Crop plot identification method and system based on time sequence vegetation characteristics

Abstract

The invention relates to a crop plot identification method and system based on time sequence vegetation characteristics, and the method comprises the steps: constructing a normalized difference vegetation index time sequence according to a multi-temporal satellite remote sensing image, and generating a crop probability distribution diagram through a crop classification model; performing connected domain analysis on the crop probability distribution map, extracting the center of gravity of an effective connected domain as a forward attention point, scanning the crop probability distribution map by using a sliding window, and generating a reverse attention point; based on the forward attention point and the reverse attention point, segmenting the high-resolution remote sensing base map through a visual segmentation model, generating a candidate mask set, and screening to reserve a mask with the maximum area as a land parcel segmentation mask of the forward attention point; and combining all the plot segmentation masks to generate a crop plot identification graph, and converting the boundary of each plot segmentation mask into a geodetic coordinate sequence to obtain plot vector boundary data. The method improves the recognition accuracy, and achieves the automatic and precise segmentation of the boundary of the land parcel.

Description

Technical Field

[0001] This disclosure relates to the fields of remote sensing image processing and computer vision technology, and in particular to a method and system for identifying crop plots based on temporal vegetation characteristics. Background Technology

[0002] Accurate and efficient identification and delineation of crop plot boundaries is a core task in agricultural remote sensing, crucial for crop area statistics, growth monitoring, and yield forecasting. Currently, traditional crop identification methods largely rely on single-temporal remote sensing images, classifying and segmenting crops based on spectral features or simple vegetation indices. However, these methods have significant limitations in practical applications: firstly, single-temporal images are easily affected by meteorological factors such as clouds and rain, leading to missing effective data and insufficient stability; secondly, single-temporal information alone cannot fully capture the unique temporal growth patterns of different crops, resulting in limited classification accuracy, especially in areas with complex crop planting structures.

[0003] To overcome the limitations of single-temporal data, existing technologies have begun to utilize multi-temporal remote sensing data, analyzing temporal features such as the Normalized Difference Vegetation Index (NDVEI) to improve recognition accuracy. However, even with more accurate crop probability distribution maps, automatically and accurately converting them into plot units with well-defined vector boundaries remains a technical challenge. Traditional image segmentation methods (such as thresholding and edge detection) are prone to blurring, breakage, or adhesion at plot edges, often requiring extensive manual post-processing, resulting in low efficiency and failing to meet the needs of large-scale, operational applications. Therefore, there is an urgent need for a technical solution that can automate and accurately complete the integrated process from crop identification to plot vector boundary generation. Summary of the Invention

[0004] To address the problem that traditional methods cannot automatically and accurately identify and extract boundaries of specific crops in complex agricultural environments using temporal remote sensing imagery, this disclosure proposes a crop plot identification method based on temporal vegetation features to solve the above problems.

[0005] According to one aspect of this disclosure, a method for identifying crop plots based on temporal vegetation characteristics is provided. The method is executed upon receiving a task instruction and includes the following steps:

[0006] S10. Acquire multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area. Based on the multi-temporal satellite remote sensing images, construct a normalized differential vegetation index time series and generate a crop probability distribution map of the target area through a crop classification model.

[0007] S20. Perform connected component analysis on the crop probability distribution map, extract the centroids of effective connected components as positive attention points, and use a sliding window to scan the crop probability distribution map to generate negative attention points.

[0008] S30. Based on the positive attention points and the negative attention points, the high-resolution remote sensing base map is segmented using a visual segmentation model to generate a candidate mask set for each positive attention point.

[0009] S40. Filter the candidate mask set and retain the mask with the largest area as the plot segmentation mask corresponding to the positive attention point.

[0010] S50. Merge the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extract the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

[0011] Preferably, based on the multi-temporal satellite remote sensing images, a normalized difference vegetation index time series is constructed, including:

[0012] The multi-temporal satellite remote sensing images are processed scene by scene, and the normalized difference vegetation index of each scene of satellite remote sensing image is calculated.

[0013] All the calculated normalized difference vegetation indices are integrated in chronological order to construct a normalized difference vegetation index time series.

[0014] Preferably, generating a crop probability distribution map of the target area using a crop classification model includes:

[0015] The normalized difference vegetation index time series is processed by a trained crop classification model to generate a crop probability distribution map of the target area.

[0016] The crop classification model includes one of the following: Long Short-Term Memory Network model, Random Forest model, or Support Vector Machine model.

[0017] Preferably, the connected component analysis includes:

[0018] Connectivity analysis is performed on the probability distribution map of the crops. Connectivity regions with an area smaller than a preset threshold are removed, and the centroids of the remaining connected regions are extracted as positive attention points.

[0019] Preferably, the process of using a sliding window to scan the crop probability distribution map and generating reverse attention points includes:

[0020] Use a sliding window to scan the crop probability distribution map and determine whether the crop probability values ​​of all pixels within the area covered by the current window are all zero.

[0021] If so, the center point of the current window will be used as the reverse attention point.

[0022] Preferably, generating a candidate mask set for each positive attention point includes:

[0023] The high-resolution remote sensing base map is divided into multiple sub-maps of the same size;

[0024] For each subgraph, each of the positive attention points contained therein is sequentially used as a positive cue point, and all the negative attention points in the current subgraph are used as negative cue points;

[0025] The sub-graph is segmented using a visual segmentation model to generate a candidate mask set for each positive attention point.

[0026] Preferably, the screening of the candidate mask set includes:

[0027] Remove all non-closed masks;

[0028] Remove all masks whose area is greater than the preset upper limit threshold;

[0029] Remove all masks with an area smaller than the preset lower threshold.

[0030] According to one aspect of this disclosure, a crop plot identification system based on temporal vegetation characteristics is provided, comprising:

[0031] The crop probability distribution map generation module acquires multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area, constructs a normalized differential vegetation index time series based on the multi-temporal satellite remote sensing images, and generates a crop probability distribution map of the target area through a crop classification model.

[0032] The attention point extraction module performs connected component analysis on the crop probability distribution map and extracts the centroids of effective connected components as positive attention points. It then uses a sliding window to scan the crop probability distribution map and generates reverse attention points.

[0033] The candidate mask set generation module, based on the positive attention point and the negative attention point, segments the high-resolution remote sensing base map using a visual segmentation model to generate a candidate mask set for each positive attention point;

[0034] The candidate mask set filtering module filters the candidate mask set and retains the mask with the largest area as the plot segmentation mask corresponding to the positive attention point;

[0035] The plot vector boundary data generation module merges the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extracts the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

[0036] According to one aspect of this disclosure, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: execute the above-described method for identifying crop plots based on temporal vegetation characteristics.

[0037] According to one aspect of this disclosure, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the above-described method for identifying crop plots based on temporal vegetation characteristics.

[0038] Compared to the prior art, the beneficial effects of this disclosure are as follows:

[0039] 1) This disclosure constructs a crop probability distribution map by using the time series features of the normalized difference vegetation index, which effectively overcomes the recognition limitations of single-period optical images under cloudy and rainy conditions and significantly improves the stability and reliability of crop identification.

[0040] 2) The automatic generation mechanism of “positive and negative attention points” proposed in this disclosure, combined with the visual segmentation model, realizes automatic location and accurate segmentation of land parcels without human intervention, effectively improving the problems of fuzziness and missed detection in the processing of land parcel edges in traditional methods.

[0041] 3) This disclosure, by adding anomaly mask screening and automatic vectorization output process, directly generates land parcel vector boundaries in the geodetic coordinate system while ensuring mask quality, providing efficient and directly usable structured data support for agricultural geographic information systems.

[0042] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0043] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to illustrate the technical solutions of this disclosure.

[0045] Figure 1 A flowchart of a method for identifying crop plots based on temporal vegetation features according to an embodiment of this disclosure is shown;

[0046] Figure 2A flowchart illustrating the identification and shaping of plots for a specific crop in an embodiment of this disclosure is shown;

[0047] Figure 3 A block diagram of a crop plot identification system based on temporal vegetation features according to an embodiment of this disclosure is shown. Detailed Implementation

[0048] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0049] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0050] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0051] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this disclosure, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0053] Based on the above ideas, this disclosure proposes a method for identifying crop plots based on temporal vegetation characteristics. Figure 1 A flowchart illustrating a method for identifying crop plots based on temporal vegetation characteristics is provided. The method includes:

[0054] S10. Acquire multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area. Based on the multi-temporal satellite remote sensing images, construct a normalized differential vegetation index time series and generate a crop probability distribution map of the target area through a crop classification model.

[0055] S20. Perform connected component analysis on the crop probability distribution map, extract the centroids of effective connected components as positive attention points, and use a sliding window to scan the crop probability distribution map to generate negative attention points.

[0056] S30. Based on the positive attention points and the negative attention points, the high-resolution remote sensing base map is segmented using a visual segmentation model to generate a candidate mask set for each positive attention point.

[0057] S40. Filter the candidate mask set and retain the mask with the largest area as the plot segmentation mask corresponding to the positive attention point.

[0058] S50. Merge the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extract the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

[0059] The above embodiments construct a crop probability distribution map by using the time series features of the normalized difference vegetation index, which effectively overcomes the recognition limitations of single-period optical images under cloudy and rainy conditions and significantly improves the stability and reliability of crop recognition. The automatic generation mechanism of "positive and negative attention points" proposed in the above embodiments, combined with the visual segmentation model, realizes automatic location and accurate segmentation of plots without human intervention, effectively improving the problems of fuzziness and missed detection in the edge processing of plots in traditional methods.

[0060] Figure 2This document presents a flowchart illustrating the identification and segmentation process for specific crops. First, a classification confidence map for a specific crop (e.g., rice) is generated based on long-term NDVI remote sensing imagery. Connected regions formed by pixels with a confidence score > 0 are filtered, and clutter regions with areas smaller than a threshold are removed. The centroids of the remaining connected regions are then extracted as positive cue points. Simultaneously, an adjustable-size window scans the confidence map, and the center point of the window where all pixels have a confidence score of 0 is set as a negative cue point. Next, the high-resolution base map is divided into 1024×1024 sub-maps and input into the SAM model. Using positive cue points as segmentation targets and negative cue points to exclude the background, a corresponding mask set is generated. Subsequently, three types of abnormal masks—those that are not closed, have areas exceeding a threshold, or have areas below a threshold—are removed. The largest of the remaining masks is selected as the segmentation mask for that positive cue point. Finally, the masks of all positive cue points and the segmentation results of each sub-image are merged to form a complete segmentation image. Then, the boundary points of the segmentation image are extracted and converted into vector boundaries in the WGS84 coordinate system, completing the entire process from data processing to accurate vector output.

[0061] This disclosure further extends the above-described steps S10-S50 with detailed possible implementations, specifically including the following steps:

[0062] S10. Acquire multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area. Based on the multi-temporal satellite remote sensing images, construct a normalized differential vegetation index time series and generate a crop probability distribution map of the target area through a crop classification model.

[0063] In one embodiment, multi-temporal satellite remote sensing images covering the target area are acquired to form a probability confidence distribution map in the continuous time domain, thereby obtaining stable crop identification feature input even under cloudy and rainy conditions. The multi-temporal satellite remote sensing images should cover at least one complete growing season of crops in the target area to ensure that the NDVI (Normalized Differential Vegetation Index) time series can reflect the phenological characteristics of crops.

[0064] Furthermore, based on the multi-temporal satellite remote sensing images, a normalized differential vegetation index time series is constructed, including: processing the multi-temporal satellite remote sensing images scene by scene, calculating the normalized differential vegetation index for each scene of the satellite remote sensing images; integrating all the calculated normalized differential vegetation indices in chronological order to construct a normalized differential vegetation index time series.

[0065] The multi-temporal satellite remote sensing images are processed scene by scene, including radiometric calibration, atmospheric correction, cloud and shadow detection and removal, image registration and cropping, etc., to ensure that the multi-temporal satellite remote sensing images have consistent spectral and spatial references, and the normalized difference vegetation index of each satellite remote sensing image is calculated.

[0066] All calculated normalized difference vegetation indices are integrated in chronological order to construct the NDVI time series for each pixel within the target area. For missing values ​​caused by factors such as clouds and shadows, temporal interpolation methods (such as linear interpolation, Savitzky-Golay filtering, etc.) can be used to fill in the missing values ​​to form a continuous and smooth NDVI curve.

[0067] Generating a crop probability distribution map of a target area using a crop classification model includes: processing the normalized difference vegetation index time series using a trained crop classification model to generate a crop probability distribution map of the target area.

[0068] The crop classification model includes one of the following: Long Short-Term Memory (LSTM) network model, Random Forest model, or Support Vector Machine (SVM) model. The model is trained using historical sample data (including NDVI time-series labels for both target and non-target crops). After training, the NDVI time-series data of the target region is input into the trained crop classification model, which outputs the probability value of each pixel belonging to a crop (e.g., rice), forming a crop probability distribution map with spatial resolution consistent with the satellite remote sensing image. The value of each pixel in the map represents the confidence level that the pixel is a crop, typically ranging from [0,1].

[0069] S20. Perform connected component analysis on the crop probability distribution map, extract the centroids of effective connected components as positive attention points, and use a sliding window to scan the crop probability distribution map to generate negative attention points.

[0070] In this embodiment, the connected component analysis includes: performing connected component analysis on the crop probability distribution map, removing connected components with an area smaller than a preset threshold, and extracting the centroid of the remaining connected components as positive attention points.

[0071] Connectivity analysis is performed on the probability distribution map of crops to identify potential crop plots. The specific steps are as follows:

[0072] Connected component extraction: Using pixels with a confidence score greater than 0 as foreground pixels, the confidence distribution map is labeled with connected components to identify all interconnected foreground pixel regions. The connectivity rule preferably uses 8-neighborhood connections to ensure that spatially adjacent high-confidence pixels are grouped into the same region.

[0073] Small-area clutter removal: Calculate the number of pixels contained in each connected region as its area. Based on the typical planting scale of crops and the spatial resolution of the image, set an area threshold (e.g., corresponding to an actual area of ​​0.2 hectares or 200 pixels). Connected regions with an area smaller than the area threshold are considered cluttered areas or noise (such as field ridge vegetation, scattered water bodies, shadows, etc.) and are removed.

[0074] Centroid Calculation and Positive Attention Point Generation: For each remaining valid connected component with an area greater than the area threshold, calculate its centroid coordinates (center point). Use the centroid coordinates of each valid connected component as a positive attention point; these points serve as positive prediction points for the subsequent visual segmentation model, indicating to the model that "target crop plots exist near these locations."

[0075] The process of generating a reverse attention point by scanning the crop probability distribution map using a sliding window includes: scanning the crop probability distribution map using a sliding window and determining whether the crop probability values ​​of all pixels within the area covered by the current window are all zero; if so, the center point of the current window is taken as the reverse attention point. The specific steps are as follows:

[0076] Sliding window settings: Define an adjustable rectangular sliding window, whose width and height can be set according to actual needs (e.g., an initial value of 50×50 pixels, and the parameters can be adjusted). The window sliding step size is usually set to 50%~80% of the window size to balance computational efficiency and coverage integrity.

[0077] Zero-confidence region determination: Starting from the top left corner of the image, the sliding window is moved sequentially to each possible position. For each window position, the values ​​of all pixels within its coverage area are checked in the confidence distribution map.

[0078] Reverse attention point extraction: If the confidence value of all pixels within the current window is 0 (i.e., the window area does not belong to the target crop at all), then the area is determined to be a "definite non-plot area". At this time, the coordinates of the center point of the sliding window are recorded as a reverse attention point.

[0079] Reverse attention point set generation: After traversal, all window center points that meet the conditions constitute the reverse attention point set.

[0080] This embodiment extracts positive prediction points by connecting component centroids and generates negative prediction points by scanning blank areas with a sliding window, forming a "guiding signal" for the visual model. The positive and negative attention point mechanism, automatically generated based on the probability distribution map, guides the visual segmentation model in target area localization and background exclusion, thereby achieving automated land parcel segmentation without human intervention.

[0081] S30. Based on the positive attention points and the negative attention points, the high-resolution remote sensing base map is segmented using a visual segmentation model to generate a candidate mask set for each positive attention point.

[0082] In this embodiment, a high-resolution remote sensing base map, forward attention points, and reverse attention points are used as inputs and imported into a pre-trained visual segmentation model (such as the Segment Anything Model, or SAM for short). Using the Segment Anything visual segmentation model, guided by automatically generated forward and reverse cue points, accurate segmentation of complex land parcel boundaries is achieved, avoiding the blurring and missed detection problems in land parcel edge processing of traditional optical or topographic segmentation methods. Specifically, the following steps are included:

[0083] Tile Size Setting: To match the optimal input size for the visual segmentation model, the high-resolution remote sensing base map is divided into a regular grid of 1024×1024 pixels, generating a series of non-overlapping or partially overlapping sub-maps (tiles). If the size of the high-resolution remote sensing base map is not an integer multiple of 1024, it can be handled by zero-padding or overlapping sliding windows.

[0084] Attention point mapping: The positive and negative attention points of the entire domain are mapped to their corresponding subgraphs based on their spatial coordinates. For connected components that cross the boundaries of subgraphs, their centroids are assigned to the subgraph containing that point; negative attention points are also assigned according to their coordinates.

[0085] Generating a candidate mask set for each positive attention point includes: splitting the high-resolution remote sensing base map into multiple sub-maps of the same size; for each sub-map, sequentially using each of the positive attention points contained therein as a positive cue point, and using all the negative attention points in the current sub-map as negative cue points; segmenting the sub-map using a visual segmentation model to generate a candidate mask set for each positive attention point.

[0086] Attention-based subgraph segmentation specifically includes the following steps: For each subgraph containing at least one positive attention point, segmentation and mask optimization are performed sequentially, guided by each positive attention point. For a given positive attention point in the subgraph, it is used as a positive prediction point; all negative attention points within the subgraph are used as negative prediction points. These prediction points are input into the visual segmentation model, which outputs multiple candidate segmentation masks, forming a mask set.

[0087] S40. Filter the candidate mask set and retain the mask with the largest area as the plot segmentation mask corresponding to the positive attention point.

[0088] In this embodiment, to improve the quality of the mask, the candidate mask set is screened, and the following three types of abnormal masks are removed. The visual segmentation model output is adaptively optimized to ensure that each plot mask has spatial continuity and morphological integrity.

[0089] All non-closed masks: Masks with breakpoints at the boundary or that do not form a closed polygon.

[0090] All masks with an area greater than the preset upper limit threshold: Masks with an area exceeding the preset upper limit threshold (e.g., corresponding to an actual area of ​​5 hectares) may be due to over-segmentation or inclusion of non-target features.

[0091] All masks with an area smaller than a preset lower threshold: Masks with an area smaller than a preset lower threshold (e.g., corresponding to an actual area of ​​0.1 hectares) may be noise or broken patches.

[0092] After the above screening, if the mask set is not empty, the mask with the largest area is selected as the final plot segmentation mask corresponding to the positive attention point. This strategy is based on the assumption that "an effective plot should generate the largest continuous area near its center point," which can effectively eliminate small-area masks caused by local missegmentation.

[0093] S50. Merge the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extract the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

[0094] In this embodiment, after the segmentation mask is output, boundary extraction and geographic vectorization are automatically performed to form land parcel boundary data in the WGS84 coordinate system, realizing a fully automated closed loop for land parcel mapping. Specifically, the following steps are included:

[0095] Mask stitching and deduplication: The plot segmentation masks obtained from all sub-maps are stitched together according to their high-resolution remote sensing base map coordinates to merge them into a complete crop plot identification map. For edge duplication or overlapping masks that may occur due to plot segmentation, fusion or deduplication processing is performed to ensure that each plot has a unique mask in the final output.

[0096] Boundary extraction and geocoding: For each plot mask, extract its boundary pixel sequence. Using the geographic transformation parameters (such as affine transformation coefficients) of the high-resolution remote sensing base map, convert the boundary pixel coordinates into geographic coordinates (such as latitude and longitude sequences in the WGS84 coordinate system), generating a vector polygon representation of the plot.

[0097] Output format: The final output is a geographic vector file (such as GeoJSON or Shapefile format), where each polygon feature corresponds to an identified crop plot, along with its geographic boundary geodetic coordinate sequence. This result can be directly imported into agricultural geographic information systems, plot management databases, or used for subsequent applications such as area statistics and spatial analysis.

[0098] This disclosure achieves stability and reliability of crop identification input under cloudy and rainy conditions by constructing multi-temporal NDVI time-series features to generate crop probability distribution maps, overcoming the limitations of traditional single-period optical remote sensing images that are easily affected by weather factors. Based on this, it innovatively proposes a mechanism for automatically generating positive and negative attention points based on probability distribution maps, which can guide the visual segmentation model to automatically locate the target area and exclude the background without manual intervention. Furthermore, it combines the Segment Anything visual segmentation model to achieve accurate segmentation of complex plot boundaries, effectively avoiding the blurring and missed detection problems in plot edge processing of traditional methods. By introducing a mask quality control mechanism to adaptively optimize the segmentation results, it ensures the spatial continuity and morphological integrity of each plot mask. Finally, it achieves fully automated closed-loop output from remote sensing images to plot vector maps. The resulting WGS84 coordinate system boundary data can be directly used in agricultural geographic information systems and plot databases, providing a complete, stable, and efficient technical solution for crop plot identification and mapping.

[0099] As another aspect of this disclosure, a crop plot identification system 100 based on temporal vegetation characteristics is also provided, such as... Figure 3 As shown, it includes:

[0100] The crop probability distribution map generation module 1 acquires multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area, constructs a normalized differential vegetation index time series based on the multi-temporal satellite remote sensing images, and generates a crop probability distribution map of the target area through a crop classification model.

[0101] Attention point extraction module 2 performs connected component analysis on the crop probability distribution map and extracts the centroids of effective connected components as positive attention points. It then uses a sliding window to scan the crop probability distribution map and generates reverse attention points.

[0102] The candidate mask set generation module 3, based on the positive attention point and the negative attention point, segments the high-resolution remote sensing base map using a visual segmentation model to generate a candidate mask set for each positive attention point;

[0103] Candidate mask set filtering module 4 filters the candidate mask set and retains the mask with the largest area as the plot segmentation mask corresponding to the positive attention point;

[0104] The plot vector boundary data generation module 5 merges the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extracts the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

[0105] Without causing contradictions, the above-described modules in the system of the present disclosure embodiments can implement any of the above-described methods.

[0106] Based on the description of the above embodiments, it can be seen that the embodiments of this disclosure can achieve the following technical effects:

[0107] 1) The embodiments of this disclosure construct a crop probability distribution map by using the normalized difference vegetation index time series features, which effectively overcomes the recognition limitations of single-period optical images under cloudy and rainy conditions and significantly improves the stability and reliability of crop recognition.

[0108] 2) The automatic generation mechanism of “positive and negative attention points” proposed in this embodiment, combined with the visual segmentation model, realizes automatic location and accurate segmentation of land parcels without human intervention, effectively improving the problems of fuzziness and missed detection in the edge processing of land parcels in traditional methods.

[0109] 3) By adding anomaly mask screening and automatic vectorization output process, this embodiment of the disclosure can directly generate the land parcel vector boundary in the geodetic coordinate system while ensuring mask quality, thus providing efficient and directly usable structured data support for agricultural geographic information systems.

[0110] This disclosure also proposes an electronic device, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured for the above-described method for identifying crop plots based on temporal vegetation characteristics. The electronic device may be provided as a terminal, a server, or other form of device.

[0111] This disclosure also proposes a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the aforementioned method for identifying crop plots based on temporal vegetation characteristics. The computer-readable storage medium may be a non-volatile computer-readable storage medium.

[0112] Those skilled in the art will understand that, in the above-described method and system for identifying crop plots based on temporal vegetation characteristics in specific embodiments, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0113] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0114] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for identifying crop plots based on temporal vegetation characteristics, characterized in that, Includes the following steps: S10. Acquire multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area. Based on the multi-temporal satellite remote sensing images, construct a normalized differential vegetation index time series and generate a crop probability distribution map of the target area through a crop classification model. S20. Perform connected component analysis on the crop probability distribution map, extract the centroids of effective connected components as positive attention points, and use a sliding window to scan the crop probability distribution map to generate negative attention points. S30. Based on the positive attention points and the negative attention points, the high-resolution remote sensing base map is segmented using a visual segmentation model to generate a candidate mask set for each positive attention point. S40. Filter the candidate mask set and retain the mask with the largest area as the plot segmentation mask corresponding to the positive attention point. S50. Merge the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extract the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

2. The method according to claim 1, characterized in that, Based on the aforementioned multi-temporal satellite remote sensing images, a normalized difference vegetation index time series is constructed, including: The multi-temporal satellite remote sensing images are processed scene by scene, and the normalized difference vegetation index of each scene of satellite remote sensing image is calculated. All the calculated normalized difference vegetation indices are integrated in chronological order to construct a normalized difference vegetation index time series.

3. The method according to claim 2, characterized in that, Generate a probability distribution map of crops in the target area using a crop classification model, including: The normalized difference vegetation index time series is processed by a trained crop classification model to generate a crop probability distribution map of the target area. The crop classification model includes one of the following: Long Short-Term Memory Network model, Random Forest model, or Support Vector Machine model.

4. The method according to claim 1, characterized in that, The connected component analysis includes: Connectivity analysis is performed on the probability distribution map of the crops. Connectivity regions with an area smaller than a preset threshold are removed, and the centroids of the remaining connected regions are extracted as positive attention points.

5. The method according to claim 1, characterized in that, Using a sliding window to scan the crop probability distribution map, reverse attention points are generated, including: Use a sliding window to scan the crop probability distribution map and determine whether the crop probability values ​​of all pixels within the area covered by the current window are all zero. If so, the center point of the current window will be used as the reverse attention point.

6. The method according to claim 1, characterized in that, Generate a candidate mask set for each positive attention point, including: The high-resolution remote sensing base map is divided into multiple sub-maps of the same size; For each subgraph, each of the positive attention points contained therein is sequentially used as a positive cue point, and all the negative attention points in the current subgraph are used as negative cue points; The sub-graph is segmented using a visual segmentation model to generate a candidate mask set for each positive attention point.

7. The method according to claim 1, characterized in that, Filtering the candidate mask set includes: Remove all non-closed masks; Remove all masks whose area is greater than the preset upper limit threshold; Remove all masks with an area smaller than the preset lower threshold.

8. A crop plot identification system based on temporal vegetation characteristics, characterized in that, include: The crop probability distribution map generation module acquires multi-temporal satellite remote sensing images and high-resolution remote sensing base maps of the target area, constructs a normalized differential vegetation index time series based on the multi-temporal satellite remote sensing images, and generates a crop probability distribution map of the target area through a crop classification model. The attention point extraction module performs connected component analysis on the crop probability distribution map and extracts the centroids of effective connected components as positive attention points. It then uses a sliding window to scan the crop probability distribution map and generates reverse attention points. The candidate mask set generation module, based on the positive attention point and the negative attention point, segments the high-resolution remote sensing base map using a visual segmentation model to generate a candidate mask set for each positive attention point; The candidate mask set filtering module filters the candidate mask set and retains the mask with the largest area as the plot segmentation mask corresponding to the positive attention point; The plot vector boundary data generation module merges the plot segmentation masks corresponding to all positive attention points to generate a crop plot identification map, and extracts the boundaries of each plot segmentation mask to convert them into geodetic coordinate sequences to obtain plot vector boundary data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the crop plot identification method based on temporal vegetation characteristics as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the crop plot identification method based on temporal vegetation characteristics as described in any one of claims 1 to 7.

Images