A remote sensing image matching method and system for farmland monitoring

By using deep learning models for cloud detection and removal, and combining them with image matching algorithms to select the optimal remote sensing images, the problem of low image utilization in farmland monitoring in cloudy and rainy areas of southern China has been solved, enabling rapid identification and precise management of farmland changes.

CN119992138BActive Publication Date: 2025-11-21SURVEYING & MAPPING INST LANDS & RESOURCE DEPT OF GUANGDONG PROVINCE
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510051119.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-01-13
Publication Date
2025-11-21
Estimated Expiration
2045-01-13

AI Technical Summary

Technical Problem

Existing technologies are unable to select the optimal remote sensing images that meet the actual requirements of farmland monitoring from a large number of satellite remote sensing images. In particular, the utilization rate of images is low in the cloudy and rainy areas of the south, resulting in inadequate identification of land type information in the monitoring area and making it impossible to quickly identify and accurately manage changes in farmland.

Method used

By acquiring a set of satellite remote sensing images that meet preset quality requirements, cloud detection and removal are performed using a deep learning model. Combined with image matching and selection algorithms, the optimal image after cloud removal, sorting, and fusion is selected and then vectorized to extract changed patches, thereby achieving the monitoring of the current status of cultivated land.

Benefits of technology

It enables rapid identification and precise management of changes in cultivated land under cloudy and rainy conditions, ensuring high resolution and timeliness of monitoring images to meet actual cultivated land monitoring requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119992138B_ABST
    Figure CN119992138B_ABST
Patent Text Reader

Abstract

The application relates to a remote sensing image matching method and system for farmland monitoring, and relates to the technical field of remote sensing image processing. The method comprises the following steps: using a satellite remote sensing image set to produce an image landing map set, then performing cloud removal processing on each image landing map to obtain an image range landing map, performing optimal matching on the basis of the landing map after cloud removal, selecting the optimal image after cloud removal and sorting and fusion from the image range landing map as a target image base map, performing vector processing on the target image base map according to obtained vector data to obtain change map patches in which vector elements and image characteristic land classes are inconsistent, and determining farmland present situation monitoring map patch results according to the change map patches. The application can realize real-time acquisition, analysis and processing of a large amount of remote sensing image data, select optimal remote sensing images from the remote sensing image data, ensure high resolution and time phase of the monitoring images, and realize rapid identification of farmland change conditions.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a remote sensing image matching method and system for cultivated land monitoring. BACKGROUND

[0002] With the increasing importance of cultivated land protection, traditional cultivated land protection methods have been unable to meet the needs of modern agricultural development. In recent years, modern information technologies such as satellite remote sensing have played an important role in dynamic monitoring and supervision of cultivated land protection. However, in the actual dynamic monitoring of cultivated land protection in southern regions, there are problems such as low image utilization rate due to cloudy and rainy weather, and incomplete identification of land information in the monitoring area.

[0003] To solve the above problems, the existing scheme mostly uses cloud detection methods when selecting remote sensing images, including threshold method, feature analysis method, homomorphic filtering method, and clustering analysis method, etc. However, these methods often have certain limitations. Specifically, the existing remote sensing image selection method only removes the cloud area of the remote sensing image, and the obtained cloud detection map is not the truly optimal remote sensing image. Moreover, the image obtained after removing the cloud area is not clear and cannot reflect the actual situation of cultivated land, so it cannot achieve rapid identification and accurate management of cultivated land changes. It can be seen that the existing technology cannot select the optimal remote sensing image that meets the actual cultivated land monitoring requirements from a large number of satellite remote sensing images. SUMMARY

[0004] The present application provides a remote sensing image matching method and system for cultivated land monitoring. In the dynamic monitoring of cultivated land protection, the optimal satellite remote sensing image is selected according to the needs of monitoring content and monitoring period, combined with various image conditions, and according to the principles, for comprehensive monitoring of the present situation of cultivated land. It can real-time acquire, analyze and process a large amount of remote sensing image data, realize rapid identification and accurate management of cultivated land changes, and ensure high resolution and timely phase of the monitoring image. It solves the problem that the existing technology cannot select the optimal remote sensing image that meets the actual monitoring requirements from a large number of satellite remote sensing images, and also solves the problem of limitation of available image range caused by cloud coverage under cloudy and rainy conditions.

[0005] In a first aspect, the present application provides a remote sensing image matching method for cultivated land monitoring, comprising:

[0006] acquiring a set of satellite remote sensing images that meet the preset quality requirements, and making an image footprint set according to the set of satellite remote sensing images;

[0007] performing cloud detection and removal processing on each image in the image footprint set through a preset image processing model, to obtain a cloud-removed image range footprint set;

[0008] According to the preset image matching and image selection algorithm, each image range map in the image range map set is optimally matched to obtain a target image base map, the target image base map being an optimal image selected from the image range map after cloud removal and sorting fusion;

[0009] According to the obtained vector data, the target image base map is vector processed to obtain a change patch, the change patch being a patch in which a vector element in the vector data is inconsistent with an image feature land class;

[0010] According to the change patch, a cultivated land status monitoring patch result is determined.

[0011] Optionally, a satellite remote sensing image set meeting preset quality requirements is obtained, including:

[0012] According to preset resolution requirements and image quality requirements, a first remote sensing image set and a second remote sensing image set are obtained from an image source;

[0013] The first satellite remote sensing image set and the second satellite remote sensing image set are respectively preprocessed and set corrected to obtain a orthographic image set meeting preset quality requirements after geometric distortion is eliminated, serving as the satellite remote sensing image set;

[0014] The orthographic image set includes a first orthographic image set corresponding to the first satellite remote sensing image set and / or a second orthographic image set corresponding to the second satellite remote sensing image set.

[0015] Optionally, each image in the image map set is cloud detected and removed by a preset image processing model to obtain a cloud-removed image range map set, including:

[0016] A cloud detection model in the preset image processing model is used to perform cloud detection on the image map set to obtain a two-class cloud mask map set;

[0017] According to the cloud mask map set, cloud region identification and post-processing are performed to obtain a batch cloud detection map set;

[0018] According to each cloud detection map in the batch cloud detection map set, a cloud region extracted from a corresponding image map is vector processed to obtain an image range map set after cloud layer range removal, the image range map set including an image range map.

[0019] Optionally, according to the preset image matching and image selection algorithm, each image range map in the image range map set is optimally matched to obtain a target image base map, including:

[0020] The image range fall map set is classified and marked to obtain corresponding marking data of the image range fall map, and the marking data includes a marking field with an assignment value;

[0021] According to the assignment value of the marking field, each image range fall map is sorted and matched in resolution and timeliness according to a preset image matching rule to obtain an effective image range fall map set with an image use order;

[0022] According to the effective image range fall map set, a mosaic fusion process is performed to obtain a fused remote sensing image fall map;

[0023] According to the fused remote sensing image fall map, a scene-by-scene retrieval matching is performed by using an image automatic retrieval algorithm to obtain a cloud-free optimal image as a target image base map;

[0024] The assignment value in the marking field includes a resolution assignment value and a time phase assignment value, and the time phase assignment value is used to determine the timeliness of the image fall map.

[0025] Optionally, the classification and marking of each image range fall map in the image range fall map set to obtain corresponding marking data of the image range fall map includes:

[0026] Each image range fall map is traversed, and a marking type is generated according to the resolution and time phase of the image range fall map;

[0027] According to the marking type, the resolution and time phase of the image range fall map are combined for assignment processing to obtain the marking data corresponding to the image range fall map.

[0028] Optionally, according to the assignment value of the marking field, each image range fall map is sorted and matched in resolution and timeliness according to a preset image matching rule to obtain an effective image range fall map set with an image use order, including:

[0029] The marking data of each image range fall map is used as a sorting element, a bubble sorting method is used, and a two-part matching pre-sorting is performed based on the marking field to obtain a pre-sorting result, the pre-sorting result including a resolution pre-sorting result corresponding to the resolution assignment value and a timeliness pre-sorting result corresponding to the time phase assignment value;

[0030] According to the resolution pre-sorting result, each image range fall map is divided in resolution based on a resolution priority rule to obtain a resolution sorting image set with at least one resolution;

[0031] According to the timeliness pre-sorting result, each image range fall map in the resolution sorting image set is divided in timeliness based on a timeliness priority rule to obtain an effective image range fall map set with an image use order;

[0032] The preordering result is used as an effective data basis of a data processing and matching process to form an ordered list for fast searching.

[0033] Optionally, the effective image range fall map set is subjected to mosaic fusion processing to obtain a fused remote sensing image fall map, including:

[0034] A target image range fall map with the first order of cloud removal is extracted from the effective image range fall map set, and the target image range fall map contains a target range vector.

[0035] A target cloud removal range is determined according to the target range vector, and a fused image range fall map set of a fused image range vector is selected from the effective image range fall map set according to the target cloud removal range.

[0036] The target range vector of the target image range fall map and the image range vector of each image range fall map in the fused image range fall map set are subjected to effective range fusion in a one-by-one fusion order until the target image range fall map is a complete layer, so as to obtain a fused remote sensing image fall map.

[0037] Optionally, according to the fused remote sensing image fall map, an image automatic retrieval algorithm is used for scene-by-scene retrieval matching to obtain a cloud-free optimal image, including:

[0038] An image automatic retrieval algorithm is used to search for an image on a target server according to an image name and a storage path corresponding to the fused remote sensing image fall map, so as to obtain a retrieval image.

[0039] The retrieval image is used as an optimal image selected in a final actual monitoring range.

[0040] Optionally, according to the obtained vector data, a vector processing is performed on the target image base map to obtain a change map patch, including:

[0041] A semantic segmentation model is used to extract a feature land class from the target image base map, and a business vector range is extracted from the obtained vector data.

[0042] The feature land class and the business vector range are superimposed to obtain a change map patch.

[0043] In a second aspect, the present application provides a remote sensing image matching system for cultivated land monitoring, including:

[0044] A remote sensing image set acquisition module is configured to acquire a satellite remote sensing image set meeting a preset quality requirement.

[0045] An image fall map set manufacturing module is configured to manufacture an image fall map set according to the satellite remote sensing image set.

[0046] a cloud removal module configured to perform cloud detection and removal processing on each image in the image footprint set by using a preset image processing model to obtain a cloud-removed image range footprint set;

[0047] an optimal matching module configured to perform optimal matching on each image range footprint in the image range footprint set according to a preset image matching and image selection algorithm to obtain a target image base map, the target image base map being an optimal image selected from the image range footprints after cloud removal, sorting and fusion;

[0048] a vector processing module configured to perform vector processing on the target image base map according to the obtained vector data to obtain a change plot, the change plot being a plot in which a vector element in the vector data is inconsistent with an image feature land class;

[0049] a monitoring module configured to determine a cultivated land status monitoring plot result according to the change plot.

[0050] In summary, the embodiment of the present application uses the obtained satellite remote sensing image set to generate an image footprint set, performs cloud removal processing on each image footprint in the image footprint set to obtain an image range footprint, then performs optimal matching on the cloud-removed footprints to select an optimal image after cloud removal, sorting and fusion from the image range footprints as a target image base map, performs vector processing on the target image base map according to the obtained vector data to obtain a change plot in which a vector element is inconsistent with an image feature land class, and determines a cultivated land status monitoring plot result according to the change plot. The embodiment of the present application can obtain, analyze and process a large amount of remote sensing image data in real time, select an optimal remote sensing image therefrom, ensure high resolution and timely phase of the monitoring image, realize rapid identification of cultivated land change, meet actual cultivated land monitoring requirements, and solve the problem that the prior art cannot select an optimal remote sensing image meeting actual cultivated land monitoring requirements from a large amount of satellite remote sensing images. BRIEF DESCRIPTION OF DRAWINGS

[0051] The accompanying drawings, which are incorporated herein and form a part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the present application.

[0052] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

[0053] Figure 1 a flowchart of a remote sensing image matching method for cultivated land monitoring provided by the embodiment of the present application;

[0054] Figure 2 is a step flow diagram of a remote sensing image matching method for farmland monitoring provided by an optional embodiment of the present application;

[0055] Figure 3 is a technical line diagram of a remote sensing image matching method for farmland monitoring provided by an optional example of the present application;

[0056] Figure 4 is a schematic diagram of an optimal remote sensing image selection system provided by an optional example of the present application;

[0057] Figure 5 is an image classification label diagram provided by an optional embodiment of the present application;

[0058] Figure 6 is a schematic diagram of image drop map fusion processing provided by an optional embodiment of the present application;

[0059] Figure 7 is a use image sequential list diagram provided by an optional embodiment of the present application;

[0060] Figure 8 is a structural block diagram of a remote sensing image matching system for farmland monitoring provided by an embodiment of the present application;

[0061] Figure 9 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in a clear and complete manner with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0063] To facilitate the understanding of the embodiments of the present application, further explanation and description will be made in combination with the accompanying drawings and specific embodiments, and the embodiments do not constitute a limitation on the embodiments of the present application.

[0064] Figure 1 is a flow diagram of a remote sensing image matching method for farmland monitoring provided by an embodiment of the present application. As shown in Figure 1 the remote sensing image matching method for farmland monitoring provided by the embodiment of the present application can specifically include the following steps:

[0065] Step 110, acquiring a satellite remote sensing image set meeting a preset quality requirement, and producing an image drop map set according to the satellite remote sensing image set.

[0066] In the embodiment, the satellite remote sensing image set includes at least two satellite remote sensing images, which can include but are not limited to commercial satellite remote sensing images (such as Beijing No. 2, Jilin No. 1) and public satellite remote sensing images (such as GF-1, ZY-1) and the like. Specifically, the satellite remote sensing image meeting the preset quality requirement can be selected in the embodiment, and preferably, the resolution of the satellite remote sensing image can cover 0.5 meters to 2 meters, so as to meet the requirements of large data coverage and high resolution at the same time.

[0067] In the specific implementation, the satellite remote sensing image meeting the preset quality requirement can be acquired as the current image data, and then whether the color and texture of the current image data are abnormal is checked to make an image drop map meeting the actual image range, so as to obtain an image drop map set.

[0068] In step 120, cloud detection and removal processing are performed on each image in the image drop map set by using a preset image processing model, so as to obtain a cloud-removed image range drop map set.

[0069] In the embodiment, the image range drop map set can include one or more cloud-removed image range drop maps. The deep learning model pre-trained in the embodiment can be used for image cloud detection and drop map cloud removal, so as to obtain a cloud-removed image range drop map.

[0070] In the specific implementation, the deep learning model can be a U-Net model improved by a deep residual network. The model can be used to extract features from the input image drop map, and then the cloud area and non-cloud area are distinguished according to the extracted features, so as to realize cloud detection. Then, the cloud area is removed according to the detected cloud area, so as to obtain a cloud-removed image range drop map. Thus, the embodiment realizes the problem of cloud shadow in the remote sensing image, and solves the limitation problem caused by the cloud coverage of the image available range in the southern region and other regions under the condition of cloudy and rainy weather.

[0071] In step 130, each image range drop map in the image range drop map set is optimally matched according to a preset image matching and image selection algorithm, so as to obtain a target image base map.

[0072] The target image base map is an optimal image obtained by cloud-sequencing fusion from the image range drop map.

[0073] In the embodiment, the optimal image can also be referred to as the best / optimal remote sensing image; the image matching principle can be understood as the image selection principle, which mainly includes but is not limited to resolution priority and time sequence priority, and the image selection algorithm can be a bubble algorithm, which is not limited in the embodiment.

[0074] In the embodiment, the all falling maps in the cloud cover removal range are sorted in a certain order based on the image selection principle, for example, in ascending order, then the sorted image range falling maps are fused, and then a secondary sorting is performed according to a certain algorithm, so as to obtain the optimal matching image as the target image base map.

[0075] In a specific implementation, the falling maps can be sorted in the resolution priority first, and then a secondary sorting is performed in the time sequence to obtain a falling map set with a certain order, and then a bubble algorithm is used to sort the falling map set, so as to provide an ordered data basis for subsequent data processing and matching process, and the subsequent matching and retrieval operation is optimized through sorting.

[0076] In step 140, the target image base map is vector processed according to the obtained vector data to obtain a change map patch.

[0077] The change map patch is a map patch in which the vector element in the vector data is inconsistent with the image feature land class.

[0078] In step 150, a cultivated land status monitoring map patch result is determined according to the change map patch.

[0079] The steps 140-150 are uniformly described as follows:

[0080] In the embodiment, the image feature land class can include but is not limited to cultivated land class, and the land class of the change map patch can be set according to actual use requirements. For example, when the change monitoring of cultivated land is applied, the cultivated land class can be identified from the target image base map. The vector data is mainly used to determine the business vector range, which refers to the monitoring range such as the cultivated land range.

[0081] In a specific implementation, for the change monitoring of cultivated land, the latest national land change survey result cultivated land range is taken as the monitoring range, the land class in the optimal image is extracted through vector+image interpretation, and then the business vector range is superimposed, so as to find the map patch in which the vector element is inconsistent with the image feature land class, and realize the extraction of the change map patch in which the remote sensing image feature is inconsistent with the cultivated land class in the latest annual land survey database.

[0082] After the change map patch is extracted in the embodiment, the cultivated land status monitoring map patch result can be obtained according to the change map patch, and the dynamic and accurate monitoring of the change of the regional cultivated land range is realized.

[0083] As can be seen, this embodiment utilizes the acquired satellite remote sensing image set to create an image map set. Each image map in the set is declouded to obtain an image range map. Then, optimal matching is performed on the declouded map, and the optimal image selected from the image range map after declouding, sorting, and fusion is used as the target image base map. Vector processing is then performed on the target image base map based on the acquired vector data to obtain change patches where the vector elements do not match the image's land cover features. Based on these change patches, the results of farmland status monitoring are determined. This embodiment can acquire, analyze, and process large amounts of remote sensing image data in real time, selecting the optimal remote sensing image to ensure high resolution and timeliness of the monitoring images. This enables rapid identification of farmland changes, meets actual farmland monitoring requirements, and solves the problem that existing technologies cannot select the optimal remote sensing image from a large number of satellite remote sensing images that meets actual farmland monitoring requirements.

[0084] Reference Figure 2 This illustration shows a flowchart of a remote sensing image matching method for farmland monitoring, provided in an optional embodiment of this application. The method specifically includes the following steps:

[0085] Step 210: Obtain a set of satellite remote sensing images that meet the preset quality requirements, and create an image map set based on the set of satellite remote sensing images.

[0086] In one optional embodiment, the present application embodiment obtains a satellite remote sensing image set that meets preset quality requirements, which may specifically include: obtaining a first remote sensing image set and a second remote sensing image set from an image source according to preset resolution requirements and image quality requirements; performing preprocessing and ensemble correction on the first satellite remote sensing image set and the second satellite remote sensing image set respectively to obtain an orthorectified image set that meets preset quality requirements after eliminating geometric distortion, as the satellite remote sensing image set; wherein, the orthorectified image set includes a first orthorectified image set corresponding to the first satellite remote sensing image set and / or a second orthorectified image set corresponding to the second satellite remote sensing image set.

[0087] In this embodiment, the first remote sensing image set may include commercially procured satellite remote sensing images acquired through commercially procured satellites (such as Beijing-2 and Jilin-1); the second remote sensing image set may include public welfare satellite remote sensing images acquired through public welfare satellites (such as Gaofen-1 and Ziyuan-1).

[0088] In this specific implementation, the first and second remote sensing image sets are processed to select a satellite remote sensing image set. Each satellite remote sensing image in this set can be composed of the optimal images selected from the two sets that meet the requirements. For example, all the optimal images can come from the same remote sensing image set, or they can come from the first and second remote sensing image sets respectively. For instance, they can mainly come from the first remote sensing image set, and the remote sensing images from the first set can be used as a supplement to areas not covered by the images in the first set. This example does not limit the method of selecting the satellite remote sensing image set from the first and second remote sensing image sets.

[0089] In the specific implementation, refer to Figure 3 The remote sensing image matching method for farmland monitoring implemented in this embodiment mainly involves steps such as data collection and quality inspection, cloud detection and removal based on deep learning models, automatic matching of optimal remote sensing images, automatic selection algorithm of optimal remote sensing images, and extraction of farmland change patches based on semantic segmentation models.

[0090] In the relevant technologies, the actual dynamic monitoring of farmland protection in some areas suffers from problems such as low image utilization and inadequate identification of land type information in the monitoring area due to the cloudy and rainy weather in the southern region. In addition, some areas also rely on manual selection of monitoring images and identification of farmland changes on a scene-by-scene basis, which leads to untimely response.

[0091] To improve the efficiency and accuracy of farmland protection, this embodiment constructs an efficient and intelligent dynamic monitoring system for farmland protection, specifically addressing the aforementioned steps, to select the optimal image. Figure 4 As shown, the system mainly includes: a data integration and preprocessing module, an image declouding module, an optimal image automatic matching module, an image automatic selection module, and a change patch extraction module. The data integration and preprocessing module primarily involves the acquisition and preprocessing of commercial satellite remote sensing images, public satellite remote sensing images, monitoring-related vector range data, and thematic management data.

[0092] For data collection and quality inspection, this embodiment can select satellite remote sensing images that meet the resolution and image quality requirements from the acquired satellite remote sensing images. Preferably, the acquired satellite remote sensing images have a resolution coverage of 0.5 meters to 2 meters, thereby simultaneously meeting the needs of monitoring for large amounts of data coverage and high resolution.

[0093] In practical processing, after acquiring satellite remote sensing images, this embodiment can perform preliminary processing on the satellite remote sensing images, including but not limited to: image registration, noise removal, etc. Subsequently, geometric correction is performed on the satellite remote sensing images to produce orthophotos, in order to eliminate geometric distortions in the images and ensure the accuracy of their geographical location and scale.

[0094] In practical implementation, refer to Figure 3 In this embodiment, after acquiring the satellite remote sensing image set, each satellite remote sensing image in the set can be used as the current image. Subsequently, the color and texture of the current image can be checked for anomalies, and an image map conforming to the actual image range can be created. The image quality requirements include, but are not limited to: ① Images should be free of large areas of noise and banding, and distorted images should be avoided as much as possible; ② Images should have natural colors, clear textures, and no blurring or ghosting; ③ Images should have clear details of ground features, moderate contrast, distinct layers, and basically balanced colors, clearly showing surface cover and boundaries.

[0095] In one optional approach, this embodiment creates an image map set based on the satellite remote sensing image set, which may specifically include: performing color and texture detection processing based on the satellite remote sensing image set to create an image map that conforms to the actual image range.

[0096] In practical implementation, image mapping can be understood as vector mapping. In addition to key information such as image data name, acquisition date, and storage path, it can also contain vector information. For example, if a satellite remote sensing image contains the monitoring range of a target area, the corresponding monitoring range can be delineated in the satellite remote sensing image in a vector manner, and relevant information (such as the area name, whether there are cloud areas, etc.) can be marked.

[0097] Step 220: Use the cloud detection model in the preset image processing model to perform cloud detection on the image atlas to obtain a binary cloud mask atlas.

[0098] Step 230: Perform cloud region identification and post-processing based on the cloud mask atlas to obtain a batch cloud detection atlas.

[0099] Steps 220 and 230 are described uniformly as follows:

[0100] In its specific implementation, this embodiment can utilize a U-Net model improved by a deep residual network to construct a cloud detection model, and pre-train the model through a series of model training steps to obtain a trained cloud detection model. For example... Figure 3 As shown, in this embodiment, each image map in the image map set is input into the cloud detection model for cloud detection. The model's residual module continuously downsamples and extracts deep features from the input images to improve the detection accuracy of fragmented and thin clouds. These features can capture multi-scale information and contextual relationships within the image.

[0101] The encoder in the model then extracts features from the input image, with each layer containing a convolutional layer, a batch normalization layer, and a ReLU activation function. During decoding, the features extracted by the encoder are recombined and upsampled to the original image size through upsampling and convolution operations, thereby generating a high-resolution segmentation map. The decoder also employs convolutional layers, batch normalization layers, and the ReLU activation function to ensure effective feature transfer and fusion.

[0102] The decoded feature map is then fused with the input image and convolved again. The model outputs a binary result indicating whether each pixel represents a cloud, resulting in a binary cloud mask map to distinguish between cloud and non-cloud regions. Finally, post-processing such as denoising and boundary smoothing is applied to the output to obtain batch cloud detection maps.

[0103] Step 240: Based on each cloud detection map in the batch cloud detection map set, the cloud region extracted from the corresponding image map is vectorized to obtain the image range map set after removing the cloud range.

[0104] The image range map set includes image range maps.

[0105] In the specific implementation, refer to Figure 3 This implementation vectorizes the cloud regions extracted from each cloud detection map, creates a buffer to address the issues of small object boundary information and fragmented small clouds, sets the width parameter of the buffer (i.e., the outward expansion range), removes the cloud-covered areas from the image range map, and forms an image range map with cloud coverage removed after topological inspection.

[0106] Reference Figure 4 As shown, after acquiring remote sensing images, the image declouding module performs cloud removal processing. This module mainly involves the cloud detection model to perform cloud masking, vectorizing the cloud coverage area, and combining it with the image range mapping data to form effective image range mapping data, thereby reducing the impact of cloud coverage on inaccurate land cover identification in subsequent monitoring.

[0107] In practical implementation, the cloud-detected images can be combined with the image map of the actual image range to vectorize the extracted cloud areas. A buffer is created to solve the problems of small object boundary information and fragmented small clouds. The width parameter of the buffer (i.e. the outward expansion range) and the format and path of the output layer are set. Then, the batch cloud detection images are matched with the corresponding image map range. The cloud detection range is removed from the image map. After topological inspection, an image map with the cloud coverage area removed is formed.

[0108] Step 250: Classify and label each image range map in the image range map set to obtain the label data corresponding to the image range map.

[0109] The labeled data includes a labeled field carrying a value, which includes a resolution value and a temporal value. The temporal value is used to determine the timeliness of the image landing.

[0110] In the specific implementation, each image range map has a corresponding resolution and time phase (preferably, the time phase can be the current quarter being monitored, used to determine the timeliness of the image range map). This embodiment sorts the image range maps primarily based on their resolution and time phase. Therefore, before sorting the image range maps, they can first be classified and labeled according to their resolution. A classification label field is set for different resolutions and assigned values ​​to obtain the resolution assignment. Then, based on this, further classification and labeling is performed according to the time phase. A classification label field is set for different time phases at the same resolution and assigned values ​​to obtain the time phase assignment.

[0111] Optionally, in this embodiment, classifying and labeling each image range map in the image range map set to obtain the label data corresponding to the image range map may include: traversing each image range map, generating a label type based on the resolution and time phase of the image range map; and assigning values ​​based on the label type, combined with the resolution and time phase of the image range map, to obtain the label data corresponding to the image range map.

[0112] For example, refer to Figure 5 The flowchart shown is for the classification and labeling process. The specific process of the optimal image automatic matching module is as follows: ① Classification and Labeling: The input image map data is traversed, and the image resolution and time phase recorded in the image map are divided into 6 label types. Then, the classification label field is assigned values ​​(e.g., "1", "2", "3", "4", "5", "6"). 1 corresponds to an image resolution (imagegsd, the same below) ≤ 0.5 and a time phase (scenetime, the same below) of the current monitoring quarter; 2 corresponds to 0.5 < resolution ≤ 0.75 and a time phase of the current monitoring quarter; 3 corresponds to 0.75 < resolution ≤ 0.8 and a time phase of the current monitoring quarter; 4 corresponds to resolution ≤ 0.5 and a time phase of the previous monitoring quarter; 5 corresponds to 0.5 < resolution ≤ 0.8 and a time phase of the previous monitoring quarter; 6 corresponds to a resolution > 0.8 in the remaining images.

[0113] Step 260: According to the preset image matching rules, each image range map is sorted by resolution and timeliness based on the assigned value of the marker field to obtain a set of effective image range maps with image usage order.

[0114] In the specific implementation, refer to Figure 3 Image matching rules, also known as image selection principles, include two aspects: resolution and timeliness. In this embodiment, based on the resolution and timeliness assignments in each image range map, resolution is prioritized first, followed by timeliness, and a binary matching sort is performed on each image range map. For example, images with a resolution of 0.5m are used first, and the image timeliness is sorted from newest to oldest (taking a monitoring time in June as an example, images from June, May, and April of the current quarter are acquired and sorted by June, May, and April); areas not covered by 0.5m resolution are supplemented with images of a resolution better than 1m, and the image timeliness is sorted from newest to oldest (June, May, and April). Thus, a set of effective image range maps with a usable order is obtained.

[0115] Optionally, this embodiment, according to preset image matching rules, performs binary matching and sorting of each image range map based on resolution and timeliness according to the assigned value of the marker field, to obtain a set of valid image range maps with image usage order. This may include: using the marker data of each image range map as elements to be sorted, performing binary matching pre-sorting based on the marker field using bubble sort, to obtain pre-sorting results, the pre-sorting results including resolution pre-sorting results corresponding to resolution assignment and timeliness pre-sorting results corresponding to timeliness assignment; based on the resolution priority rule, dividing each image range map by resolution according to the resolution pre-sorting results to obtain a resolution-sorted image set with at least one resolution; based on the timeliness priority principle, dividing each image range map in the resolution-sorted image set by timeliness according to the timeliness pre-sorting results to obtain a set of valid image range maps with image usage order; wherein, the pre-sorting results serve as an effective data basis for data processing and matching processes, used to form an ordered list for fast searching.

[0116] In this implementation, the labeled data of each image area is used as the elements to be sorted. Then, a bubble sort algorithm is used to pre-sort the fields of the image area, sorting them according to higher resolution and temporal update, forming an ordered list, and outputting the sorted image area. Specifically, based on the principles of monitoring image usage, a colon sorting method is used to automatically match high-resolution optical remote sensing images that meet the monitoring requirements as the monitoring base map. By using bubble sort to pre-sort all fields of the image area, an ordered data foundation is provided for subsequent data processing and matching processes. Optimizing subsequent matching and retrieval operations through sorting not only improves the efficiency of data processing but also enhances the accuracy of data retrieval and matching, providing strong support for the optimization of the entire data processing flow.

[0117] For example, refer to Figure 4The system and modules shown, particularly the optimal image automatic matching module, primarily involve sorting image plots based on the principles of image usage to generate the most suitable effective images for generating change patches. Specifically, after obtaining the cloud-removed image plots, all plots are first initially classified and labeled, with label fields added and assigned specific values. Then, a bubble sort algorithm is used to pre-sort the plot fields according to higher resolution and temporal update, forming an ordered list, and the sorted image plots are output.

[0118] Furthermore, for bubble sort, this embodiment first considers the classification label, resolution, and temporal field of the image as elements to be sorted. These fields are then sorted according to the following order: higher ordinal number among different classification labels, more recent temporal phase within the same classification label, and higher resolution within the same temporal phase, forming an ordered list. During image matching, the ordered fields can speed up the search, allowing the use of more efficient search algorithms such as binary search instead of linear search. In addition, sorting can help to more easily identify and process duplicate or similar image data, thereby improving the efficiency of data cleaning and deduplication.

[0119] Step 270: Perform mosaic fusion processing on the effective image range map set to obtain the fused remote sensing image map.

[0120] In this specific implementation, the embodiments utilize each image in the effective image range set matched in the previous step as the optimal remote sensing image, and then perform mosaic fusion processing on the optimal remote sensing image. Image image fusion is performed within the monitoring range according to the sorted image data. For example, the image image fusion processing can refer to... Figure 6 As shown above, among which, Figure 6 In each image, Y represents the portion of the aforementioned image range after removing clouds. A, B, C, and D are the sorted image plots. The list of sorted images can be found in [reference needed]. Figure 7 As shown.

[0121] In one optional embodiment, this application embodiment performs mosaic fusion processing based on the effective image range map set to obtain a fused remote sensing image map. Specifically, it may include: extracting the target image range map with cloud range removed, ranked first, from the effective image range map set, the target image range map map containing a target range vector; determining the target cloud removal range based on the target range vector, and selecting a fused image range map set with fused image range vectors from the effective image range map set based on the target cloud removal range; and performing effective range fusion of the target range vector of the target image range map map and the image range vectors of each image range map map in the fused image range map set according to the fusion order, until the target image range map map is a complete layer, thereby obtaining the fused remote sensing image map.

[0122] In related technologies, traditional cloud detection methods for imagery include thresholding, feature analysis, homomorphic filtering, and clustering analysis, but these methods often have certain limitations. In recent years, deep learning-based cloud detection methods have gradually become mainstream. For example, the U-Net method is a U-shaped convolutional neural network cloud detection method with strong feature extraction and reconstruction capabilities. In cloud detection, it can accurately locate and segment cloud regions. By inputting the remote sensing image to be detected into a trained U-Net model, the model outputs a binary result indicating whether each pixel is a cloud. Post-processing of the output results, such as denoising and boundary smoothing, yields the final cloud detection map. However, while existing cloud detection methods can segment cloud regions, the occluded areas are also segmented, leaving unusable gaps that fail to reflect the actual condition of farmland.

[0123] To address the aforementioned issues, this embodiment utilizes image fusion to remove cloud areas while fully preserving occluded areas. Specifically, the effective image fusion data within this range after image-to-image fusion processing can be referenced... Figure 6 As shown. The specific fusion process includes: merging the image range vector ranked first (after removing cloud cover) with the image range vectors of the next order. For example, if the second-ranked image range vector also has no effective range in the area due to cloud removal, then the third-ranked vector is merged in, and so on, until the first-ranked image data in the monitoring area becomes a complete layer, resulting in a fused remote sensing image map. Alternatively, if the first-ranked image range vector has no gaps (no clouds), the process proceeds directly to the next step, using it as the fused remote sensing image map. Thus, through image map fusion, this embodiment obtains the optimal image map that is cloud-free and has no gaps, which is then used as the fused remote sensing image map for subsequent scene-by-scene retrieval and matching.

[0124] In practical implementation, combined with Figure 6 Detailed supplementary description: For the fusion of image range maps, this embodiment mainly uses the image range map ranked first as the basis, and takes it as the target image range map, that is... Figure 6 Before processing, the unprocessed image A, in its represented range vector, can include the portion of the area that has no effective range due to cloud removal, as the target range vector (refer to...). Figure 6 The Y region of image A is missing due to cloud removal. Then, based on the target extent vector, other image extent maps are selected from the set of valid image extent maps to fill the gaps caused by cloud removal. Specifically, for the target extent vector, the image extent vectors of other image extent maps are traversed in sorted order to determine if there are also cases where there is no valid extent due to cloud removal in the region corresponding to the target extent vector. For example, the second-ranked image extent map (…Figure 6 In the unprocessed image map B, if there is no situation where the area corresponding to the target range vector is completely devoid of effective range due to cloud removal, meaning that the image range map has an effective range, then the effective range of this image range map is fused with the target range vector to compensate for the gaps in the target image range map caused by cloud removal. Subsequent image range maps are judged and fused sequentially (refer to...). Figure 6 (Supplementing with map C and map D in the original image) until the missing map data is a complete layer, thus obtaining the fused remote sensing image map.

[0125] The effective scope can be understood as: reference Figure 6 In the case of plot A and plot B, the Y regions of the two plots are not completely overlapping or identical. Therefore, the position of the Y region in plot B corresponding to the Y region in plot A is considered as the non-overlapping part and is understood as the effective range. This is used to fill the cloud removal gaps in the Y region of plot A.

[0126] Step 280: Based on the fused remote sensing image map, use the automatic image retrieval algorithm to perform scene-by-scene retrieval and matching to obtain the optimal image without cloud cover, which will be used as the target image base map.

[0127] In this specific implementation, based on the fused image map data, an automatic image retrieval algorithm is used to match image regions within the effective image range, and matching images are retrieved scene by scene in sequence according to the range. Specifically, the automatic image retrieval algorithm searches for images on the server by image name and storage path, and the retrieved images are used as the final image map selected within the actual monitoring range, outputting an optimal image base map without cloud cover, which is also the target image base map.

[0128] In one optional embodiment, the present application embodiment uses an automatic image retrieval algorithm to perform scene-by-scene retrieval and matching based on the fused remote sensing image map to obtain the optimal image without cloud cover. This may include: using the automatic image retrieval algorithm to search for images on the target server with the image name and storage path corresponding to the fused remote sensing image map to obtain the retrieved image; and using the retrieved image as the optimal image selected within the final actual monitoring range.

[0129] In this implementation, to achieve image retrieval on the target server using an automatic image retrieval algorithm, this embodiment first automatically captures image data from the server and then uses a hash list and string matching (Knuth-Morris-Pratt, KMP) algorithm for efficient image matching. This enables rapid retrieval and accurate matching of image resources, improving the efficiency and accuracy of data processing. The acquired image data includes image names and storage paths, which are stored in a hash list. The selection of the hash list is based on its fast data access and retrieval capabilities, which is crucial for the subsequent matching process.

[0130] In this embodiment, the core matching algorithm KMP is used in the image name matching process. By constructing a partial matching table (prefix function), invalid backtracking of the pattern string is avoided, thereby improving matching efficiency. The partial matching table pre-calculates the length of the longest common element between the prefix and suffix at each position in the pattern string. This allows the algorithm to determine where the pattern string should backtrack to if a mismatch is found during the matching process, reducing the number of comparisons. The time complexity of the KMP algorithm is O(n + m), where n is the text length and m is the pattern string length, significantly improving matching efficiency. In this application scenario, the KMP algorithm matches the list of image names against the server image list, filtering out used images and their server addresses. This precise filtering mechanism not only improves resource management efficiency but also provides a solid foundation for data analysis and processing. Through this refined data management, efficient utilization and precise control of image resources can be achieved, ensuring the efficiency and accuracy of the data processing workflow.

[0131] In practical implementation, refer to Figure 4 The image automatic selection module shown primarily involves matching and selecting images layer by layer according to the image sorting results, based on the effective image range using an automatic image retrieval algorithm, to form an optimal monitoring image without cloud cover. Specifically, in this embodiment, the system obtains an optimal image base map without cloud cover through the automatic image retrieval algorithm, forming a single map for monitoring the current status of cultivated land. Specifically, the system inputs image map data with the image usage order, reads the specified field names (name, scenetime, imagegsd, YXBH) from the image map data attribute table, and the sorted list of image names is as follows. Figure 7 As shown, image mosaicking and fusion are performed in ascending order according to the sorted image number (YXBH) field. Then, an automatic image retrieval algorithm is used to search for images on the server by image name and storage path. The retrieved images are used as the final image map selected within the actual monitoring range, and an optimal image base map without cloud cover is output.

[0132] Furthermore, to enhance the visual effect and quality of the images, this embodiment can perform color homogenization processing on the stitched images to reduce color differences between different images caused by factors such as shooting conditions and sensor characteristics, ensuring uniform color throughout the entire image. In addition, through techniques such as adjusting contrast, reducing noise, sharpening, and atmospheric correction, the image clarity and physical realism are further improved.

[0133] Step 290: Perform vector processing on the target image base map based on the acquired vector data to obtain changed patches. The changed patches are patches in the vector data where the vector elements and image feature land types are inconsistent.

[0134] Optionally, the above-mentioned vector processing of the target image base map based on the acquired vector data to obtain the changed patch may specifically include: extracting feature land classes from the target image base map using a semantic segmentation model, and extracting business vector ranges from the acquired vector data; and superimposing the feature land classes and the business vector ranges to obtain the changed patch.

[0135] In its specific implementation, this embodiment achieves farmland change patch extraction based on a semantic segmentation model. Specifically, referring to... Figure 3 and Figure 4 The change patch extraction module is used to extract change patches within the cultivated land area, providing data support for assessing the cultivated land area in the region and ensuring the balance between cultivated land occupation and replenishment, and between inflow and outflow. This embodiment uses the cultivated land area from the latest land change survey results as the monitoring scope. Utilizing the optimal image obtained from the aforementioned integrated cultivated land status monitoring, this embodiment performs "vector + image" interpretation on the image. Specifically, after extracting the land type from a single period image using a semantic segmentation model, it is overlaid with the operational vector range to identify patches where the vector elements and image feature land types are inconsistent. This achieves the extraction of change patches where remote sensing image features are inconsistent with cultivated land types in the latest annual land survey database.

[0136] Step 300: Determine the results of the farmland status monitoring map based on the changed map patches.

[0137] Therefore, the optimal remote sensing image automatic matching method for monitoring the current status of cultivated land proposed in this application realizes dynamic and accurate monitoring of changes in the scope of cultivated land in a region. This embodiment utilizes a deep learning model for cloud removal and image ranking, which not only solves the problem of missing land type information data in cloud-covered areas due to the susceptibility of optical remote sensing to cloud and rain weather, but also achieves dynamic monitoring of cultivated land protection using the most recent, highest-resolution remote sensing images through the image ranking model. This ensures the accuracy and timeliness of land type monitoring, providing data support and decision support for dynamic supervision of cultivated land protection and ensuring food security. It also addresses the problem that existing technologies cannot select the optimal remote sensing images that meet the actual requirements of cultivated land monitoring from a large number of satellite remote sensing images.

[0138] In summary, this application embodiment creates an image map set based on the acquired satellite remote sensing image set. Then, it uses a cloud detection model in a preset image processing model to perform cloud detection on the image map set, obtaining a binary cloud mask map set. Cloud region identification and post-processing are then performed based on the cloud mask map set to obtain a batch cloud detection map set. Next, each cloud detection map in the batch cloud detection map set is combined with the cloud region extracted from the corresponding image map set for vectorization processing, resulting in an image range map with cloud extent removed. Finally, according to preset image matching rules, each image range map is sorted by resolution and timeliness. The method involves obtaining labeled data corresponding to the time-ordered image set and image extent maps. Based on the labeled data, a pre-defined bubble sort algorithm is used to perform binary matching sorting on the time-ordered image set to obtain an effective image extent map set. Then, mosaicking and fusion processing is performed on the effective image extent map set to obtain a fused remote sensing image map. Based on the fused remote sensing image map, an automatic image retrieval algorithm is used to perform scene-by-scene retrieval and matching to obtain the optimal image without cloud cover. Vector processing is then performed on the vector data to obtain variation patches where vector elements do not match the image feature land types, thereby determining the results of farmland status monitoring patches. The optimal automatic remote sensing image matching method for farmland status monitoring proposed in this embodiment aims to solve the problem that high-quality optical satellite image data is extremely limited under the cloudy and rainy climate conditions in southern regions, and that farmland extent cannot be accurately identified in dynamic monitoring of farmland protection. This embodiment first utilizes a deep learning model for cloud detection and removal to address the cloud shadow problem in remote sensing images. Then, it sorts all image patches containing the cloud-covered areas and performs image patch fusion in ascending order based on image selection principles. The scene number of the fused image is read and the image is automatically loaded, thereby enabling the rapid selection of the optimal remote sensing monitoring image from images with multiple cloud coverage, multiple time phases, and multiple resolutions. This provides technical support for achieving the balance of total cultivated land and the protection of cultivated land in my country.

[0139] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should know that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps may be performed in other orders or simultaneously.

[0140] like Figure 8 As shown in the figure, this application embodiment also provides a remote sensing image matching system 800 for farmland monitoring, including:

[0141] The remote sensing image set acquisition module 810 is used to acquire satellite remote sensing image sets that meet preset quality requirements;

[0142] Image atlas creation module 820 is used to create image atlases based on the satellite remote sensing image atlas;

[0143] The cloud removal module 830 is used to perform cloud detection and removal processing on each image in the image set using a preset image processing model to obtain a cloud-removed image range set.

[0144] The optimal matching module 840 is used to perform optimal matching on each image range map in the image range map set according to a preset image matching and image selection algorithm to obtain a target image base map. The target image base map is the optimal image selected from the image range map after cloud removal, sorting and fusion.

[0145] The vector processing module 850 is used to perform vector processing on the target image base map according to the acquired vector data to obtain changed patches, wherein the changed patches are patches in the vector data whose vector elements are inconsistent with the image feature land types.

[0146] The monitoring module 860 is used to determine the results of the farmland status monitoring based on the changed patches.

[0147] Optionally, the remote sensing image set acquisition module 810 includes:

[0148] The remote sensing image set acquisition submodule is used to acquire the first and second remote sensing image sets from the image source according to preset resolution and image quality requirements.

[0149] The remote sensing image set processing submodule is used to preprocess and perform ensemble correction on the first satellite remote sensing image set and the second satellite remote sensing image set respectively, to obtain an orthorectified image set that meets the preset quality requirements after eliminating geometric distortion, which is used as the satellite remote sensing image set; wherein, the orthorectified image set includes a first orthorectified image set corresponding to the first satellite remote sensing image set and / or a second orthorectified image set corresponding to the second satellite remote sensing image set.

[0150] Optionally, the cloud removal module 830 includes:

[0151] The cloud detection submodule is used to perform cloud detection on the image set using the cloud detection model in the preset image processing model, and obtain a binary cloud mask set.

[0152] The identification and post-processing submodule is used to identify and post-process cloud regions based on the cloud mask atlas to obtain a batch cloud detection atlas.

[0153] The vectorization processing submodule is used to perform vectorization processing on each cloud detection map in the batch cloud detection map set, combined with the cloud region extracted from the corresponding image map, to obtain an image range map set after removing the cloud range. The image range map set includes the image range map.

[0154] Optionally, the optimal matching module 840 includes:

[0155] The classification and labeling submodule is used to classify and label each image range map in the image range map set to obtain the labeling data corresponding to the image range map. The labeling data includes a labeling field carrying an assigned value.

[0156] The sorting submodule is used to sort each image range map by resolution and timeliness according to the preset image matching rules and the value of the marker field, so as to obtain a set of effective image range maps with image usage order.

[0157] The fusion processing submodule is used to perform mosaic fusion processing based on the effective image range map set to obtain the fused remote sensing image map.

[0158] The retrieval and matching submodule is used to perform scene-by-scene retrieval and matching based on the fused remote sensing image map and the automatic image retrieval algorithm to obtain the optimal image without cloud cover, which is used as the target image base map; wherein, the assignment in the tag field includes resolution assignment and temporal assignment, and the temporal assignment is used to determine the timeliness of the image map.

[0159] Optionally, the classification tagging submodule includes:

[0160] The marker type generation unit is used to traverse each image range map and generate a marker type based on the resolution and time of the image range map.

[0161] The assignment unit is used to perform assignment processing based on the marker type, combined with the resolution and time phase of the image range map, to obtain the marker data corresponding to the image range map.

[0162] Optional, sorting submodule, including:

[0163] The pre-sorting unit is used to take the marked data of each image range as the elements to be sorted, and use the bubble sort method to perform binary matching pre-sorting based on the marked field to obtain the pre-sorting result. The pre-sorting result includes the resolution pre-sorting result corresponding to the resolution assignment and the time pre-sorting result corresponding to the time phase assignment.

[0164] The resolution partitioning unit is used to partition the range of each image based on the resolution priority rule and the resolution pre-sorting result, so as to obtain a resolution-sorted image set with at least one resolution.

[0165] The timeliness division unit is used to divide the image range of each image in the resolution sorted image set according to the timeliness priority principle and the timeliness pre-sorting result, so as to obtain an effective image range map set with the image usage order; wherein, the pre-sorting result serves as an effective data basis for data processing and matching processes, and is used to form an ordered list for fast search.

[0166] Optionally, the fusion processing submodule includes:

[0167] The fused image range map set acquisition unit is used to determine the target decloud range based on the target range vector, and select the fused image range map set of the fused image range vector from the effective image range map set based on the target decloud range;

[0168] The fusion unit is used to effectively fuse the target range vector of the target image range map and the image range vector of each image range map in the fused image range map set according to the fusion order, until the target image range map is a complete layer, thus obtaining the fused remote sensing image map.

[0169] Optionally, the retrieval matching submodule includes:

[0170] The search unit is used to search for images on the target server using an automatic image retrieval algorithm, based on the image name and storage path corresponding to the fused remote sensing image plot, and to obtain the retrieved images.

[0171] The optimal image selection unit is used to select the retrieved image as the optimal image within the final actual monitoring range.

[0172] Optionally, the vector processing module includes:

[0173] The land use extraction submodule is used to extract feature land use types from the target image base map using a semantic segmentation model, and to extract business vector ranges from the acquired vector data;

[0174] The overlay submodule is used to overlay the feature land type and the business vector range to obtain the change patch.

[0175] It should be noted that the remote sensing image matching system for farmland monitoring provided in the embodiments of this application can execute the remote sensing image matching method for farmland monitoring provided in any embodiment of this application, and has the corresponding functions and beneficial effects of the method.

[0176] In practical implementation, the aforementioned remote sensing image matching system for farmland monitoring can be integrated into the device, enabling the device to acquire, analyze, and process large amounts of remote sensing image data in real time. Through cloud regional processing and other methods, the optimal / best remote sensing image is selected for comprehensive monitoring of the current farmland. This device can be composed of two or more physical entities, or it can be composed of a single physical entity. For example, the electronic device can be a personal computer (PC), a computer, a server, etc. This application embodiment does not impose specific limitations in this regard.

[0177] like Figure 9 As shown, this application embodiment provides an electronic device, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114. The processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114. The memory 113 is used to store computer programs. When the processor 111 executes the program stored in the memory 113, it implements the steps of the remote sensing image matching method for farmland monitoring provided in any of the aforementioned method embodiments. For example, the steps of a remote sensing image matching method for farmland monitoring may include the following: acquiring a set of satellite remote sensing images that meet preset quality requirements, and creating an image map set based on the satellite remote sensing image set; performing cloud detection and removal processing on each image in the image map set using a preset image processing model to obtain a cloud-removed image range map set; performing optimal matching on each image range map in the image range map set according to a preset image matching and image selection algorithm to obtain a target image base map, wherein the target image base map is the optimal image selected from the image range map after cloud removal, sorting, and fusion; performing vector processing on the target image base map based on the acquired vector data to obtain changed patches, wherein the changed patches are patches in the vector data whose vector elements are inconsistent with the image feature land type; and determining the farmland status monitoring patch results based on the changed patches.

[0178] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the remote sensing image matching method for farmland monitoring as provided in any of the foregoing method embodiments.

[0179] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "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 a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0180] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A remote sensing image matching method for farmland monitoring, characterized in that, include: Acquire a set of satellite remote sensing images that meet preset quality requirements, and create an image map set based on the set of satellite remote sensing images; By using a preset image processing model, cloud detection and removal are performed on each image in the image set to obtain a cloud-removed image range set. According to the preset image matching and image selection algorithm, the optimal matching is performed on each image range map in the image range map set to obtain the target image base map. The target image base map is the optimal image selected from the image range map after cloud removal, sorting and fusion. The target image base map is processed by vector data to obtain changed patches, which are patches in the vector data where the vector features are inconsistent with the image feature land types. The results of the farmland status monitoring map were determined based on the changed map patches; The process involves: performing optimal matching on each image range map in the image range map set according to a preset image matching and image selection algorithm to obtain a target image base map. This includes: classifying and labeling each image range map in the image range map set to obtain label data corresponding to the image range map, wherein the label data includes a label field carrying an assigned value; performing binary matching and sorting of each image range map based on resolution and timeliness according to preset image matching rules and the assigned value of the label field to obtain a valid image range map set with image usage order; performing mosaic fusion processing on the valid image range map set to obtain a fused remote sensing image map; and performing scene-by-scene retrieval matching on the fused remote sensing image map using an automatic image retrieval algorithm to obtain the optimal image without cloud cover, which is used as the target image base map. The assigned value in the label field includes a resolution assignment and a temporal assignment, wherein the temporal assignment is used to determine the timeliness of the image map.

2. The method according to claim 1, characterized in that, Acquire a set of satellite remote sensing images that meet preset quality requirements, including: According to the preset resolution and image quality requirements, the first and second remote sensing image sets are acquired from the image source. The first remote sensing image set and the second remote sensing image set are preprocessed and ensemble corrected respectively to obtain an orthophoto set that meets the preset quality requirements after eliminating geometric distortion, which is then used as the satellite remote sensing image set. The orthophoto set includes a first orthophoto set corresponding to the first remote sensing image set and / or a second orthophoto set corresponding to the second remote sensing image set.

3. The method according to claim 1, characterized in that, By using a preset image processing model, cloud detection and removal are performed on each image in the image set to obtain a cloud-removed image range set, including: The cloud detection model in the preset image processing model is used to perform cloud detection on the image set to obtain a binary cloud mask set; Cloud region identification and post-processing are performed based on the cloud mask atlas to obtain a batch cloud detection atlas; Based on each cloud detection map in the batch cloud detection map set, the cloud region extracted from the corresponding image map is vectorized to obtain an image range map set after cloud range removal, and the image range map set includes image range maps.

4. The method according to claim 1, characterized in that, Each image range map in the image range map set is classified and labeled to obtain the label data corresponding to the image range map, including: Iterate through each image range map and generate a marker type based on the resolution and time of the image range map; Based on the marker type, the marker data corresponding to the image range map is obtained by assigning values ​​according to the resolution and time phase of the image range map.

5. The method according to claim 1, characterized in that, According to preset image matching rules, each image range map is sorted by resolution and timeliness based on the assigned value of the marker field, resulting in a set of valid image range maps with image usage order, including: The marked data of each image range is used as the elements to be sorted. The bubble sort method is used to perform binary matching pre-sorting based on the marked fields to obtain the pre-sorting results. The pre-sorting results include the resolution pre-sorting results corresponding to the resolution assignment and the time pre-sorting results corresponding to the time phase assignment. Based on the resolution priority rule, the resolution of each image range is divided according to the resolution pre-sorting result to obtain a resolution sorted image set with at least one resolution. Based on the principle of timeliness priority, the timeliness of each image range map in the resolution sorted image set is divided according to the timeliness pre-sorting result to obtain an effective image range map set with the image usage order. The pre-sorted results serve as an effective data foundation for data processing and matching, and are used to form an ordered list for fast searching.

6. The method according to claim 1, characterized in that, Mosaic fusion processing is performed on the effective image range map set to obtain the fused remote sensing image map, including: Extract the target image range map, sorted first after removing the cloud range, from the effective image range map set; the target image range map contains the target range vector. The target cloud removal range is determined based on the target range vector, and the fused image range map set is selected from the effective image range map set based on the target cloud removal range. Following the fusion sequence, the target range vector of the target image range map and the image range vector of each image range map in the fused image range map set are effectively fused until the target image range map is a complete layer, thus obtaining the fused remote sensing image map.

7. The method according to claim 1, characterized in that, Based on the fused remote sensing imagery, an automatic image retrieval algorithm is used to perform scene-by-scene matching to obtain the optimal cloud-free imagery, including: Using an automatic image retrieval algorithm, images are searched on the target server using the image name and storage path corresponding to the fused remote sensing image plot, and the retrieved images are obtained. The retrieved images will be selected as the optimal images within the final actual monitoring range.

8. The method according to claim 1, characterized in that, The target image base map is processed using the acquired vector data to obtain variable patches, including: The semantic segmentation model is used to extract feature land classes from the target image base map, and the business vector range is extracted from the acquired vector data; The change patch is obtained by overlaying the feature land type and the business vector range.

9. A remote sensing image matching system for farmland monitoring, characterized in that, include: The remote sensing image set acquisition module is used to acquire satellite remote sensing image sets that meet preset quality requirements; The image atlas creation module is used to create image atlases based on the satellite remote sensing image set; The cloud removal module is used to perform cloud detection and removal processing on each image in the image set using a preset image processing model, so as to obtain a cloud-removed image range set. The optimal matching module is used to perform optimal matching on each image range map in the image range map set according to a preset image matching and image selection algorithm to obtain a target image base map. The target image base map is the optimal image selected from the image range map after cloud removal, sorting and fusion. The vector processing module is used to perform vector processing on the target image base map based on the acquired vector data to obtain changed patches, wherein the changed patches are patches in the vector data whose vector features are inconsistent with the image feature land types; The monitoring module is used to determine the monitoring results of cultivated land status based on the changed patches; The optimal matching module includes: a classification and labeling submodule, used to classify and label each image range map in the image range map set to obtain label data corresponding to the image range map, wherein the label data includes a label field carrying an assigned value; a sorting submodule, used to perform binary matching and sorting of each image range map according to a preset image matching rule and the assigned value of the label field to obtain an effective image range map set with image usage order; a fusion processing submodule, used to perform mosaic fusion processing on the effective image range map set to obtain a fused remote sensing image map; and a retrieval and matching submodule, used to perform scene-by-scene retrieval and matching on the fused remote sensing image map using an automatic image retrieval algorithm to obtain the optimal image without cloud cover as the target image base map; wherein the assigned value in the label field includes a resolution assignment and a temporal assignment, wherein the temporal assignment is used to determine the timeliness of the image map.

Citation Information

Patent Citations

  • Remote sensing image cloud detection method based on feature multi-scale perception and adaptive aggregation

    CN114494821A

  • Intelligent change detection method for buildings in multi-cloud and multi-fog cultivated land protection area

    CN115272848A