Intelligent human settlement space inside and outside environment ai recognition and comprehensive evaluation system

Abstract

The application discloses an intelligent human settlement space internal and external environment AI identification and comprehensive evaluation system, and belongs to the technical field of artificial intelligence image identification and human settlement environment evaluation. The system comprises: an indoor real scene acquisition and AI identification module, which is used for full-element identification of indoor real scene images, extraction of space structure, visual environment, furnishing layout and microenvironment parameters; an outdoor close-range acquisition and semantic identification module, which is used for identification of residential surrounding structures, road traffic, landscape water systems and outdoor microclimate parameters; a satellite landform data acquisition and macro analysis module, which is used for analysis of global terrain, water system pattern and road network air flow field characteristics. The application realizes full-dimension linkage evaluation of indoor micro observation, outdoor medium observation and geographical macro observation, has the advantages of high identification precision, comprehensive evaluation dimension, strong intelligent adaptability and the like, and can be widely applied to fields such as civil residential self-evaluation, home design, real estate transaction and the like.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence image recognition and intelligent assessment of human living environment technology, specifically involving an intelligent AI recognition and comprehensive assessment system for the internal and external environment of human living space. Background Technology

[0002] Currently, with the popularization of artificial intelligence vision technology and geographic information technology, software tools for residential unit type analysis and human living environment assessment have begun to appear on the market. Existing technical solutions usually rely on users uploading standardized unit type drawings, and then use preset rules or simple image recognition algorithms to identify basic building structures such as walls, doors, and windows, and score a few indicators such as the squareness of the unit type and functional zoning according to fixed templates.

[0003] However, the aforementioned prior art has the following significant drawbacks: (1) Existing technologies can only process paper or electronic floor plans, and cannot collect and analyze the fine visual elements such as the actual color matching, soft furnishings, furniture placement and size ratio, and object occlusion relationship in the interior, resulting in a serious disconnect between the evaluation results and the actual living experience of users, and poor evaluation accuracy; (2) Existing technical solutions all focus on a single indoor space and fail to include nearby outdoor environmental factors such as buildings outside the windows and balconies, road directions, building spacing, greening and water bodies in the evaluation system. Therefore, they cannot quantify and analyze the actual impact of the outdoor environment on indoor lighting, ventilation, privacy and visual comfort. (3) Existing technologies do not have the ability to access and analyze satellite remote sensing data or urban spatial big data, and cannot obtain macro-geographic information such as the large-scale topographic undulations, water system network, and road network pattern around the residence. Therefore, they cannot complete the overall assessment of the spatial enclosure, airflow circulation and residential stability of the area.

[0004] (4) Existing technologies all use fixed weight rules or templates for standardized scoring, which cannot dynamically adjust the weight of each evaluation dimension according to different application scenarios (such as decoration, home purchase, and building planning) and the actual characteristics of different regions and different house types. This results in serious homogenization of evaluation results and cannot provide truly personalized and accurate optimization solutions.

[0005] In summary, there is an urgent need in this field for an intelligent human settlement environment analysis system that can cover all dimensions of indoor micro-level, outdoor meso-level, and geographical macro-level, and has dynamic adaptive evaluation capabilities, in order to fill the gaps in existing technologies. Summary of the Invention

[0006] This invention aims to provide an AI-based intelligent residential space environment recognition and comprehensive evaluation system, which solves the technical problems of existing technologies that can only analyze standardized floor plans and cannot identify indoor real-world details, ignore the influence of outdoor and macro-geographical environments, and have rigid evaluation modes that lead to one-sided and insufficient personalization in the evaluation.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: An AI-powered intelligent living space environment recognition and comprehensive assessment system, including: Indoor Real-Scene Acquisition and AI Recognition Module: This module receives indoor real-scene images captured by users through mobile devices and uses a deep learning model to perform full-element recognition on the indoor real-scene images, extracting indoor spatial structure features, visual environment features, furnishing layout features, and micro-environment parameters. Outdoor close-up acquisition and semantic recognition module: used to receive real-world images of the exterior of the residence taken by the user through a mobile device, and to identify and quantify the surrounding outdoor structures, road traffic environment, landscape water system environment and outdoor microclimate parameters using semantic segmentation and target detection algorithms. Satellite geomorphology data acquisition and macroscopic analysis module: Based on the geographical location information of the user's residence, it automatically captures and analyzes satellite remote sensing data or urban spatial big data of the surrounding area of ​​the residence to generate the overall topographic pattern, water system pattern and road network airflow field characteristics. Multi-source data fusion evaluation module: It is connected to the indoor real-scene acquisition and AI recognition module, the outdoor close-up acquisition and semantic recognition module, and the satellite landform data acquisition and macro analysis module, respectively. It is used to standardize and cross-validate the received indoor micro data, outdoor meso data and geographic macro data, and perform weighted fusion calculation according to the preset or dynamically adjusted weight allocation strategy, and finally generate a quantitative comprehensive evaluation report and layout optimization scheme.

[0008] Furthermore, the indoor spatial structure features identified by the indoor real-scene acquisition and AI recognition module include at least: the overall layout of the house, functional zoning, walls, door and window locations, and beam and column distribution; the visual environment features include at least: the color tone of the whole house, the intensity of natural lighting, and the range of light and shadow distribution; the furnishing layout features include at least: furniture type, location, size ratio, and orientation angle; and the micro-environment parameters include at least: spatial comfort index calculated based on lighting angle, spatial transparency, and the area obstructed by objects.

[0009] Furthermore, the outdoor surrounding structures identified by the outdoor close-up acquisition and semantic recognition module include at least: the height of adjacent buildings, the distance between buildings, and the building obstruction structure; the road traffic environment includes at least: the road direction, the road network distribution, and the vehicle flow lines; the landscape water system environment includes at least: the water area, the outline, and the direction; and the outdoor microclimate parameters include at least: the ventilation adaptability index calculated based on the proportion of building obstruction area and the proportion of open space.

[0010] Furthermore, the overall terrain pattern analyzed by the satellite geomorphology data acquisition and macroscopic analysis module includes at least: the topographic undulations, elevation, and enclosing structure of the residential area; the water system pattern includes at least: the distribution of surrounding water bodies and the direction of rivers; and the road network airflow field characteristics include at least: the regional spatial airflow field distribution formed by the urban main and secondary road network.

[0011] Furthermore, the weight allocation strategy in the multi-source data fusion evaluation module is an adaptive weight allocation strategy. This strategy dynamically adjusts the fusion weights of indoor micro data, outdoor meso data, and geographic macro data according to different application scenarios. The application scenarios include decoration scenarios, home purchase scenarios, and architectural planning scenarios.

[0012] Furthermore, the mobile device on which the system relies includes a smartphone or tablet; the mobile device at least invokes the following hardware modules to interact with the system software: The rear camera module is used to capture indoor and outdoor real-world images with a resolution that meets the clarity requirements of the recognition algorithm; The positioning module is used to obtain the geographical location of the user's residence, so as to provide a positioning reference for the satellite topographic data acquisition and macroscopic analysis module; The processor module is used to accelerate the operation of the deep learning model and fusion algorithm; The display module is used to show the real-scene shooting interface, the comprehensive evaluation report, and the layout optimization scheme.

[0013] Furthermore, the deep learning model used in the indoor real-scene acquisition and AI recognition module is an "indoor environment full-element recognition model" based on convolutional neural networks. Its training process includes: training with a dataset containing multiple indoor real-scene images labeled with wall, door, window, furniture and color elements; and performing preprocessing operations such as denoising, normalization and size adjustment on the input indoor real-scene images in sequence.

[0014] Furthermore, the outdoor close-up acquisition and semantic recognition module adopts an improved model based on Mask R-CNN, and sets the semantic segmentation threshold to 0.7 and the target detection confidence to 0.8; and performs preprocessing operations such as dehazing and / or light compensation on the input outdoor real-scene image.

[0015] Furthermore, when the multi-source data fusion evaluation module fuses indoor micro data, outdoor meso data, and geographic macro data, it performs the following operations: normalizes the quantitative indicators of all data to the same numerical range; calculates the comprehensive score using a weighted summation formula; and compares the consistency of the three types of data using a cross-validation algorithm. If the deviation exceeds a preset threshold, the weights are readjusted.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention achieves significant benefits by constructing a three-layer linkage evaluation system of indoor real-scene AI recognition, outdoor near-scene semantic recognition, and satellite topographic big data analysis, and by adopting an adaptive weighted fusion algorithm. First, this invention abandons the traditional analysis mode based on single floor plan drawings, and accurately identifies refined real-scene elements such as indoor color, soft furnishings, furniture layout, lighting, and ventilation through a self-developed "indoor environment full-element recognition model." The evaluation results highly match the user's actual living scenario, greatly improving the accuracy and authenticity of the living environment assessment. Second, this invention, for the first time, incorporates nearby environmental factors such as surrounding buildings, roads, traffic, and landscape water systems outside the windows and balconies of the residence. By incorporating a quantitative evaluation system and combining satellite remote sensing data to analyze the overall terrain, water system, and road network pattern, this invention overcomes the one-sidedness of existing technologies that only focus on indoor spaces, achieving a comprehensive evaluation across all dimensions: indoor micro-level, outdoor meso-level, and geographical macro-level. Furthermore, this invention employs an adaptive weight allocation strategy, dynamically adjusting the fusion weights of the three types of data according to different application scenarios such as decoration, home purchase, and architectural planning. This overcomes the homogenization problem caused by traditional fixed templates, enabling the output of personalized and refined spatial optimization solutions. Finally, this invention is implemented using a pure software solution on mobile devices, requiring no additional hardware, and can be quickly promoted to multiple fields such as residential housing, home design, and real estate transactions, possessing excellent industrial application prospects and commercial value. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall system architecture and data flow of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be further described in detail below with reference to the embodiments.

[0020] like Figure 1 As shown, this invention provides an AI-based intelligent recognition and comprehensive evaluation system for the internal and external environments of intelligent living spaces. The system adopts a layered architecture, comprising, from top to bottom, a user interaction layer, a data acquisition layer, an algorithm processing layer, a data storage layer, and an application output layer. Each layer operates independently yet interconnects, forming a complete closed loop of "acquisition-processing-storage-output." The system relies on mobile intelligent devices (such as smartphones or tablets running iOS or Android operating systems) to achieve data acquisition and user interaction.

[0021] The intelligent living space indoor and outdoor environment AI recognition and comprehensive evaluation system includes an indoor real-scene acquisition and AI recognition module, an outdoor close-up acquisition and semantic recognition module, a satellite topographic data acquisition and macro-analysis module, and a multi-source data fusion evaluation module.

[0022] The indoor real-scene acquisition and AI recognition module is used to receive indoor real-scene images taken by users through mobile devices and to perform full-element recognition using a deep learning model.

[0023] After the user launches the mobile app, they authorize the use of the rear camera (≥12MP, supports autofocus) to sequentially capture real-world images of the living room, bedroom, kitchen, and other rooms, with an image resolution of at least 1080P. After capturing the images, the module automatically triggers an image preprocessing algorithm, sequentially performing noise reduction (Gaussian filtering, 3×3 convolution kernel), image normalization (pixel values ​​normalized to the [0,1] range), and size adjustment (uniform to 640×480 pixels).

[0024] This module employs a self-developed deep learning model, named the "Indoor Environment Full-Element Recognition Model," which is based on an improved Convolutional Neural Network (CNN). During the model training phase, it is jointly trained using the COCO public dataset and a self-built indoor environment dataset. The self-built dataset contains over 100,000 real-world indoor images of different apartment types and decoration styles, with annotations including elements such as walls, doors and windows, beams and columns, furniture, colors, and lighting. Training parameters: learning rate 0.001, 100 iterations, batch size 32, resulting in a recognition accuracy of ≥95% after training.

[0025] In the inference phase, the preprocessed images are input into the model. Through feature extraction, object detection (YOLOv8), and semantic segmentation, the following element recognition is achieved: extraction of indoor spatial structural features (house layout, functional zoning, walls, door and window locations, beam and column distribution), visual environment features (color tone analysis through color histograms, and quantification of lighting intensity and light and shadow distribution through brightness detection), furnishing layout features (furniture type, location, size ratio, and orientation angle), and calculation of micro-environmental parameters (based on lighting angle, spatial transparency, and object occlusion area, calculated using the formula "comfort = 0.4 × lighting intensity + 0.3 × transparency + 0.3 × (1 - occlusion rate)"). The recognition results are organized into structured data and transmitted to the multi-source data fusion evaluation module.

[0026] The outdoor close-up acquisition and semantic recognition module is used to receive real-world images of the exterior of a residence taken by the user through a mobile device, and to perform recognition using semantic segmentation and object detection algorithms.

[0027] Users capture real-world images of the view outside their windows, balconies, or public areas of their residential community, encompassing surrounding buildings, roads, greenery, and other elements. This module employs a fusion scheme of semantic segmentation and object detection, based on an improved Mask R-CNN model. Parameter settings: input resolution 1080P, semantic segmentation threshold 0.7, object detection confidence ≥ 0.8. For the input image, dehazing (dark channel prior algorithm) and lighting compensation (adaptive histogram equalization) preprocessing are first performed.

[0028] The preprocessed image is input into the model, and the following recognition processes are performed: Surrounding structure recognition (segmenting building areas, measuring height error ≤ 1 meter, building spacing error ≤ 2 meters, and identifying occlusion structures); Road traffic recognition (extracting road direction, determining traffic density and movement patterns); Landscape water system recognition (segmenting water systems based on color features within the RGB range [0,50,100]-[50,100,255], extracting contours and directions); Outdoor microclimate parameter calculation (calculated using the formula "ventilation adaptability = 0.6 × (1 - occlusion rate) + 0.4 × open space ratio"). The recognition results are transmitted in structured data form to the multi-source data fusion evaluation module.

[0029] The satellite geomorphology data acquisition and macroscopic analysis module automatically captures and analyzes satellite remote sensing data or urban spatial big data based on the geographical location information of users' residences.

[0030] The system acquires the latitude and longitude coordinates (accuracy ≤ 10 meters) of the user's residence via the mobile terminal's GPS / BeiDou positioning module. Based on these coordinates, it sends a request to the satellite data service interface to retrieve the digital elevation model (DEM), satellite imagery, and road network vector data within a radius of 1-5 kilometers centered on the residence. Then, it performs: a comprehensive topographic pattern analysis (topographic relief, elevation, and landform enclosure structure); a water system pattern analysis (calculating water body distribution and river direction using the NDWI water index); and a road network airflow field analysis (analyzing the direction and density of main and secondary roads to assess regional airflow field distribution). The macroscopic data, after structured processing, is transmitted to the multi-source data fusion and evaluation module.

[0031] The multi-source data fusion evaluation module is connected to the three modules mentioned above. It is used to standardize and cross-validate indoor micro data, outdoor meso data and geographic macro data, and perform weighted fusion calculation according to the weight allocation strategy to finally generate a comprehensive evaluation report and layout optimization scheme. The specific steps are as follows: (1) Standardization: normalize all quantitative indicators to the range of [0,100]; supplement missing data using cubic spline interpolation. (2) Adaptive weight allocation: dynamically adjust the weights according to the application scenario - indoor:outdoor:satellite weights are 0.5:0.3:0.2 in the decoration scenario, 0.4:0.3:0.3 in the home purchase scenario, and 0.2:0.3:0.5 in the building planning scenario. (3) Weighted fusion and cross-validation: calculate the comprehensive score using the weighted summation formula; at the same time, cross-validate the consistency of the three types of data (such as the matching degree of indoor lighting and outdoor shading). If the deviation is ≥10%, the weights are readjusted. (4) Output: generate a quantitative evaluation report and layout optimization scheme and transmit it to the application output layer.

[0032] The following describes the specific implementation of the various features identified by the indoor real-scene acquisition and AI recognition module. These features include the recognition of spatial structure features, visual environment features, furnishing layout features, and the calculation of micro-environment parameters.

[0033] Spatial structural features are identified by extracting edge features of walls, doors, windows, beams, and columns using an edge detection algorithm (Canny operator). The model's fully convolutional layer outputs a semantic segmentation map, with different color channels corresponding to categories such as walls, doors, windows, and beams / columns. In the post-processing stage, connected component analysis is used to convert the segmentation results into vector boundaries, thereby calculating the closed contours of each functional area. Testing showed that the planar coordinate error of door and window locations is ≤5 cm.

[0034] Visual environment feature recognition involves converting the image to the HSV color space and statistically analyzing the hue distribution histogram. If warm tones (0~30° and 330~360°) account for more than 60% of the pixels, the image is classified as a warm tone; if cool tones (180~270°) account for more than 60%, the image is classified as a cool tone. Light intensity is quantified using the mean of the luminance channel (V channel), and the range of light and shadow distribution is identified by analyzing the direction of the luminance gradient.

[0035] For the identification of furnishing layout features, the YOLOv8 object detection network is used to output the bounding box and category label of each detected object (sofa, bed, dining table, etc.). Based on this, the center position of the furniture, the aspect ratio of the bounding box (as a size ratio, with an error of ≤10%), and the orientation angle are calculated.

[0036] The calculation of microenvironment parameters is based on the lighting angle, spatial transparency (the ratio of door and window area to room area) and the area obstructed by objects. The comfort index of 0-100 points is calculated according to the formula "spatial comfort = 0.4 × normalized value of lighting intensity + 0.3 × transparency + 0.3 × (1 - shading rate)".

[0037] The following describes the specific implementation of the various features identified by the outdoor close-up acquisition and semantic recognition module. These features include the identification of surrounding structures, road and traffic environments, landscape and water system environments, and outdoor microclimate parameters.

[0038] Implementation of surrounding structure recognition: In the Mask R-CNN model, the building instance segmentation result is output as a polygonal mask. The actual height is calculated using the known camera focal length and the pixel height of the building in the image, combined with the principle of similar triangles (error ≤ 1 meter). The distance between buildings is converted from the horizontal distance between the centroids of the two building masks. Occlusion structures are classified by judging the overlap ratio between the building mask and the residential area mask on the image plane (no occlusion: <5%; partial occlusion: 5%~50%; full occlusion: >50%).

[0039] Implementation of road traffic environment recognition: For semantically segmented road regions, Hough line detection is applied to extract the angles of the main lines as the road direction. Road network density is calculated by the proportion of road pixels to total image pixels. Vehicle flow lines are calculated based on the difference images of three consecutive frames, and the principal direction of the motion vector is calculated.

[0040] Implementation of landscape water system environment recognition: Regions satisfying the pixel range [0,50,100]-[50,100,255] for water systems are extracted in RGB space. Morphological closing operations are applied to fill the holes, and then Canny edge detection is used to extract the contours. The direction of the water system is determined by the direction angle of the contour skeleton lines.

[0041] Calculation of outdoor microclimate parameters: The degree of light shading is calculated using the formula "shading rate = shading area / total field of view area". The ventilation adaptability index is calculated using the formula "ventilation adaptability = 0.6 × (1 - shading rate) + 0.4 × open space ratio", where the open space ratio is obtained from the proportion of unshaded area in the semantic segmentation results.

[0042] The following section further describes the specific implementation of the features analyzed by the satellite geomorphology data acquisition and macroscopic analysis module.

[0043] Analysis of the overall terrain pattern: Based on digital elevation model (DEM) data, the maximum elevation difference (undulation) within a 1km radius centered on the residence is calculated, as well as the elevation percentile of the residence relative to the surrounding area (determining whether it is a slope top, mid-slope, or depression). The geomorphic enclosure structure is identified by calculating the elevation distribution in a 360° direction around the residence to determine if there is continuous high terrain forming an enclosure.

[0044] Analysis of the water system pattern: Water body pixels are extracted by calculating the improved Normalized Difference Water Index (MNDWI), and the water area and shortest distance to residences are statistically analyzed. Skeletonization is performed on continuous water body pixels to obtain the river centerline, and its direction angle is fitted.

[0045] Analysis of road network airflow characteristics: Open-source road network data was acquired, the geometric lines of main and secondary roads were extracted, the azimuth angle of each road was calculated, and a directional rose diagram was drawn. Combining meteorological principles (airflow accelerates along the road direction), the potential of the area as a ventilation corridor was quantitatively assessed through road density and connectivity.

[0046] This section further explains the adaptive weight allocation strategy in the multi-source data fusion evaluation module. When a user launches the app, a scene selection interface is displayed, allowing the user to manually select their current purpose ("I want to renovate," "I want to buy a house," or "I want to plan"). If the user does not manually select, the system can automatically infer based on the characteristics of the uploaded data: if the user uploads multiple interior detail photos without explicitly requesting macroscopic analysis, it is determined to be a renovation scene; if the user enters their home purchase intention or searches for properties, it is determined to be a home purchase scene; if the user uploads a floor plan with topographical annotations or requests regional analysis, it is determined to be an architectural planning scene. The specific weight allocation values ​​are as follows: Renovation scene—Indoor weight 0.5, Outdoor weight 0.3, Satellite weight 0.2, emphasizing the dominant role of interior layout while considering the influence of the near-view outside the window; Home purchase scene—Indoor weight 0.4, Outdoor weight 0.3, Satellite weight 0.3, a balance among the three, taking into account indoor quality, the surrounding area of ​​the building, and the macroscopic environment of the community; Architectural planning scene—Indoor weight 0.2, Outdoor weight 0.3, Satellite weight 0.5, focusing on macroscopic topography, water systems, and road network patterns. The aforementioned weight values ​​are directly substituted into the weighted summation formula for fusion calculation.

[0047] The following describes the interaction between the various hardware modules and system software on the mobile device.

[0048] Rear camera module: The system calls the AVFoundation (iOS) or Camera2 (Android) API to set the capture resolution to 1080P and enable autofocus and auto white balance. After the user presses the shutter, the system reads the image data from the image buffer and directly passes it to the preprocessing unit of the recognition module, automatically triggering noise reduction and normalization processing.

[0049] Positioning Module: The system calls Core Location (iOS) or Fused Location Provider (Android) to request high-precision positioning permission. Outdoors, GPS / BeiDou is prioritized, with a positioning accuracy ≤10 meters; indoors, it automatically switches to WiFi-assisted positioning (by scanning surrounding WiFi hotspot SSIDs and signal strengths, combined with a WiFi location database). The acquired latitude and longitude coordinates are directly transmitted to the satellite terrain data acquisition module as the positioning reference.

[0050] Processor Module: The system calls the GPU via Metal (iOS) or OpenCL / Neural Networks API (Android) to compile deep learning models into GPU-executable instructions. Recognition tasks are executed in an asynchronous queue, with the processor automatically scheduling CPU and GPU resources to prioritize real-time recognition tasks, achieving a processing speed of ≤3 seconds / frame.

[0051] Storage Module: The system uses the mobile device's built-in storage to create an application-specific directory, categorizing and storing raw images, intermediate feature data, and final reports. It also integrates a cloud storage SDK, automatically synchronizing data to the cloud over Wi-Fi, supporting both local backup and cloud synchronization.

[0052] Display module: The system renders visual charts (such as radar charts and bar charts) in the evaluation report using Metal / OpenGL. The display interface supports multi-touch, allowing users to zoom in on real-world image details with two-finger zoom gestures and switch between different evaluation dimensions (such as switching from the "Lighting" view to the "Ventilation" view) with a single-finger swipe.

[0053] This section describes the specific applications of the system in three areas: residential housing, home design, and real estate transactions.

[0054] For individual users in residential buildings: Ordinary residential users can complete the following operations through the APP: ① Authorize location and camera permissions; ② Take real-time images of each room in the room (living room, bedroom, kitchen, etc.) and upload the floor plan; ③ Take real-time outdoor images of the windows and balconies (surrounding buildings, roads, greenery); ④ The system automatically captures satellite data, processes it through algorithms, and outputs a quantitative evaluation report, such as "Spatial layout rationality: 85 points; Lighting comfort: 78 points; Ventilation suitability: 82 points", and provides optimization suggestions such as "Adjust the placement of bedroom furniture to avoid blocking light" and "Plant greenery on the balcony to improve the suitability of the outdoor landscape".

[0055] In a home design company scenario: Designers visit the client's residence with their mobile devices, photographing the interior and exterior views and uploading the floor plan. The system quickly generates an assessment report, identifying the shortcomings of the existing space (e.g., "unreasonable kitchen and dining room layout, cramped furniture placement," "severe outdoor obstruction, insufficient indoor lighting"). Designers then utilize the system's built-in optimization modules to adjust the design (e.g., optimizing furniture layout, adjusting soft furnishing colors, suggesting increased lighting), and display before-and-after comparisons through the system to improve the design's rationality and relevance, reducing rework rates.

[0056] Real estate transaction scenario: Before listing a property, real estate agents use this system to photograph the interior and exterior of the property and generate an evaluation report. In addition to displaying the property's scores across various dimensions, the report automatically compares the scores with the average scores of other evaluated properties in the same neighborhood. The agent then presents this report to potential buyers, allowing them to intuitively understand the property's suitability for living, improving transaction efficiency, and differentiating it from traditional property descriptions to enhance their core competitiveness.

[0057] This section further supplements the description of the training dataset composition and preprocessing operations for the deep learning model in the indoor real-scene acquisition and AI recognition module.

[0058] The self-built indoor environment dataset comprises over 100,000 real-world indoor images, covering 50 different apartment types (one- to four-bedroom) and 20 interior design styles (modern, European, Chinese, etc.). Each image is pixel-level labeled using the LabelMe tool, with 30 categories including walls, doors, windows, beams, ceilings, floors, sofas, beds, dining tables, chairs, lighting fixtures, curtains, and wall color areas. Data augmentation strategies such as random horizontal flipping, random rotation (±15°), and random brightness adjustment (0.8~1.2x) were employed during training.

[0059] The specific parameters for the preprocessing operations are as follows: Denoising uses OpenCV's GaussianBlur function with a kernel size of (3,3) and sigmaX=0; Normalization converts the image data type from uint8 to float32 and divides it by 255.0; Resizing uses the cv2.resize function with bilinear interpolation and an output size of 640×480 pixels.

[0060] This section provides further explanation of the specific parameter settings and preprocessing operations for the model in the outdoor close-up acquisition and semantic recognition module.

[0061] In this embodiment, the backbone network of the Mask R-CNN model is ResNet-101-FPN. The parameters set during inference are: detection_min_confidence is 0.8, meaning only object detection results with a confidence level ≥ 0.8 are retained; mask_threshold is 0.7, meaning the probability map output by semantic segmentation is binarized with a threshold of 0.7.

[0062] The specific implementation of preprocessing operations is as follows: The dehazing algorithm adopts the dark channel prior method, with a window size of 15×15 and a dehazing intensity parameter ω=0.95; the lighting compensation adopts the CLAHE algorithm, which divides the image into 8×8 blocks, limits the contrast to 2.0, and expands the histogram distribution to the full range of 0-255. The above preprocessing is performed automatically and sequentially after image acquisition.

[0063] This section further explains the normalization, weighted summation, and cross-validation operations in the multi-source data fusion evaluation module and provides numerical examples.

[0064] The normalization process involves mapping all quantitative indicators, such as indoor comfort (original score 0-100), outdoor ventilation suitability (original score 0-100), and terrain openness (original score 0-100), to the range [0, 100]. Indicators already in the 0-100 range are retained. For indicators with other dimensions, a linear mapping formula is used: Normalized value = (original value - minimum value) / (maximum value - minimum value) × 100.

[0065] The specific implementation of the weighted summation operation: Taking the home purchase scenario as an example, let the indoor data score be 82.5, the outdoor data score be 67.3, and the satellite data score be 74.8. The weights for indoor data are 0.4, outdoor data is 0.3, and satellite data is 0.3. Then the weighted summation comprehensive score is 82.5×0.4+67.3×0.3+74.8×0.3=33.0+20.19+22.44=75.63 points.

[0066] The cross-validation process involves mapping indoor daylight intensity (lux) to a score of 0-100 and comparing it against outdoor shading rate (converted from shading rate to a score). If the indoor daylight score is 80 and the outdoor shading score is 50, the relative deviation is calculated using the formula |80-50| / 80 = 37.5%, exceeding the preset 10% threshold. The system then triggers a weight reallocation: reducing the weight of outdoor data from 0.3 to 0.15 and correspondingly increasing the weight of indoor data from 0.4 to 0.55 (keeping the total weight at 1), before recalculating the overall score. This mechanism ensures that the evaluation results rely more heavily on more reliable data sources.

[0067] It should be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent human living space indoor and outdoor environment AI recognition and comprehensive evaluation system, characterized in that, include: Indoor Real-Scene Acquisition and AI Recognition Module: This module receives indoor real-scene images captured by users through mobile devices and uses a deep learning model to perform full-element recognition on the indoor real-scene images, extracting indoor spatial structure features, visual environment features, furnishing layout features, and micro-environment parameters. Outdoor close-up acquisition and semantic recognition module: used to receive real-world images of the exterior of the residence taken by the user through a mobile device, and to identify and quantify the surrounding outdoor structures, road traffic environment, landscape water system environment and outdoor microclimate parameters using semantic segmentation and target detection algorithms. Satellite geomorphology data acquisition and macroscopic analysis module: Based on the geographical location information of the user's residence, it automatically captures and analyzes satellite remote sensing data or urban spatial big data of the surrounding area of ​​the residence to generate the overall topographic pattern, water system pattern and road network airflow field characteristics. Multi-source data fusion evaluation module: It is connected to the indoor real-scene acquisition and AI recognition module, the outdoor close-up acquisition and semantic recognition module, and the satellite landform data acquisition and macro analysis module, respectively. It is used to standardize and cross-validate the received indoor micro data, outdoor meso data and geographic macro data, and perform weighted fusion calculation according to the preset or dynamically adjusted weight allocation strategy, and finally generate a quantitative comprehensive evaluation report and layout optimization scheme.

2. The intelligent living space indoor and outdoor environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The indoor spatial structure features identified by the indoor real-scene acquisition and AI recognition module include at least: the overall layout of the house, functional zoning, walls, door and window locations, and beam and column distribution; the visual environment features include at least: the color tone of the whole house, the intensity of natural lighting, and the range of light and shadow distribution; the furnishing layout features include at least: furniture type, location, size ratio, and orientation angle; and the micro-environment parameters include at least: spatial comfort index calculated based on lighting angle, spatial transparency, and the area obstructed by objects.

3. The intelligent human living space internal and external environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The outdoor surrounding structures identified by the outdoor close-up acquisition and semantic recognition module include at least: the height of adjacent buildings, the distance between buildings, and the building obstruction structure; the road traffic environment includes at least: the road direction, the road network distribution, and the vehicle flow line; the landscape water system environment includes at least: the water area, the outline, and the direction; and the outdoor microclimate parameters include at least: the ventilation adaptability index calculated based on the proportion of building obstruction area and the proportion of open space.

4. The intelligent human living space internal and external environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The overall topographic pattern analyzed by the satellite geomorphological data acquisition and macroscopic analysis module includes at least: the topographic undulations, elevation, and enclosing structure around the residence; the water system pattern includes at least: the distribution of surrounding water bodies and the direction of rivers; the road network airflow field characteristics include at least: the regional spatial airflow field distribution formed by the urban main and secondary roads.

5. The intelligent human living space indoor and outdoor environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The weight allocation strategy in the multi-source data fusion evaluation module is an adaptive weight allocation strategy. This strategy dynamically adjusts the fusion weights of indoor micro data, outdoor meso data, and geographic macro data according to different application scenarios. The application scenarios include decoration scenarios, home purchase scenarios, and architectural planning scenarios.

6. The intelligent living space indoor and outdoor environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The system relies on mobile devices including smartphones or tablets; the mobile device at least invokes the following hardware modules to interact with the system software: The rear camera module is used to capture indoor and outdoor real-world images with a resolution that meets the clarity requirements of the recognition algorithm; The positioning module is used to obtain the geographical location of the user's residence, so as to provide a positioning reference for the satellite topographic data acquisition and macroscopic analysis module; The processor module is used to accelerate the operation of the deep learning model and fusion algorithm; The display module is used to show the real-scene shooting interface, the comprehensive evaluation report, and the layout optimization scheme.

7. The intelligent human living space internal and external environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The deep learning model used in the indoor real-scene acquisition and AI recognition module is an "indoor environment full-element recognition model" based on convolutional neural networks. Its training process includes: training with a dataset containing multiple indoor real-scene images labeled with wall, door, window, furniture and color elements; and performing preprocessing operations such as denoising, normalization and size adjustment on the input indoor real-scene images in sequence.

8. The intelligent human living space indoor and outdoor environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, The outdoor close-up acquisition and semantic recognition module adopts an improved model based on Mask R-CNN, and sets the semantic segmentation threshold to 0.7 and the target detection confidence to 0.8; in addition, it performs preprocessing operations such as dehazing and / or light compensation on the input outdoor real-scene image.

9. The intelligent living space indoor and outdoor environment AI recognition and comprehensive evaluation system according to claim 1, characterized in that, When the multi-source data fusion evaluation module fuses indoor micro data, outdoor meso data, and geographic macro data, it performs the following operations: normalizes the quantitative indicators of all data to the same numerical range. The weighted summation formula is used to calculate the overall score; the cross-validation algorithm is used to compare the consistency of the three types of data, and if the deviation exceeds the preset threshold, the weights are readjusted.

Images