A greenhouse interpretation and extraction method and system based on sub-meter remote sensing images

Abstract

The present invention discloses a greenhouse interpretation and extraction method, system, electronic device, and computer-readable storage medium based on sub-meter remote sensing imagery, belonging to the field of remote sensing image processing technology. The present invention targets high-resolution remote sensing images and, based on water network and road network vector files, clips the region of interest mask file, dividing it into dense greenhouse areas and non-dense greenhouse areas. Based on the distribution of different greenhouses, different algorithms are selected to effectively extract the edges of the greenhouses, thereby improving the efficiency and accuracy of greenhouse extraction and ensuring the integrity of greenhouse extraction. This solves the existing problems of "difficulty in greenhouse extraction, low extraction accuracy, and low efficiency in smart agriculture" in the prior art.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a greenhouse interpretation and extraction method and system based on sub-meter remote sensing images. Background Art

[0002] The statements in this section merely mention the background technology related to this application and do not necessarily constitute the prior art.

[0003] With the rapid development of modern agriculture in my country, the number of greenhouses has increased significantly to meet the demand for greenhouse vegetables. In order to facilitate agricultural management and soil protection, it is particularly important to obtain accurate and timely spatial distribution information of greenhouses.

[0004] In the context of informatization, smart agriculture is gaining increasing attention. In the process of interpreting and extracting greenhouse information, traditional manual visual interpretation is time-consuming and labor-intensive. Machine learning, a more in-depth study of artificial intelligence, is now being applied to remote sensing image extraction, achieving intelligent and automated remote sensing image interpretation, extracting land information, and significantly saving human resources. However, in the field of remote sensing image extraction for agricultural greenhouses, the following problems still exist:

[0005] (1) Existing deep learning methods require large-scale labeled samples to train convolutional neural networks (CNNs) to achieve excellent performance. Labeled samples applied to greenhouses often cannot guarantee the accuracy of greenhouse edge extraction.

[0006] (2) The extraction integrity and accuracy of agricultural greenhouses are not high enough;

[0007] (3) It is difficult to extract the dense greenhouse interpretation into a single greenhouse interpretation. Summary of the Invention

[0008] In order to address the shortcomings of the existing technology, the present application provides a greenhouse interpretation and extraction method and system based on sub-meter remote sensing images, which adopts a deep learning method that combines the edge detection network (HED) model and the semantic segmentation network (D_LinkNet) model to make the greenhouse extraction more accurate, thereby improving the accuracy of remote sensing image extraction of agricultural greenhouses.

[0009] In the first aspect, the present application provides a greenhouse interpretation and extraction method based on sub-meter remote sensing images;

[0010] A greenhouse interpretation and extraction method based on sub-meter remote sensing images, comprising:

[0011] Acquire sub-meter remote sensing images of the mission area and perform preprocessing to obtain remote sensing images of the area of ​​interest;

[0012] Based on the water surface vector file and the road surface vector file, the remote sensing images of the area of ​​interest are divided into images of greenhouse-intensive areas and images of greenhouse-unintensive areas;

[0013] Based on the pre-processed sub-meter remote sensing images of the mission area, a sample dataset is constructed to train edge line detection models and texture surface detection models.

[0014] The trained texture surface detection model is used to predict images of dense greenhouse areas and extract continuous dense greenhouses; the trained edge line detection model is used to predict images of non-dense greenhouse areas and extract discrete individual greenhouses;

[0015] The extraction results are post-processed to obtain greenhouse prediction extraction results.

[0016] In the second aspect, this application provides a greenhouse interpretation and extraction system based on sub-meter remote sensing images;

[0017] A greenhouse interpretation and extraction system based on sub-meter remote sensing images, including:

[0018] The preprocessing module is configured to: acquire sub-meter remote sensing images and perform preprocessing to obtain remote sensing images of the area of ​​interest;

[0019] The image partitioning module is configured to: divide the remote sensing image of the area of ​​interest into images of greenhouse-dense areas and images of greenhouse-non-dense areas based on the water surface vector file and the road surface vector file;

[0020] The model training module is configured to: construct a sample dataset based on the pre-processed sub-meter remote sensing images of the mission area, and train the edge line detection model and the texture surface detection model;

[0021] The greenhouse extraction module is configured to: use the trained texture surface detection model to predict images of densely populated areas with greenhouses and extract continuous dense greenhouses; use the trained edge line detection model to predict images of non-dense areas with greenhouses and extract discrete individual greenhouses; post-process the extraction results to obtain greenhouse prediction and extraction results.

[0022] In a third aspect, the present application provides an electronic device;

[0023] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps of the above-mentioned greenhouse interpretation and extraction method based on sub-meter remote sensing images are completed.

[0024] In a fourth aspect, the present application provides a computer-readable storage medium;

[0025] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps of the greenhouse interpretation and extraction method based on sub-meter remote sensing images are completed.

[0026] Compared with the prior art, the present invention has the following advantages:

[0027] 1. This application uses high-resolution remote sensing imagery, applies edge detection algorithms and semantic segmentation network model algorithms, selects different algorithms for the distribution of different greenhouses, conducts intelligent interpretation of remote sensing image information, accurately identifies agricultural land information within the map, and realizes intelligent extraction of fine vector patches of greenhouses in areas of interest, thereby improving the efficiency and accuracy of greenhouse extraction and solving the problem of difficult agricultural greenhouse extraction in smart agriculture under the background of informatization.

[0028] 2. This application innovatively combines the extraction of water networks, road networks and greenhouses, interprets and extracts greenhouse-dense areas into single greenhouses for further extraction, reduces missed greenhouse extraction, improves the extraction rate of greenhouses, and effectively ensures the integrity and accuracy of the extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings in the specification, which constitute a part of this application, are used to provide further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute improper limitations on this application.

[0030] Figure 1 A schematic diagram of the logic flow of the greenhouse interpretation and extraction method based on sub-meter remote sensing images provided in an embodiment of the present application;

[0031] Figure 2 A flow chart of a greenhouse interpretation and extraction method based on sub-meter remote sensing images provided in an embodiment of the present application;

[0032] Figure 3 A schematic diagram of an example partition provided in an embodiment of the present application;

[0033] Figure 4 A schematic diagram of the sample production process provided in the embodiment of this application;

[0034] Figure 5 A schematic diagram of an example of sample point coverage provided in an embodiment of the present application;

[0035] Figure 6 A schematic diagram of a greenhouse line sample example provided in an embodiment of the present application;

[0036] Figure 7 A schematic diagram of a greenhouse surface sample example provided in an embodiment of the present application;

[0037] Figure 8 This is an example diagram of extracting continuous and dense greenhouses provided in an embodiment of the present application;

[0038] Figure 9 This is an example diagram of discrete greenhouse extraction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by those skilled in the art to which this application belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0042] Example 1

[0043] In the existing technology, the deep learning method used for greenhouse extraction cannot guarantee the accuracy and integrity of greenhouse edge extraction, and it is difficult to cope with different greenhouse distributions; therefore, this application provides a greenhouse interpretation and extraction method based on sub-meter remote sensing images.

[0044] A greenhouse interpretation and extraction method based on sub-meter remote sensing images, comprising:

[0045] Acquire sub-meter remote sensing images of the mission area and perform preprocessing to obtain remote sensing images of the area of ​​interest;

[0046] Based on the water surface vector file and the road surface vector file, the remote sensing images of the area of ​​interest are divided into images of greenhouse-intensive areas and images of greenhouse-unintensive areas;

[0047] Based on the pre-processed sub-meter remote sensing images of the mission area, a sample dataset is constructed to train edge line detection models and texture surface detection models.

[0048] The trained texture surface detection model is used to predict images of dense greenhouse areas and extract continuous dense greenhouses; the trained edge line detection model is used to predict images of non-dense greenhouse areas and extract discrete individual greenhouses;

[0049] The extraction results are post-processed to obtain greenhouse prediction extraction results.

[0050] Furthermore, the sub-meter remote sensing images of the mission area are pre-processed. The specific steps to obtain the remote sensing images of the area of ​​interest are as follows:

[0051] Perform orthorectification, image registration, image fusion, and uniform color mosaic operations on sub-meter remote sensing images;

[0052] According to the distribution range of greenhouses, the pre-processed sub-meter remote sensing images are cropped to obtain remote sensing images of the area of ​​interest.

[0053] Furthermore, based on the pre-processed sub-meter remote sensing images of the mission area, the specific steps for constructing a sample dataset are as follows:

[0054] Based on the remote sensing images of the area of ​​interest, the sub-meter remote sensing images of the pre-processed mission area are cropped and sampled with the same size, and binary labels corresponding to the sampled samples are produced to fill the sample dataset.

[0055] Furthermore, the trained texture surface detection model is used to predict the image of the dense greenhouse area. The specific steps for extracting continuous dense greenhouses include:

[0056] Use the trained texture surface detection model to predict the image of the greenhouse dense area and generate predicted grayscale raster data;

[0057] Convert the predicted grayscale raster data into surface vector data, and perform square and edge simplification on the surface vector data;

[0058] Based on the road surface vector file, remove the greenhouse patches on the road surface in the surface vector data.

[0059] Furthermore, the trained edge line detection model is used to predict the image of the non-dense greenhouse area. The specific steps to extract and separate individual greenhouses include:

[0060] Use the trained edge line detection model to predict the image of the non-dense area of ​​the greenhouse and generate predicted grayscale raster data;

[0061] Convert the predicted grayscale raster data into line vector data, and construct the line vector data into surface vector data;

[0062] Based on the road surface vector file, remove the greenhouse patches on the road surface in the surface vector data.

[0063] Furthermore, the extraction results are post-processed to obtain greenhouse prediction extraction results in the following specific steps:

[0064] The extracted discrete single greenhouses and continuous dense greenhouses are merged, and based on the road network vector file and water network vector file, the incorrectly extracted rivers and roads are removed to obtain the greenhouse prediction extraction results.

[0065] Furthermore, the edge line detection model is the HED edge detection model, and the texture surface detection model is the D_LinkNet semantic segmentation network model.

[0066] Next, combine Figure 1-9 A greenhouse interpretation and extraction method based on sub-meter remote sensing images disclosed in this embodiment is described in detail. The greenhouse interpretation and extraction method based on sub-meter remote sensing images includes the following steps:

[0067] S1. Select and pre-process sub-meter remote sensing images of the mission area. Resolution represents the smallest unit that can distinguish surface images. Common satellite resolutions range from 0.5m to 15m. Since the width of a greenhouse is as small as a few meters, higher-resolution remote sensing images are required. In this embodiment, remote sensing images from the Gaojing-1 satellite with a full-color resolution of 0.5m are selected. The specific steps include:

[0068] S101. Perform orthorectification, image registration, image fusion, and uniform color mosaic operations on sub-meter remote sensing images.

[0069] S102. Based on the distribution range of the greenhouses, the pre-processed sub-meter remote sensing image is cropped to obtain a remote sensing image of the area of ​​interest.

[0070] S2. Based on the water network and road network vector files, the mask file of the mission area is cropped and divided into images of greenhouse-dense areas and images of greenhouse-non-dense areas. Among them, the mask file refers to a specific vector file that blocks the processed image to control the image processing area. The control partition generally refers to the background information such as terrain, topography, and landform. According to the linear features with obvious edge features in the remote sensing image such as roads, water systems, ridges, and valleys, a closed (preferably convex polygon) surface vector range is formed. In this embodiment, based on the 0.5-meter resolution remote sensing image and the road network data, the geographical partition is drawn according to the different textures, colors, and forms presented by the greenhouses. The mask file is a specific vector file used to process the remote sensing image of the mission area.

[0071] The basis for controlling the drawing of partitions is:

[0072] (1) The primary criterion for partitioning boundaries is not to destroy the extraction target;

[0073] (2) The area of ​​a single zone shall be controlled within 400 square kilometers;

[0074] (3) The types of features in the same area should be kept consistent as much as possible.

[0075] S3. Based on the remote sensing image of the area of ​​interest, the sub-meter remote sensing image of the pre-processed task area is cropped and sampled with the same size, and binary labels corresponding to the sample samples are produced to fill the sample data set; specifically, the sample production is divided into three steps: (1) Based on the greenhouse image, sample points are drawn based on the "draw" tool in ArcMap, and sample points (areas) are drawn randomly; (2) The remote sensing image is cropped with the sample point as the center: the image is cropped according to the sample point selected in the previous step. The cropping method is to draw a square as the sample drawing area with a certain distance from the sample point as the center, forming a 1000*1000 pixel image and the corresponding outer boundary vector data; (3) Draw the sample: in each sample vector cropped above, the greenhouse is drawn with the image as the base map.

[0076] S4. Based on the sample dataset, train an edge line detection model and a texture surface detection model; wherein the edge line detection model is the HED edge detection model, and the texture surface detection model is the D_LinkNet semantic segmentation network model;

[0077] 1. The HED edge detection model uses VGG-16 as the backbone network. The sample images in the sample dataset are input into the HED edge detection model to train the model. The specific process is as follows:

[0078] The HED edge detection model uses cropped 1000×1000, dimension 3 image files from the training set. Five convolutions and four max pooling operations are performed on the image files. The last layer of each convolution is used as the outer output of the network, resulting in five outer outputs. These outer outputs are images with enhanced edges after these operations.

[0079] Since the sizes of the outer outputs are inconsistent, each outer output is deconvolved in the network. The deconvolution uses the transposed convolution operation to expand it to the output size, and the five layers of output are fused to obtain a fusion layer. The HED edge detection model has a total of six layers of output; the final output result is a feature map with a size of 1000×1000 and a dimension of one after the five layers of side output layers are expanded to the size of the original input image and cascaded and fused.

[0080] The loss function is used to assess the degree of deviation between the model's predictions and the true values. Generally speaking, the better the loss function, the better the model's performance, and the smaller the loss function, the more robust the model. Due to the significant difference in the number of edge and non-edge pixels in edge extraction, the HED edge detection model improves on the traditional loss function by weighting the loss calculation for edges and non-edges accordingly.

[0081] The outer total loss function of the HED edge detection model is:

[0082]

[0083] Where W is all the parameter values ​​of the VGG-16 network, w (M) is the parameter value of the mth side branch. If the network has M side branches, then the parameter value of the side branch is defined as w=(w (1) ,…,w (M) ),α m Represents the weight of the loss function of each side branch, which can be adjusted according to the training log. is the inter-class balanced cross entropy loss function for the mth outer output, defined as:

[0084]

[0085] ∑ j∈Y- logP r (y j =0|X;W,w (m) )

[0086] Among them, β is suitable for balancing the imbalanced category balance weights of positive and negative samples in edge detection, where Y - Y is the set of pixels marked as edges in the ground truth; + is a set of non-edge pixels; represents the edge value predicted by the mth side branch at the jth pixel, and σ() is the sigmoid activation function.

[0087] The loss function of the fusion layer is:

[0088]

[0089]

[0090] Where: h m is the fusion weight; Output value for the outer side of each layer; is the value after the outer output is fused and activated, then the fusion layer loss function is calculated The distance between Y, Dist is calculated using the cross entropy loss function, and σ is the sigmoid activation function.

[0091] The final optimization objective function is:

[0092] (W,w,h) *=argmin(L fuse (W,w,h)+L side (W,w)) (4)

[0093] That is, minimize the side branch loss L side (W, w) and fuse loss L fuse The sum of (W,w,h).

[0094] 2. The D-LinkNet semantic segmentation network model uses LinkNet, a pre-trained encoder, as its backbone, with additional dilated convolutional layers in the center. LinkNet uses ResNet34 as its encoder, a highly efficient semantic segmentation neural network that utilizes skip connections, residual blocks, and an encoder-decoder architecture. The training process for the D-LinkNet semantic segmentation network model is as follows:

[0095] The D-LinkNet semantic segmentation network model's encoding structure consists of a ResNet34 architecture. Image files of size 1000×1000 and dimension 3, cropped from the training set, are fed into the model. The images are first downsampled, resulting in feature maps of size 500×500 and dimension 64. After four further downsampling steps, the feature maps are 32×32 and dimension 512. The dilation rates of the cascaded dilated convolutional layers are 1, 2, 4, and 8, respectively. After adding the dilated convolutional layers, the feature receptive fields at the center of the feature maps increase to 3, 7, 5, and 3. After the dilated convolutional layers, the feature maps maintain their size and number of layers at 32×32 and dimension 512. The decoding structure of the network adopts the decoding structure of the LinkNet network. After four upsampling, the features are converted back to a 500×500×64 feature map, and finally converted to a 1000×1000 feature map through two deconvolutions. The pixels in the feature map are used for binary classification output using the sigmoid function.

[0096] The loss function of the D-LinkNet34 neural network is: BCE Loss + soft Dice Loss. The loss function dice_loss is combined with bce_loss to improve the stability of model training.

[0097] (1) In BCE Loss, for a batch of N samples D(x,y), each sample may have M labels, and the loss is calculated as follows:

[0098]

[0099] in, is the loss corresponding to the nth sample.

[0100]

[0101] Indicates that the true category in the nth sample is equal to i and takes 1, otherwise it takes 0 (0 or 1), Indicates the probability that the nth sample belongs to category i, and the probability is obtained by the sigmoid function. i is a hyperparameter used to address the problem of sample imbalance between labels. For a batch of training sets, if a label appears less frequently, it should be given a higher weight when calculating the loss.

[0102] (2) In soft Dice Loss, soft Dice Loss = 1-soft dice coeff, where softdice coeff is the dice coefficient, and the formula is:

[0103]

[0104] X represents the Ground Truth segmentation image, and Y represents the predicted segmentation image.

[0105] Based on existing sample data and deep learning techniques, model parameters are determined through model training. The resulting model is the training result. During training, the absolute paths of sample images and corresponding labeled images are recorded in the "train.txt" file. A specific number of samples are selected and fed into the deep learning network. Training is repeated for a set number of training cycles, ultimately forming a model. By training on samples from the task area, a specific mathematical result is used to obtain a model for target extraction in the task area, which is then used to make predictions for the task area.

[0106] S5. Extract greenhouses. Greenhouse plots are land use units formed by artificial surface transformation. Their texture features are richer than their edge information and are less susceptible to interference. However, in areas with dense greenhouses, texture structures cannot be well distinguished. Therefore, the trained HED edge detection model is used in areas with dense greenhouses, and the trained D_LinkNet semantic segmentation network model is used in areas with less dense greenhouses. The specific steps include:

[0107] S501, continuous and dense greenhouse extraction:

[0108] (1) Use the trained HED edge detection model to predict the image of the greenhouse dense area and generate predicted grayscale raster data;

[0109] (2) Raster to vector conversion: Convert the predicted grayscale raster data into surface vector data in ArcMap software;

[0110] (3) Rectangularization and edge simplification: Simplify the curved map boundary into a regular rectangle in ArcMap software or open source GIS tools;

[0111] (4) Road and water network masking: In ArcMap software, the existing road network vector data is used to mask the greenhouse surface extraction results to achieve the purpose of removing the greenhouse spots on the road surface that were mistakenly extracted;

[0112] (5) Debris removal: In ArcMap software, the patches are filtered based on the area, perimeter, and perimeter-to-area ratio (the square of the perimeter divided by 16 times the area) to remove small patches and incorrectly selected patches.

[0113] S502, discrete separation of single greenhouse extraction:

[0114] (1) Use the trained D_LinkNet semantic segmentation network model to predict the non-dense area image of the greenhouse and generate predicted grayscale raster data;

[0115] (2) Raster to vector conversion: converting the predicted grayscale raster data into line vector data;

[0116] (3) Line vector data is constructed into surface vector data;

[0117] (4) Road water network mask: Use the existing road surface vector data to mask the greenhouse line extraction results to achieve the purpose of removing the greenhouse spots on the road surface that were mistakenly extracted.

[0118] (5) Debris removal: Filter patches based on area, perimeter, and perimeter-to-area ratio (the square of the perimeter divided by 16 times the area) to remove small patches and incorrectly selected patches.

[0119] S503: Merge the extracted discrete single greenhouses and continuous dense greenhouses, remove the erroneously extracted rivers and roads based on the road network vector file and the water network vector file, and obtain the greenhouse prediction extraction results.

[0120] Example 2

[0121] This embodiment discloses a greenhouse interpretation and extraction system based on sub-meter remote sensing images, including:

[0122] The preprocessing module is configured to: acquire sub-meter remote sensing images and perform preprocessing to obtain remote sensing images of the area of ​​interest;

[0123] The image partitioning module is configured to: divide the remote sensing image of the area of ​​interest into images of greenhouse-dense areas and images of greenhouse-non-dense areas based on the water surface vector file and the road surface vector file;

[0124] The model training module is configured to: construct a sample dataset based on the pre-processed sub-meter remote sensing images of the mission area, and train the edge line detection model and the texture surface detection model;

[0125] The greenhouse extraction module is configured to: use the trained texture surface detection model to predict images of densely populated areas with greenhouses and extract continuous dense greenhouses; use the trained edge line detection model to predict images of non-dense areas with greenhouses and extract discrete individual greenhouses; post-process the extraction results to obtain greenhouse prediction and extraction results.

[0126] It should be noted that the preprocessing module, image partitioning module, model training module, and greenhouse extraction module described above correspond to the steps in Example 1. The examples and application scenarios implemented by these modules and the corresponding steps are the same, but are not limited to the content disclosed in Example 1. It should be noted that the above modules, as part of a system, can be executed in a computer system, such as a set of computer-executable instructions.

[0127] Example 3

[0128] A third embodiment of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the above steps are completed.

[0129] Example 4

[0130] A fourth embodiment of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the above steps are completed.

[0131] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0134] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0135] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A greenhouse interpretation and extraction method based on sub-meter remote sensing images, characterized by: include: Acquire sub-meter remote sensing images of the mission area and perform preprocessing to obtain remote sensing images of the area of ​​interest; Based on the water surface vector file and the road surface vector file, the remote sensing images of the area of ​​interest are divided into images of greenhouse-intensive areas and images of greenhouse-unintensive areas; Based on the pre-processed sub-meter remote sensing images of the mission area, a sample dataset is constructed to train edge line detection models and texture surface detection models. The trained texture surface detection model is used to predict images of dense greenhouse areas and extract continuous dense greenhouses; the trained edge line detection model is used to predict images of non-dense greenhouse areas and extract discrete individual greenhouses; The extraction results are post-processed to obtain greenhouse prediction extraction results.

2. The greenhouse interpretation and extraction method based on sub-meter remote sensing images according to claim 1 is characterized in that: The specific steps for preprocessing the sub-meter remote sensing images of the mission area and obtaining remote sensing images of the area of ​​interest are as follows: Perform orthorectification, image registration, image fusion, and uniform color mosaic operations on sub-meter remote sensing images; According to the distribution range of greenhouses, the pre-processed sub-meter remote sensing images are cropped to obtain remote sensing images of the area of ​​interest.

3. The greenhouse interpretation and extraction method based on sub-meter remote sensing images according to claim 1 is characterized in that: The specific steps of constructing a sample dataset based on the pre-processed sub-meter remote sensing image of the mission area are as follows: Based on the remote sensing images of the area of ​​interest, the sub-meter remote sensing images of the pre-processed mission area are cropped and sampled with the same size, and binary labels corresponding to the sampled samples are produced to fill the sample dataset.

4. The greenhouse interpretation and extraction method based on sub-meter remote sensing images according to claim 1 is characterized in that: The specific steps of using the trained texture surface detection model to predict the image of the dense greenhouse area and extracting continuous dense greenhouses include: Use the trained texture surface detection model to predict the image of the greenhouse dense area and generate predicted grayscale raster data; Convert the predicted grayscale raster data into surface vector data, and perform square and edge simplification on the surface vector data; Based on the road surface vector file, remove the greenhouse patches on the road surface in the surface vector data.

5. The greenhouse interpretation and extraction method based on sub-meter remote sensing images according to claim 1 is characterized in that: The specific steps of using the trained edge line detection model to predict the image of the non-dense area of ​​greenhouses and extracting and separating individual greenhouses include: Use the trained edge line detection model to predict the image of the non-dense area of ​​the greenhouse and generate predicted grayscale raster data; Convert the predicted grayscale raster data into line vector data, and construct the line vector data into surface vector data; Based on the road surface vector file, remove the greenhouse patches on the road surface in the surface vector data.

6. The greenhouse interpretation and extraction method based on sub-meter remote sensing images according to claim 1 is characterized in that: The specific steps of post-processing the extraction results to obtain greenhouse prediction extraction results are: The extracted discrete single greenhouses and continuous dense greenhouses are merged, and based on the road network vector file and water network vector file, the incorrectly extracted rivers and roads are removed to obtain the greenhouse prediction extraction results.

7. The greenhouse interpretation and extraction method based on sub-meter remote sensing images according to claim 1 is characterized in that: The edge line detection model is the HED edge detection model, and the texture surface detection model is the D_LinkNet semantic segmentation network model.

8. A greenhouse interpretation and extraction system based on sub-meter remote sensing images, characterized by: include: The preprocessing module is configured to: acquire sub-meter remote sensing images and perform preprocessing to obtain remote sensing images of the area of ​​interest; The image partitioning module is configured to: divide the remote sensing image of the area of ​​interest into images of greenhouse-dense areas and images of greenhouse-non-dense areas based on the water surface vector file and the road surface vector file; The model training module is configured to: construct a sample dataset based on the pre-processed sub-meter remote sensing images of the mission area, and train the edge line detection model and the texture surface detection model; The greenhouse extraction module is configured to: use the trained texture surface detection model to predict images of densely populated areas with greenhouses and extract continuous dense greenhouses; use the trained edge line detection model to predict images of non-dense areas with greenhouses and extract discrete individual greenhouses; post-process the extraction results to obtain greenhouse prediction and extraction results.

9. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the steps described in any one of claims 1 to 7 are completed when the computer instructions are executed by the processor.

10. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the steps described in any one of claims 1 to 7.

Images