Space distribution prediction method for household air conditioners
Through high-resolution remote sensing images and street scene image data combined with deep learning technology, the spatial distribution information of household air conditioners is extracted, and the problems of lack and uncertainty of household air conditioners in the existing technology are solved, and efficient and accurate statistics of air conditioners and space are achieved. Distribution analysis improves the ability to deal with extreme high temperature events.
Patent Information
- Application Number
- CN202411145079.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, there is a lack of data on the use of household air conditioners and spatial distribution, traditional survey methods are uncertain and time-consuming, and there is a lack of efficient survey methods.
Using high-resolution remote sensing image and street scene image data combined with deep learning technology, the semantic segmentation model, height estimation model and instance segmentation model are used to extract the outline, height and pixel ratio of the air conditioner outside the city building, and calculate the side wall area and the number of air conditioners. , realize the spatial distribution prediction of household air conditioners.
It significantly improves the efficiency and accuracy of household air conditioner use surveys, provides reliable statistical information, and improves the ability to respond to extreme high temperature events and control level of power and energy dispatch.
Smart Images

Figure CN119091298A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of image processing and relates to a prediction method, in particular to a spatial distribution prediction method for household air conditioners. Background Art
[0002] Air conditioning is the main solution for urban residents to cope with extreme heat, but there is currently a lack of data on the number and spatial distribution of air conditioning usage. Existing air conditioning usage surveys still rely on traditional manual methods and sampling surveys to estimate statistical data. This survey method is outdated, the data uncertainty is high, and the spatial distribution of household air conditioners is not clear. So far, there has been no research results on efficient methods for investigating the use of household air conditioners.
[0003] High-resolution remote sensing images are an important data source containing rich information on the shape, structure and texture of surface targets, and are of great value for urban research. High-resolution aerial remote sensing images, satellite remote sensing images, drone remote sensing images and radar images are all widely used to extract the outlines of urban buildings. In recent years, there have also been studies using high-resolution remote sensing images as a data basis to extract the shadows and foundation pit information of buildings, and to estimate the height of buildings by establishing an inversion model of the geometric parameters between the sun and satellite positions and the buildings. Among them, a method with high feasibility and high accuracy is to extract the shadows of buildings using high-resolution remote sensing images, and to estimate the height of buildings by calculating the width of the shadows perpendicular to the direction of the buildings. Existing technologies have used Sentinel satellite images to extract and map the height of buildings in many cities across the country. In addition, methods based on high-resolution satellite remote sensing data combined with building height analysis have been further applied to study the fine distribution of urban population density. It can be foreseen that with the passage of time, the application space of high-resolution remote sensing image data will be broader.
[0004] In recent years, the development of street view services such as Google Street View and Baidu Panorama has gradually improved, and the coverage has continued to expand, covering more than half of the world's population. Street view image data spreads rapidly, and it has become easier to obtain street view image data for research, making it possible to conduct large-scale, comprehensive street view image research. Current street view images have a wide coverage range, can provide street-level landscape information, and have low data collection costs, which provides a large sample data source and new research ideas for urban environmental evaluation research. Street view images are image data that depict the visual environment of urban streets from a parallel perspective of people, which is not available from other data sources such as satellites and aviation. Therefore, street view image data is also described as a type of image data corresponding to remote sensing images.
[0005] Image segmentation is an important branch of computer vision. Its purpose is to classify each pixel according to the target neighborhood characteristics, and finally obtain a pixel-level semantically annotated image. The semantic segmentation results of the image can provide simple and effective image feature information, providing basic technical support for many applications. Early traditional image segmentation methods are often simple and effective, mainly including methods based on thresholds, regions, edges, clustering, graph theory, etc., but their performance in tasks that require a large amount of high-precision semantic information is often not ideal. Nowadays, with the rapid development of hardware and the large amount of data generated, technologies such as deep learning have developed rapidly, and have made significant progress in many fields of computer vision. In order to achieve pixel-level classification of images, the existing technology introduces the fully convolutional network (FCN) for image semantic segmentation. Most convolution-based semantic segmentation networks use the basic principles of FCN. Since then, there has been a boom in the research of regional pixel semantic understanding. For example, U-Net, which has achieved good results in medical image segmentation, has achieved good results in multi-class segmentation problems. Subsequent mainstream methods have tried to expand the receptive field and capture contextual information, such as DeepLab proposed in the prior art, which uses atrous spatial pyramid pooling (ASPP) to segment objects at multiple scales.
[0006] As climate change intensifies, extreme heat disasters pose an increasing risk to human life and safety. In this context, household air conditioners have become an important means for residents to effectively cope with extreme temperature events, and are also the main way to consume electricity during high temperatures in summer. However, the current understanding of the number of household air conditioners used by urban residents and their spatial distribution is not accurate enough. Traditional manual sampling surveys and user surveys are still the main means of household air conditioner usage statistics, and the survey results are highly uncertain, and the survey process is time-consuming and labor-intensive. Summary of the invention
[0007] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting the spatial distribution of household air conditioners, which solves the problem of inaccurate prediction of the number of household air conditioners in the prior art.
[0008] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:
[0009] A method for predicting the spatial distribution of household air conditioners comprises the following steps:
[0010] S1, acquiring remote sensing images and street view image data of the target area, preprocessing the remote sensing images, and obtaining preprocessed remote sensing images;
[0011] S2, inputting the preprocessed remote sensing image into the trained semantic segmentation model and the trained height estimation model, and outputting the urban building outline and building height of the target area through the semantic segmentation model and the height estimation model respectively;
[0012] S3, inputting the street view image data into the trained instance segmentation model, outputting the instance segmentation result of the target area through the instance segmentation model, and calculating the ratio of the air conditioner outdoor unit pixels in the instance segmentation result;
[0013] S4. Determine the side wall area of the target area based on the urban building outline and building height;
[0014] S5. Determine the number of household air conditioners in the target area based on the side wall area and the pixel ratio of the air conditioner outdoor unit.
[0015] The beneficial effects of the above scheme are:
[0016] The present invention utilizes information from high-resolution satellite remote sensing images and Internet map street view images, and combines deep learning and spatial statistical methods to focus on accurately extracting the usage and spatial distribution of urban household air conditioners, significantly improving the efficiency of household air conditioner usage surveys. At the same time, it further enhances the management and control level of urban residents and management departments in responding to extreme high temperature events and rationally dispatching electric energy. It can obtain statistical information on household air conditioners more quickly and accurately, and provide reliable data support for relevant decision-making and planning.
[0017] Furthermore, in step S1, the remote sensing image is preprocessed, specifically including:
[0018] S11, performing geometric correction on the remote sensing image using a polynomial correction model to obtain a remote sensing image after geometric correction;
[0019] The remote sensing images after geometric correction include multispectral band images and panchromatic band images;
[0020] S12. Transform and fuse the multispectral band image and the panchromatic band image to obtain a fused remote sensing image.
[0021] The beneficial effects of the above further scheme are: by performing geometric correction on the remote sensing image, the remote sensing image can be made consistent with the real ground object, and by fusing the image data of two different bands, the information content and availability of the image can be improved, thereby better supporting subsequent tasks such as building foundation pit identification, monomer extraction, shadow information extraction and building height inversion.
[0022] Further, in step S2, the trained semantic segmentation model includes a contraction path, a skip connection layer, and an expansion path connected in sequence;
[0023] The contraction path includes sequentially connected convolutional layers, rectified linear units, and max pooling layers;
[0024] The size of the convolution layer is 3*3, the size of the maximum pooling layer is 2*2, and the stride of the maximum pooling layer is 2;
[0025] The expansion path includes at least one module for recovering the feature size of the image;
[0026] Each module for restoring the image feature size consists of a feature map upsampling layer, a cascade layer, two image convolution layers, and two ReLU activation layers connected in sequence;
[0027] The size of the image convolution layer is 3*3.
[0028] The beneficial effect of the above further scheme is: improving the imbalance between positioning accuracy and context information caused by the different sizes of target image blocks, less context information of small-sized target image blocks and loss of target position information in the pooling layer image.
[0029] Furthermore, in step S2, the algorithm used by the height estimation model is a random forest model.
[0030] Furthermore, in step S3, the calculation formula used to calculate the ratio of pixels of the air conditioner outdoor unit in the instance segmentation result is:
[0031]
[0032] Among them, AB represents the pixel ratio of the air conditioner outdoor unit, A represents the number of pixels occupied by the air conditioner outdoor unit in the instance segmentation result, and B represents the number of pixels occupied by the building in the instance segmentation result.
[0033] Further, in step S4, the calculation formula used to determine the side wall area of the target area is as follows based on the urban building outline and building height:
[0034] MB=S*H
[0035] Among them, MB represents the side wall area, S represents the perimeter of the urban building outline, and H represents the building height.
[0036] Furthermore, step S5 specifically includes:
[0037] S51, partitioning the target area to obtain at least two partition areas;
[0038] The proportion of pixels of the air conditioner outdoor units in each partition area;
[0039] S52, calculating the ratio of the pixels of the air conditioner outdoor unit in each partition area according to the ratio of the pixels of the air conditioner outdoor unit in each partition area;
[0040] S53, determining the total area of the air conditioner outdoor units in the target area according to the side wall area and the pixel ratio of the air conditioner outdoor units in each partition area;
[0041] S54. Determine the number of household air conditioners in the target area based on the total area of the air conditioner outdoor units in the target area.
[0042] The beneficial effects of the above further scheme are: from easily accessible street view images, the pixel ratio of air-conditioning outdoor units and building walls is calculated, and the number of air conditioners is inferred; the building outline, height and distribution of air-conditioning outdoor units are combined and applied to the investigation and statistics of household air conditioners; deep learning methods are combined with ground survey methods to extract, quickly investigate and count household air conditioner information, saving manpower and material resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 The figure is a flow chart of a method for predicting the spatial distribution of household air conditioners.
[0044] Figure 2 Schematic diagram of remote sensing images of the target area.
[0045] Figure 3 A schematic diagram of the urban building outline of the target area.
[0046] Figure 4 Schematic diagram of street view image data.
[0047] Figure 5 This is a schematic diagram of the pixel ratio of the air conditioner outdoor unit.
[0048] Figure 6 It is a schematic diagram of a coordinate point diagram with AB ratio information. DETAILED DESCRIPTION
[0049] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.
[0050] like Figure 1 As shown, a method for predicting the spatial distribution of household air conditioners includes the following steps:
[0051] S1. Acquire remote sensing images and street view image data of a target area, preprocess the remote sensing images, and obtain preprocessed remote sensing images.
[0052] In this embodiment, in step S1, the remote sensing image is preprocessed, specifically including:
[0053] S11, performing geometric correction on the remote sensing image using a polynomial correction model to obtain a remote sensing image after geometric correction;
[0054] The remote sensing images after geometric correction include multispectral band images and panchromatic band images;
[0055] S12. Transform and fuse the multispectral band image and the panchromatic band image to obtain a fused remote sensing image.
[0056] For example, the remote sensing images are corrected and image fusion enhanced, and the fused images can be used for later object-oriented building foundation pit recognition, monomer extraction, shadow information extraction and building height inversion. At the same time, the feasibility of the recognition and extraction results is verified and the accuracy is evaluated.
[0057] For example, Figure 2 As shown, Figure 2 The figure is a schematic diagram of the remote sensing image of the target area. In the remote sensing image, the acquired ground object image information is inconsistent with the real ground object in combination, size, scale and direction, so geometric correction is required. A polynomial correction model can be used to geometrically transform the remote sensing image to make it consistent with the real ground object. High-resolution remote sensing image fusion is the process of transforming and fusing multi-spectral band image data and higher-resolution panchromatic band image data. By fusing these two different bands of image data, the amount of information and availability of the image can be increased, thereby better supporting subsequent tasks such as building foundation pit identification, monomer extraction, shadow information extraction, and building height inversion.
[0058] Optionally, in order to ensure the feasibility and accuracy of the recognition and extraction results, they can be verified and evaluated. By comparing and verifying with the actual situation, the accuracy and reliability of the recognition and extraction results can be determined. At the same time, the accuracy evaluation can be performed to evaluate the performance of the method in different scenarios and improve the credibility and reliability of subsequent applications.
[0059] S2. Input the preprocessed remote sensing image into the trained semantic segmentation model and the trained height estimation model, and output the urban building outline and building height of the target area through the semantic segmentation model and the height estimation model respectively.
[0060] In this embodiment, in step S2, the trained semantic segmentation model includes a contraction path, a skip connection layer, and an expansion path connected in sequence;
[0061] The contraction path includes sequentially connected convolutional layers, rectified linear units, and max pooling layers;
[0062] The size of the convolution layer is 3*3, the size of the maximum pooling layer is 2*2, and the stride of the maximum pooling layer is 2;
[0063] The expansion path includes at least one module for recovering the feature size of the image;
[0064] Each module for restoring the image feature size consists of a feature map upsampling layer, a cascade layer, two image convolution layers, and two ReLU activation layers connected in sequence;
[0065] The size of the image convolution layer is 3*3.
[0066] In this embodiment, in step S2, the algorithm used by the height estimation model is a random forest model.
[0067] Exemplarily, the trained semantic segmentation model can be a UNet network. The trained semantic segmentation model can be composed of a contraction path and an expansion path. The network layers corresponding to the contraction path and the expansion path can be connected by a jump connection structure. Different network layers in the contraction path contain feature maps with different degrees of fineness of position information, and different network layers in the expansion path contain upsampled feature maps with different degrees of richness of semantic information. The jump connection structure can combine semantic abstract information with different depths and position information with different degrees of fineness in different network layers, and realize the fusion between feature maps containing dense position information and feature maps containing rich semantic information. While realizing pixel classification, the accuracy of pixel positioning is improved, and the semantic segmentation accuracy of image targets is further improved. As the network depth increases, the activation map gradually pays more attention to abstract information, and linear jump connections are used to improve the gradient flow between a large number of network layers. The UNet network improves the imbalance between positioning accuracy and context information caused by different sizes of target image blocks, less context information of small-sized target image blocks, and loss of image target position information in the pooling layer.
[0068] For example, Figure 3 As shown, Figure 3 The figure is a schematic diagram of the urban building outline in the target area. The upsampling part in the expansion path has a large number of feature channels, which helps to propagate context information to subsequent network layers with higher resolution and achieve accurate pixel prediction. The UNet network uses a weighted loss function to give a large weight to the background label, thereby achieving separation between contact targets of the same category.
[0069] For example, a UNet network can be composed of a contraction path and an expansion path. The contraction path imitates the typical convolutional network structure and can be composed of repeated 3*3 convolutional layers. Each convolutional layer can be connected to a rectified linear unit and a 2*2 maximum pooling layer with a stride of 2 to achieve image downsampling, and double the number of feature channels in each downsampling step.
[0070] In each module of the expansion path that restores the image feature size, a 2*2 feature map upsampling layer can be included, followed by a cascade layer, two 3*3 convolution layers, and two ReLU activation layers. The 2*2 upsampling layer halves the feature channel. The cascade layer cascades the upsampling layer output with the feature map corresponding to the compression path. Each 3*3 convolution layer is followed by a ReLU activation layer. Since each convolution layer has a boundary pixel loss problem, the feature map from the compression path can be clipped. The last layer of the network can be a 1*1 convolution layer, which assigns each feature vector containing 64 components to the category label predicted by the pixel.
[0071] Optionally, after extracting the urban building outlines of the target area, the urban building outlines of the target area can be imported into ArcGIS software in the form of polygon layers to prepare for the next step of data fusion.
[0072] Exemplarily, a random forest model (RF model) may be used to construct and optimize the height estimation model to obtain a trained height estimation model.
[0073] For example, an RF regression model is used to estimate building height, where a total of 519 variables such as shadow area, optical texture features, radar data, etc. can be set as predictors. The number of trees (Ntree) and the number of features randomly selected to split each node (Mtry) can be two key parameters of the RF model, so increasing Ntree can improve the performance of random forests. Since the RF classifier is computationally efficient and non-overfitting, Ntree can be set to the highest feasible value. In a feasible embodiment, the Ntree value can be set to 500, and the classification tree is stable before reaching 500. When Mtry is small, the variance of the height estimation model decreases, and the deviation increases because some key features may be ignored. When Mtry is large, the variance of the height estimation model increases, but the deviation decreases as the height estimation model considers more features.
[0074] Alternatively, the RF regression model can be directly performed on the Google Earth Engine (GEE) platform. After obtaining the building heights, the building heights can be imported into ArcGIS software to prepare for the next step of data fusion.
[0075] S3. Input the street view image data into the trained instance segmentation model, output the instance segmentation result of the target area through the instance segmentation model, and calculate the ratio of the air conditioner outdoor unit pixels in the instance segmentation result.
[0076] In this embodiment, in step S3, the calculation formula used to calculate the ratio of pixels of the air conditioner outdoor unit in the instance segmentation result is:
[0077]
[0078] Among them, AB represents the pixel ratio of the air conditioner outdoor unit, A represents the number of pixels occupied by the air conditioner outdoor unit in the instance segmentation result, and B represents the number of pixels occupied by the building in the instance segmentation result.
[0079] For example, Figure 4 As shown, Figure 4 The figure is a schematic diagram of street view image data. Baidu Maps has collected panoramic data for the streets of 652 cities in China, covering a mileage of 2.295 million kilometers. In most cities, users can access the street view of the road on Baidu Maps. Therefore, after converting the coordinate system through the API interface, the 360° street view image data of the target location or street view image data of any direction and any angle range can be obtained according to the coordinate point. Optionally, positioning once every 25m from the set road network coordinates can obtain complete and coherent street view image data.
[0080] Alternatively, if Figure 5 As shown, Figure 5 The pixel ratio of the air conditioner outdoor unit can be combined with the original longitude and latitude coordinate information of the street view image data to produce a coordinate point diagram with AB ratio information, such as Figure 6 As shown, Figure 6 The coordinate point diagram with AB ratio information is shown in Figure 1. The coordinate point diagram with AB ratio information is imported into Argis software to prepare for the next step of data fusion.
[0081] S4. Determine the side wall area of the target area based on the urban building outline and building height.
[0082] In this embodiment, in step S4, the calculation formula used to determine the side wall area of the target area based on the urban building outline and building height is:
[0083] MB=S*H
[0084] Among them, MB represents the side wall area, S represents the perimeter of the urban building outline, and H represents the building height.
[0085] S5. Determine the number of household air conditioners in the target area based on the side wall area and the pixel ratio of the air conditioner outdoor unit.
[0086] In this embodiment, step S5 specifically includes:
[0087] S51, partitioning the target area to obtain at least two partition areas;
[0088] The proportion of pixels of the air conditioner outdoor units in each partition area;
[0089] S52, calculating the ratio of the pixels of the air conditioner outdoor unit in each partition area according to the ratio of the pixels of the air conditioner outdoor unit in each partition area;
[0090] S53, determining the total area of the air conditioner outdoor units in the target area according to the side wall area and the pixel ratio of the air conditioner outdoor units in each partition area;
[0091] S54. Determine the number of household air conditioners in the target area based on the total area of the air conditioner outdoor units in the target area.
[0092] For example, due to factors such as the house type and lighting, the number of air conditioner external units on the four sides of the building is not consistent, so the proportion data of a certain point cannot represent the proportion of the entire building. In order to reduce errors in this regard, the complete urban area can be decomposed into several 100m*100m sub-areas, each of which contains several coordinate points with air-conditioned building area proportion data. These points contain the situation in all directions of the building, and the average value of these points is used to represent the unified proportion mAB of the 100m*100m sub-area:
[0093]
[0094] Among them, mAB represents the uniform ratio of the partition area, SUM represents the sum of the ratios of all coordinate points in the partition area, and N represents the number of coordinate points in the partition area.
[0095] For example, the total area of the air conditioner outdoor units in the target area can be determined based on the side wall area and the pixel ratio of the air conditioner outdoor units in each partition area. The calculation formula can be:
[0096] MA=MB*mAB
[0097] Among them, MA represents the total area of air-conditioning outdoor units in the target area, MB represents the side wall area, and mAB represents the uniform ratio of the partitioned area.
[0098] For example, since the volumes of the outdoor units of air conditioners with different capacities are different, considering the volumes and usage rates of the outdoor units of air conditioners with different capacities, 0.5 m2 can be used as the area of an outdoor unit of an air conditioner, and the number of household air conditioners in the target area can be determined based on the total area of the outdoor units of air conditioners in the target area. The calculation formula can be:
[0099]
[0100] Where n represents the number of household air conditioners in the target area, and MA represents the total area of air conditioner outdoor units in the target area.
[0101] The present invention can extract information such as the height and outline area of residential buildings from high-resolution remote sensing images, then use street view image information to extract the proportion of the area occupied by air conditioners on residential buildings, and determine the spatial distribution of these targets by extracting the spatial position information of the targets. A functional relationship between the number of building air conditioners and the building height and outline area is established to count the number of residential home air conditioners and analyze their spatial distribution, thereby improving the accuracy of the statistics on the number of residential home air conditioners.
[0102] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.
Claims
1. A method for predicting the spatial distribution of household air conditioners, characterized in that: The method comprises: S1, acquiring remote sensing images and street view image data of a target area, and preprocessing the remote sensing images to obtain preprocessed remote sensing images; S2, inputting the pre-processed remote sensing image into a trained semantic segmentation model and a trained height estimation model, and outputting the urban building outline and building height of the target area through the semantic segmentation model and the height estimation model respectively; S3, inputting the street view image data into a trained instance segmentation model, outputting the instance segmentation result of the target area through the instance segmentation model, and calculating the ratio of air conditioner outdoor unit pixels in the instance segmentation result; S4. Determine the side wall area of the target area according to the urban building outline and the building height; S5. Determine the number of household air conditioners in the target area according to the side wall area and the pixel ratio of the air conditioner outdoor unit.
2. The method according to claim 1, characterized in that: In the step S1, the remote sensing image is preprocessed, specifically including: S11, performing geometric correction on the remote sensing image using a polynomial correction model to obtain a remote sensing image after geometric correction; The geometrically corrected remote sensing image includes a multispectral band image and a panchromatic band image; S12, transforming and fusing the multispectral band image and the panchromatic band image to obtain a fused remote sensing image.
3. The method according to claim 1, characterized in that In step S2, the trained semantic segmentation model includes a contraction path, a skip connection layer, and an expansion path connected in sequence; The contraction path includes a convolutional layer, a rectified linear unit, and a maximum pooling layer connected in sequence; The size of the convolutional layer is 3*3, the size of the maximum pooling layer is 2*2, and the stride of the maximum pooling layer is 2; The expansion path includes at least one module for restoring the feature size of the image; Each module for restoring the feature size of an image includes a feature map upsampling layer, a cascade layer, two image convolution layers, and two ReLU activation layers connected in sequence; The size of the image convolution layer is 3*3.
4. The method according to claim 1, characterized in that: In step S2, the algorithm used by the height estimation model is a random forest model.
5. The method according to claim 1, characterized in that In step S3, the calculation formula used to calculate the ratio of pixels of the air conditioner outdoor unit in the instance segmentation result is: Among them, AB represents the pixel ratio of the air conditioner outdoor unit, A represents the number of pixels occupied by the air conditioner outdoor unit in the instance segmentation result, and B represents the number of pixels occupied by the building in the instance segmentation result.
6. The method according to claim 1, characterized in that In step S4, the calculation formula used to determine the side wall area of the target area based on the urban building outline and the building height is: MB=S*H Among them, MB represents the side wall area, S represents the perimeter of the urban building outline, and H represents the building height.
7. The method according to claim 1, characterized in that The step S5 specifically includes: S51, partitioning the target area to obtain at least two partitioned areas; The ratio of pixels of the air conditioner outdoor unit contained in each partition area; S52, calculating the air conditioner outdoor unit pixel ratio of each partitioned area according to the air conditioner outdoor unit pixel ratio of each partitioned area; S53, determining the total area of the air conditioner outdoor units in the target area according to the side wall area and the ratio of the pixels of the air conditioner outdoor units in each partition area; S54: Determine the number of household air conditioners in the target area according to the total area of the air conditioner outdoor units in the target area.
Citation Information
Patent Citations
Automatic extraction method for dragon fruit single plant identification in a karst region
CN109919088A
Household air conditioner identification method, device and equipment, storage medium and product
CN114187524A
Area determination method and device, electronic equipment and storage medium
CN115908535A
Image recognition method and system
CN117058417A
