quotations in this system is 3%-5%.

[0072] Test results show that the accuracy of this system's quotations is significantly higher than that of manual quotations and existing automatic quotations, with the error rate controlled within 5%, thus solving the problem of large errors in manual quotations.

[0073] Test metric: Accuracy of decoration element recognition (number of correctly recognized elements / total number of elements × 100%). Test results are as follows: The current system's recognition accuracy is 75%-85%. The system has an accuracy rate of over 95%.

[0074] Test results show that the improved deep learning algorithm of this system effectively improves the recognition accuracy of decoration elements, providing a reliable guarantee for the accuracy of quotations.

[0075] Based on the analysis of the test results, this system has the following performance advantages: High efficiency: Utilizing optimized algorithms and modular design, it significantly reduces quote generation time and improves customer response speed; High accuracy: Through improved deep learning algorithms, standardized databases, and verification mechanisms, the quotation error is controlled within 5%; Highly user-friendly: Visual display and real-time adjustment functions lower the barrier to entry for users and enhance the interactive experience; Good scalability: Modular design and standardized interfaces facilitate subsequent function upgrades and database updates.

[0076] This paper proposes an automated home renovation quotation generation system that addresses the core pain points in the quotation process of the home renovation industry. It integrates six core modules: multimodal input, intelligent parsing, scheme matching, a database of materials and processes, quotation calculation, and output display. Employing improved deep learning algorithms and standardized data support, it achieves full automation of the home renovation quotation process. Through multimodal fusion parsing technology, the system overcomes the limitations of existing systems, providing a "what you see is what you get" quotation experience. By optimizing algorithms and establishing a standardized database, it significantly improves quotation efficiency and accuracy, reducing the quotation cycle from 2-3 hours to within half an hour, and controlling the quotation error to within 5%. Through visual interaction and real-time adjustment functions, it reduces the difficulty of understanding and communication costs for customers, and solves the problem of inconsistent quotation standards in the industry.

[0077] Test results show that the system is stable and practical, effectively meeting the needs of decoration companies and customers. It provides a feasible technical solution for the standardization and intelligent transformation of the quotation process in the decoration industry, and has significant practical application value.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A system for automatically generating renovation quotes, characterized in that, It includes six core modules: multimodal input module, intelligent parsing module, scheme matching module, process and material database, quotation calculation module, and output display module. These modules work together to automate the entire quotation process. Multimodal input module: As the system's input entry point, it is responsible for receiving decoration renderings and CAD floor plans uploaded by users, supporting multiple formats, and completing file preprocessing; Intelligent parsing module: The core processing module of the system, responsible for parsing the input multimodal data and extracting key information such as decoration elements, dimensions, and structure; The scheme matching module matches the optimal renovation scheme template from the preset scheme library based on the extracted renovation element information. Process and Materials Database: The core supporting module of the system, storing various data on materials, processes, and quotas, providing data support for quotation calculation; Quotation Calculation Module: By combining analytical information, matching schemes, and database data, it automatically calculates the project quantity and various costs, and generates a complete quotation. Output display module: Displays the quotation results to users in a visual way, supporting real-time adjustment and export in multiple formats.

2. The automated renovation quotation generation system as described in claim 1, characterized in that, The system data flow is as follows: users upload files through the multimodal input module, which are then preprocessed and passed to the intelligent parsing module; the intelligent parsing module extracts key information and passes it to the solution matching module and the quotation calculation module respectively. The scheme matching module matches the optimal scheme from the preset scheme library of the process material database and transmits the scheme information to the quotation calculation module; The quotation calculation module calls the material, process, and quota data in the process material database to complete the quotation calculation; finally, the quotation result is displayed to the user through the output display module, and the user can adjust and export it.

3. The automated renovation quotation generation system as described in claim 1, characterized in that, The core function of the multimodal input module is to receive decoration renderings and CAD floor plans uploaded by users, support multiple mainstream formats, and preprocess the uploaded files to ensure that the subsequent parsing module can efficiently and accurately extract key information. Input format support: Considering the actual application scenarios in the decoration industry, the module supports uploading the following two types of files: Interior design renderings: Supports mainstream formats such as JPG, PNG, 3D Max, SketchUp, and Kujiale for rendering 3D images, covering commonly used rendering output formats in interior design to meet the usage habits of different designers; CAD floor plans: Supports mainstream CAD formats such as DWG and DXF, covering most drawing formats used by decoration companies, ensuring complete import of drawing information.

4. The automated renovation quotation generation system as described in claim 1, characterized in that, File preprocessing requires preprocessing the uploaded files. The specific steps are as follows: Image denoising: For interior design renderings, a Gaussian filtering algorithm is used to remove image noise and reduce the interference of noise on subsequent element recognition; Resolution Enhancement: For low-resolution renderings, a bilinear interpolation algorithm is used to enhance image resolution, ensuring that details of decorative elements are clearly discernible; CAD file format conversion: For different versions of DWG and DXF files, an open-source CAD parsing library is used to convert them into the standard DXF format, eliminating format differences; After preprocessing, the module transmits the standardized renderings and CAD files to the intelligent parsing module, providing a reliable data foundation for subsequent information extraction.

5. The automated renovation quotation generation system as described in claim 1, characterized in that, The core of interpreting interior design renderings is to identify various interior design elements and determine their material, color, and specification attributes. Traditional YOLOv8 object detection algorithms and U-Net semantic segmentation algorithms have shortcomings in interior design element recognition: YOLOv8 has low accuracy in recognizing small-sized elements, and U-Net is not accurate enough in segmenting the boundaries of interior design elements. Furthermore, both are difficult to effectively extract the attribute information of elements. To address the low accuracy of YOLOv8 in recognizing small elements, this system optimizes its feature extraction network: a lightweight attention mechanism is added to the backbone to enhance the ability to extract features from small elements; a feature fusion module is used in the neck to combine shallow and deep features, improving the accuracy of element boundary localization; the improved YOLOv8 algorithm loss function is as follows: in, For category loss, the cross-entropy loss function is used to determine the category of decoration elements; For bounding box loss, the CIoU loss function is adopted to improve the bounding box localization accuracy; The target confidence loss is used to determine whether the detected area is a decoration element; , The weighting coefficients are set to 1.5 and 1.0 respectively to balance various losses.

6. The automated renovation quotation generation system as described in claim 1, characterized in that, The improved YOLOv8 algorithm process is as follows: the preprocessed image is input into the network, image features are extracted through the backbone, enhanced by the attention mechanism, feature fusion is performed through the neck part, and finally the category, bounding box coordinates and confidence score of the decoration elements are output through the head part.

7. The automated decoration quotation generation system as described in claim 1, characterized in that, To address the inaccuracy of boundary segmentation in U-Net, this system adds a boundary enhancement module to the decoding part of U-Net. By capturing the edge features of the image, it improves the boundary segmentation accuracy of decorative elements. Simultaneously, Dice loss is introduced into the loss function to resolve the pixel imbalance problem of decorative elements. The improved U-Net loss function is as follows: in, The pixel values ​​predicted by the model. The actual label pixel value, The total number of pixels in the image. To avoid constants with a denominator of 0; The improved U-Net algorithm is used for semantic segmentation of decoration elements, accurately dividing the regional range of different elements. It works in conjunction with the improved YOLOv8 algorithm: YOLOv8 is responsible for identifying the element category and approximate location, while U-Net is responsible for accurately segmenting the element boundaries. The combination of the two achieves accurate identification of decoration elements.

Images