Crop classification method and device based on landscape complexity remote sensing model

Abstract

This invention provides a method and apparatus for crop classification based on a landscape complexity remote sensing model, relating to the field of agricultural remote sensing technology. The method obtains the optimal feature set for remote sensing of farmland landscape complexity based on satellite remote sensing image data. Through the landscape complexity remote sensing model, multiple sensing zones of the farmland landscape are obtained. Then, based on these multiple sensing zones, crops in the sensing zones are classified. This invention, by establishing a landscape complexity remote sensing model, achieves complex zoning of farmland landscapes without relying on prior knowledge, giving the model good transferability. Furthermore, based on the complex zoning of the farmland landscape, crops are further identified and classified, thereby obtaining a crop distribution map and improving the accuracy of crop classification.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing technology, and in particular to a method and apparatus for crop classification based on a landscape complexity remote sensing model. Background Technology

[0002] Remote sensing technology possesses objective, rapid, and large-scale monitoring capabilities, enabling the acquisition of accurate and abundant agricultural information in a short period. Utilizing remote sensing technology to identify crops and accurately map their distribution is of significant scientific and practical importance in today's society, characterized by climate change and population growth.

[0003] Current research on farmland landscapes uses remote sensing data with a coarse scale to characterize the complexity of farmland landscapes, resulting in a general understanding of this complexity and a lack of detailed characterization. Furthermore, existing studies often rely on prior knowledge such as historical crop distribution maps to summarize the relationship between farmland landscape complexity and classification accuracy, leading to poor model transferability and low accuracy in crop classification. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for crop classification based on a landscape complexity remote sensing model, in order to solve the above-mentioned problems existing in the prior art.

[0005] This invention provides a crop classification method based on a landscape complexity remote sensing model, comprising the following steps.

[0006] Acquire satellite remote sensing image data of farmland landscapes;

[0007] Based on the satellite remote sensing image data of the farmland landscape, the optimal feature set for remote sensing perception of the complexity of the farmland landscape is obtained;

[0008] The optimal feature set of the farmland landscape complexity remote sensing is input into the landscape complexity remote sensing model to obtain the farmland landscape sensing raster output by the landscape complexity remote sensing model.

[0009] Based on the farmland landscape perception grid, multiple perception zones of the farmland landscape are obtained;

[0010] Based on multiple sensing zones of the farmland landscape, the crops in the multiple sensing zones are classified to obtain a spatial distribution map of the crops.

[0011] According to the present invention, a crop classification method based on a landscape complexity remote sensing model is provided. The steps for establishing the landscape complexity remote sensing model are as follows:

[0012] Acquire drone remote sensing image data of farmland landscape;

[0013] Landscape features are extracted from the UAV remote sensing image data. These landscape features include band reflectance features, vegetation index features, and texture features, along with the standard deviation and mean of the landscape unit grid within the farmland landscape.

[0014] Based on the UAV remote sensing image data and the landscape features, construct one or more decision trees;

[0015] Based on the one or more decision trees, a remote sensing model for landscape complexity is established through ensemble learning.

[0016] According to the present invention, a crop classification method based on a landscape complexity remote sensing model is provided, wherein the optimal feature set for remote sensing perception of farmland landscape complexity is obtained based on satellite remote sensing image data of the farmland landscape, including:

[0017] Based on the satellite remote sensing image data of the farmland landscape, the landscape features of the farmland landscape are constructed;

[0018] One or more landscape features of the farmland landscape are input into the landscape complexity remote sensing model to obtain different sensing accuracy results corresponding to one or more landscape features of the farmland landscape output by the landscape complexity remote sensing model.

[0019] By selecting one or more landscape features corresponding to the highest perception accuracy results, the optimal feature set for remote sensing perception of farmland landscape complexity is obtained.

[0020] The elements in the optimal feature set include multiple optimal feature subsets; the elements in the optimal feature subset include one or more landscape features.

[0021] According to the present invention, a crop classification method based on a landscape complexity remote sensing model includes classifying crops in multiple sensing zones of the farmland landscape to obtain a spatial distribution map of the crops, comprising:

[0022] Based on the multiple sensing zones of the farmland landscape, the optimal feature set within the multiple sensing zones of the farmland landscape is selected to obtain the optimal feature set corresponding to each sensing zone.

[0023] The optimal feature set corresponding to each perceptual partition is input into multiple classifiers to obtain the classification result output by each classifier.

[0024] By comparing the classification results output by the multiple classifiers, the optimal classification result corresponding to each perceptual partition is obtained;

[0025] By stitching together the optimal classification results corresponding to each sensing partition, a spatial distribution map of the crop is obtained.

[0026] According to the present invention, a crop classification method based on a landscape complexity remote sensing model includes optimizing the optimal feature set within multiple sensing partitions of the farmland landscape to obtain the optimal feature set corresponding to each sensing partition. This includes:

[0027] Based on multiple perceptual partitions of the farmland landscape, the optimal feature set within the multiple perceptual partitions of the farmland landscape is obtained;

[0028] Based on the optimal feature set within multiple perceptual zones of the farmland landscape, the importance of each feature in the optimal feature set is evaluated by repeatedly training one or more models; wherein, the higher the importance, the greater the contribution of the feature to the model performance, and vice versa.

[0029] Based on the degree of importance, features that contribute little to the model performance in the best feature set of multiple perception zones of the farmland landscape are gradually removed until the number of remaining features in the best feature set reaches a set threshold.

[0030] The best feature set within multiple sensing zones of the farmland landscape, where the number of features reaches a set threshold, is taken as the optimal feature set for each sensing zone.

[0031] According to the present invention, a crop classification method based on a landscape complexity remote sensing model is provided, wherein the classifier includes one or more of decision tree classifier, random forest classifier, support vector machine and minimum distance classifier.

[0032] The present invention also provides a crop classification device based on a landscape complexity remote sensing model, comprising the following modules:

[0033] The acquisition module is used to acquire satellite remote sensing image data of farmland landscapes;

[0034] The feature module is used to obtain the optimal feature set for remote sensing perception of the complexity of the farmland landscape based on the satellite remote sensing image data of the farmland landscape.

[0035] The perception module is used to input the optimal feature set of the farmland landscape complexity remote sensing perception into the landscape complexity remote sensing perception model to obtain the farmland landscape perception raster output by the landscape complexity remote sensing perception model.

[0036] The partitioning module is used to obtain multiple perceptual partitions of the farmland landscape based on the farmland landscape perception grid.

[0037] The classification module is used to classify crops in multiple sensing zones of the farmland landscape to obtain a spatial distribution map of the crops.

[0038] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the crop classification method based on the landscape complexity remote sensing model as described above.

[0039] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the crop classification method based on a landscape complexity remote sensing model as described above.

[0040] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the crop classification method based on a landscape complexity remote sensing model as described above.

[0041] The present invention provides a crop classification method and apparatus based on a landscape complexity remote sensing model. This method acquires satellite remote sensing image data of farmland landscapes and obtains the optimal feature set for remote sensing of farmland landscape complexity based on this data. Then, using the landscape complexity remote sensing model, it partitions the farmland landscape into multiple sensing zones without relying on prior knowledge. Finally, based on these multiple sensing zones, it classifies the crops within each zone. This not only enhances the transferability of the landscape complexity remote sensing model but also improves the accuracy of crop classification. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 This is one of the flowcharts of the crop classification method based on a landscape complexity remote sensing model provided by the present invention.

[0044] Figure 2 This is the second schematic diagram of the crop classification method based on the landscape complexity remote sensing model provided by the present invention.

[0045] Figure 3 This is a schematic diagram of the two-dimensional complexity concept model provided by the present invention.

[0046] Figure 4 This is a flowchart illustrating the feature optimization process provided by the present invention.

[0047] Figure 5 This is a diagram showing the perception accuracy and the required number of features achieved by the optimal feature subset under different spatial resolutions, provided by this invention.

[0048] Figure 6 This is a schematic diagram of the optimal feature subset under different spatial resolutions provided by the present invention.

[0049] Figure 7 This is a landscape zoning result map of the experimental area provided by the present invention.

[0050] Figure 8 This is an evaluation chart of the classifier effect for different landscape zones provided by the present invention.

[0051] Figure 9 This invention provides a crop distribution map of an experimental area based on farmland landscape zoning.

[0052] Figure 10 This is a comparison chart of crop classification accuracy provided by the present invention.

[0053] Figure 11 This is a schematic diagram of the crop classification device based on a landscape complexity remote sensing model provided by the present invention.

[0054] Figure 12 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0055] Accurate and timely understanding of large-scale crop planting conditions is not only beneficial for the deployment of national grain subsidy policies, macro-control of market prices, and the formulation of foreign grain trade policies, but also of great significance for adjusting agricultural production layout and improving the precision of agricultural insurance services. Obtaining reliable crop distribution maps also provides fundamental data for downstream applications of agricultural remote sensing. Only by acquiring information on crop distribution can further work be carried out, such as crop area estimation, crop physiological and ecological parameter inversion, crop yield estimation, and crop disaster monitoring. These tasks are crucial for a deeper understanding of crop productivity, disaster loss estimation, and arable land quality assessment.

[0056] Remote sensing technology possesses objective, rapid, and large-scale monitoring capabilities, enabling the acquisition of accurate and abundant agricultural information in a short period. Utilizing remote sensing technology for crop identification and accurate crop distribution mapping has significant scientific and practical implications in today's society, characterized by climate change and population growth. Optimizing the identification features of crops in remote sensing imagery and selecting effective classification algorithms are key tasks for improving crop identification accuracy and completing crop distribution mapping. Previous studies have found that farmland landscape information such as crop abundance, clustering, fragmentation, plot shape, and plot size within the monitoring area significantly influences crop remote sensing identification features and classification algorithms.

[0057] Current research on farmland landscapes uses remote sensing data of relatively coarse scale to characterize the complexity of farmland landscapes, resulting in a general understanding of this complexity and a lack of detailed characterization. Furthermore, existing studies often rely on prior knowledge such as historical crop distribution maps to summarize the relationship between farmland landscape complexity and classification accuracy, leading to poor model transferability.

[0058] Therefore, this invention, based on high-resolution UAV imagery from field surveys depicting the real-world farmland landscape, analyzes the complexity of the farmland landscape. By constructing a remote sensing model of landscape complexity, it achieves farmland landscape complexity zoning without relying on prior knowledge. The impact of landscape factors on crop remote sensing classification under different complexity zoning is analyzed, the indicative role of farmland landscape information in crop remote sensing classification is summarized, and a crop remote sensing classification method based on farmland landscape complexity zoning is established, improving crop classification accuracy. The research results of this invention provide important methodological and theoretical references for using farmland landscape complexity information to guide crop remote sensing classification.

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0060] The following is combined Figures 1 to 12 This invention describes a crop classification method and apparatus based on a landscape complexity remote sensing model.

[0061] Figure 1 This is one of the flowcharts illustrating the crop classification method based on a landscape complexity remote sensing model provided by this invention, such as... Figure 1 As shown, the method includes steps 101, 102, 103, 104 and 105.

[0062] Figure 2This is the second schematic diagram of the crop classification method based on a landscape complexity remote sensing model provided by the present invention, as shown below. Figure 2 As shown.

[0063] According to the embodiments provided by the present invention, firstly, based on high-resolution image data from UAVs, the farmland landscape is quantitatively described, and then a remote sensing perception model of the farmland landscape is constructed based on the Random Forest (RF) algorithm through feature construction, feature optimization, etc.

[0064] Based on this, the complexity of farmland landscape in the study area was evaluated using satellite imagery data, and the landscape perception zoning of the study area was obtained using the natural breakpoint method.

[0065] Finally, within each landscape zone, the selection of crop remote sensing identification features and the selection of the best classifier were carried out to conduct accurate identification of crops within the zone. The optimal identification results within the zone were then stitched together to obtain the optimal classification and identification results of crops in the study area.

[0066] Step 101: Obtain satellite remote sensing image data of farmland landscape.

[0067] According to the embodiments provided by the present invention, remote sensing image data of farmland landscape is acquired via satellite.

[0068] Step 102: Based on the satellite remote sensing image data of the farmland landscape, obtain the optimal feature set for remote sensing perception of the complexity of the farmland landscape.

[0069] Furthermore, based on the satellite remote sensing image data of the farmland landscape, the optimal feature set for remote sensing perception of farmland landscape complexity is obtained, including:

[0070] Based on the satellite remote sensing image data of the farmland landscape, the landscape features of the farmland landscape are constructed;

[0071] One or more landscape features of the farmland landscape are input into the landscape complexity remote sensing model to obtain different sensing accuracy results corresponding to one or more landscape features of the farmland landscape output by the landscape complexity remote sensing model.

[0072] By selecting one or more landscape features corresponding to the highest perception accuracy results, the optimal feature set for remote sensing perception of farmland landscape complexity is obtained.

[0073] The elements in the optimal feature set include one or more optimal feature subsets; the elements in the optimal feature subset include one or more landscape features.

[0074] According to the embodiments provided by the present invention, when constructing the image features of the farmland landscape complexity perception model, based on the satellite remote sensing image data of the farmland landscape, statistical values ​​of image features such as band reflectance features, vegetation index features, and texture features within landscape units are selected to characterize the complexity of the farmland landscape. The standard deviation and mean within the landscape unit grid are selected as landscape features. Table 1 is the image feature table for farmland landscape complexity perception provided by the present invention, as shown in Table 1. A total of 30 features are used to construct the initial landscape features, including 5 original band reflectance, 9 vegetation indices, and 16 texture indices. The mean and standard deviation of each initial landscape feature are calculated within the landscape unit, resulting in 60 landscape features. The landscape features are named with "feature name_mean" or "feature name_std", such as: rededge_std, ndvi_mean, savi_std, ent_mean, etc.

[0075] Table 1. Features of Farmland Landscape Perceptual Imagery

[0076]

[0077] Step 103: Input the optimal feature set of the farmland landscape complexity remote sensing into the landscape complexity remote sensing model to obtain the farmland landscape perception raster output by the landscape complexity remote sensing model.

[0078] Furthermore, the steps for establishing the landscape complexity remote sensing model are as follows:

[0079] Acquire drone remote sensing image data of farmland landscape;

[0080] Landscape features are extracted from the UAV remote sensing image data. These landscape features include band reflectance features, vegetation index features, and texture features, along with the standard deviation and mean of the landscape unit grid within the farmland landscape.

[0081] Based on the UAV remote sensing image data and the landscape features, construct one or more decision trees;

[0082] Based on the one or more decision trees, a remote sensing model for landscape complexity is established through ensemble learning.

[0083] According to the embodiments provided by the present invention, the following requirements exist for the remote sensing perception model of landscape complexity: For farmland landscape patterns, the features representing landscape complexity are high-dimensional data, which require regression algorithms to process; the regression algorithm needs to have stable performance, and stable performance is more conducive to transfer to other farmland scenarios; the model perception of farmland landscape is to serve the subsequent remote sensing classification of crops, and the preliminary data processing process must be simple.

[0084] The advantages of the random forest regression model meet the experimental requirements. This is because random forest regression offers the following advantages: it can handle large amounts of high-dimensional data; it has high accuracy and stability; it does not require feature scaling or regularization; and it can automatically handle missing and outlier values. These advantages effectively fulfill the aforementioned requirements.

[0085] Random forest regression is an ensemble learning method based on decision trees that can be used for regression and classification problems. Random forests construct multiple decision trees by randomly selecting samples and features, and then combine them to improve predictive performance. Within each decision tree, the selection of samples and features is randomized, which reduces overfitting and improves the model's generalization ability. Its model construction process is as follows:

[0086] 1) Random sample selection: A certain number of samples are randomly selected from the original dataset (usually using the Bootstrap algorithm) to construct a new training set. Samples not selected can be used as test samples to evaluate the accuracy of the training results; these samples are statistically referred to as out-of-bag (OOB) samples.

[0087] 2) Random feature selection: Randomly select a certain number of features from all features. For regression problems, the number of features selected is generally the square root of the total number of features.

[0088] 3) Construct a decision tree: Based on the new training set and randomly selected features, construct a decision tree. At each node, select the best feature to split the data, and output the average value of the samples contained in each leaf node.

[0089] 4) Repeat steps 1 to 3: Repeat steps 1 to 3 to build multiple decision trees.

[0090] 5) Prediction Results: The generated decision trees are independent of each other and have equal importance, so there is no need to consider their weights. For new input data, the average result of the output of all individual decision trees is used as the final prediction result.

[0091] Based on the aforementioned model construction process, a remote sensing model of landscape complexity can be established by using remote sensing image data of farmland landscapes acquired by UAVs as the original dataset.

[0092] According to the embodiments provided by the present invention, Figure 3 This is a schematic diagram of the two-dimensional complexity conceptual model provided by the present invention, such as... Figure 3 As shown, Figure 3In the diagram, (a), (b), (c), and (d) represent different landscape units, and different colors represent different types of crops. Existing research on the complexity of farmland landscapes involves subjective descriptions across multiple dimensions, including plot size, plot regularity, distribution clustering, and species richness, lacking generalization and condensation. Therefore, this invention draws on ecological descriptions of landscapes and proposes a two-dimensional complexity model based on the complexity of constituent elements and spatial configuration to summarize farmland landscapes.

[0093] according to Figure 3 The two-dimensional complexity conceptual model shown in the paper indicates that the complexity of farmland landscapes mainly stems from two aspects: one is the complexity of composition brought about by different crop types, such as... Figure 3 The comparison between (a) and (c) and (b) and (d) shows that the crops are spatially consistent, but the increased variety of crops leads to a more complex farmland landscape, increasing complexity. Secondly, the different spatial distribution patterns result in greater spatial configuration complexity, such as... Figure 3 The comparisons in (a) and (b) and (c) and (d) show that the types of crops are exactly the same, but the fragmentation of their distribution leads to a more complex farmland landscape.

[0094] Figure 3 The different sub-maps in the image represent typical farmland landscape complexity. Figure 3 (a) represents a farmland landscape type with low complexity of constituent components and spatial configuration. This type of landscape is the simplest, with few types of crops and large, concentrated plots. Figure 3 (d) represents a farmland landscape type with high complexity of composition and spatial configuration. This type of landscape is the most complex, with many types of crops and small, scattered plots. Figure 3 In the diagram, (b) and (c) represent farmland landscapes where only spatial configuration complexity and component complexity are prominent, respectively. The complexity of these two types of farmland landscapes falls between... Figure 3 Between (a) and (d) in the text.

[0095] According to the embodiments provided by the present invention, a mathematical expression for the complexity of farmland landscape is also provided. Table 2 is a table showing the relationship between the farmland landscape index provided by the present invention and the accuracy of crop classification.

[0096] First, landscape indices were selected from two dimensions: the complexity of constituent elements and the complexity of spatial configuration. These included indices characterizing spatial configuration complexity: shape factor (LSI), fractal dimension (FD), and clustering index (AI), as well as an index characterizing constituent element complexity: evenness (SHEI).

[0097] Table 2. Farmland Landscape Index and its Relationship with Crop Classification Accuracy

[0098]

[0099] The four landscape factors selected in this invention represent the landscape information and classification accuracy patterns. For example, as the fragmentation increases, the classification accuracy decreases, while as the crop number increases, the planting structure becomes more uniform, the spatial distribution becomes more clustered, and the plot shape becomes more square, the recognition accuracy shows a clear upward trend.

[0100]

[0101] In the formula, For shape factor, This indicates the perimeter of a patch, which refers to fragmented plots of land within an agricultural landscape. It is the total area of ​​the landscape.

[0102]

[0103] In the formula, Let fractal dimension be the number of ... Indicates the perimeter of the plaque. It is a constant for raster landscapes. Take 4. It is the area of ​​the patch.

[0104]

[0105] In the formula, This is the aggregation index. The patches are calculated based on the number of grids. circumference, The patches are calculated based on the number of grid cells. area, It represents the total number of grid cells in the landscape.

[0106]

[0107]

[0108] In the formula, For uniformity, It is the Shannon Diversity Index. It represents the maximum possible uniformity of the landscape. This represents the total number of landscape types.

[0109] The impact of landscape information on classification can be expressed by a monotonic landscape factor. This is because classification accuracy is determined by both landscape and non-landscape information; let the classification accuracy be... Landscape information is in vector form. , representing the four landscape indices in Table 2, and other factors determining classification accuracy are vectors. This includes, but is not limited to, classifiers, classification features, image quality, etc. Therefore, a functional relationship exists:

[0110]

[0111] During the classification process When it is a constant value, Landscape information The function, let the comprehensive landscape index be... (Monotonic landscape factor), its calculation formula is:

[0112]

[0113] According to the monotonicity of combinatorial functions, With classification accuracy There is a theoretically monotonically increasing functional relationship between them. That is, this index describes the difficulty of remote sensing classification of crops for the corresponding farmland landscape. The larger the value, the higher the classification accuracy of crops under the same non-landscape conditions such as image quality, classifier, and classification features.

[0114] Step 104: Based on the farmland landscape perception grid, obtain multiple perception zones of the farmland landscape.

[0115] According to the embodiments provided by the present invention, based on satellite remote sensing image data, the optimal feature set for remote sensing perception of farmland landscape complexity is calculated, and the feature set is input into the constructed landscape complexity remote sensing perception model. A 500m×500m landscape grid is deployed in the study area, and the landscape perception grid result of the study area is calculated. The farmland landscape perception zoning of the study area is obtained using the natural breakpoint method.

[0116] Step 105: Based on the multiple sensing zones of the farmland landscape, classify the crops in the multiple sensing zones to obtain a spatial distribution map of the crops.

[0117] Furthermore, the process of classifying crops in multiple sensing zones based on the farmland landscape to obtain a spatial distribution map of the crops includes:

[0118] Based on the multiple sensing zones of the farmland landscape, the optimal feature set within the multiple sensing zones of the farmland landscape is selected to obtain the optimal feature set corresponding to each sensing zone.

[0119] The optimal feature set corresponding to each perceptual partition is input into multiple classifiers to obtain the classification result output by each classifier.

[0120] By comparing the classification results output by the multiple classifiers, the optimal classification result corresponding to each perceptual partition is obtained;

[0121] By stitching together the optimal classification results corresponding to each sensing partition, a spatial distribution map of the crop is obtained.

[0122] Furthermore, the step of optimizing the optimal feature set within the multiple sensing zones of the farmland landscape, based on the multiple sensing zones of the farmland landscape, to obtain the optimal feature set corresponding to each sensing zone, includes:

[0123] Based on multiple perceptual partitions of the farmland landscape, the optimal feature set within the multiple perceptual partitions of the farmland landscape is obtained;

[0124] Based on the optimal feature set within multiple perceptual zones of the farmland landscape, the importance of each feature in the optimal feature set is evaluated by repeatedly training one or more models; wherein, the higher the importance, the greater the contribution of the feature to the model performance, and vice versa.

[0125] Based on the degree of importance, features that contribute little to the model performance in the best feature set of multiple perception zones of the farmland landscape are gradually removed until the number of remaining features in the best feature set reaches a set threshold.

[0126] The best feature set within multiple sensing zones of the farmland landscape, where the number of features reaches a set threshold, is taken as the optimal feature set for each sensing zone.

[0127] Furthermore, the classifier includes one or more of the following: decision tree classifier, random forest classifier, support vector machine, and minimum distance classifier.

[0128] According to the embodiments provided by the present invention, the optimal feature set for crop classification is selected using the recursive feature elimination (RFE) method in different landscape zones. At the same time, four classification methods, namely decision tree classification, random forest classification, support vector machine, and minimum distance classification, are selected. The optimal feature set and the optimal classification method are selected in each landscape zone, and remote sensing identification of crops in each zone is performed. Finally, the optimal classification results of the zones are stitched together to obtain the spatial distribution map of crops in the study area.

[0129] According to embodiments of the present invention, Recursive Feature Elimination (RFE) is a feature optimization method employed in this invention. Its main idea is to gradually eliminate features that contribute little to model performance by repeatedly training the model. In each iteration, the method sorts the features according to their importance and deletes the lowest-ranked features. This process continues until the number of remaining features reaches a set threshold or a specified number of features.

[0130] During the RFE (Real-Time Evaluation) process, various models can be used to evaluate the importance of each feature, with Random Forest being a commonly used choice. Random Forest is an ensemble learning method based on decision trees. Its main idea is to reduce the model's variance by training multiple decision trees. In Random Forest, the training samples and features for each decision tree are obtained through random sampling, which effectively avoids overfitting.

[0131] Based on any of the above embodiments, the present invention also provides the following specific experimental scenarios for further explanation.

[0132] This invention selects a region to construct a remote sensing model for landscape complexity, and selects a portion of the area as an experimental zone to conduct research on remote sensing identification of crops based on landscape zoning. Within the experimental zone, a total of 22 high spatial resolution multispectral farmland landscape images from UAVs, as well as one satellite image, were collected.

[0133] Based on any of the above embodiments Figure 4 This is a diagram of the feature optimization process provided by the present invention, such as... Figure 4 As shown in the figure. The experiment modeled all feature numbers from 1 to 60. For each feature number, a random forest regression model was used to cross-validate the sample data. Each model construction used a 12:10 training-to-test ratio for data grouping, and this was repeated randomly five times to obtain the perceptual accuracy R corresponding to the five test sets. 2 The average value is used as the model score, and the optimal feature set is the best feature set consisting of the number of features corresponding to the best model score.

[0134] Specifically, by comparing grouped experiments on features, the optimal remote sensing model for farmland landscape complexity was obtained. Table 3 shows the accuracy results of the remote sensing model for landscape complexity involving different features provided in this invention.

[0135] Table 3. Accuracy Results of Remote Sensing Models for Landscape Complexity with Different Features

[0136]

[0137] According to the embodiments provided by the present invention, the present invention analyzes the spatial scale characteristics of the landscape complexity remote sensing perception model based on UAV remote sensing images of farmland landscape.

[0138] Specifically, resampling methods (such as Lansos resampling, nearest neighbor interpolation, bilinear interpolation, etc.) are used. Lansos resampling will be used as an example below. A 10m series of images was acquired through Lansos resampling, and the impact of spatial resolution on perceptual features and perceptual accuracy was analyzed. The analysis results are as follows: Figure 5 and Figure 6 As shown.

[0139] Lansos resampling is an image scaling algorithm that resamples the original image using convolution to generate a new image. The algorithm uses the Lansos kernel, which has excellent approximation properties in the frequency domain, producing smoother images. Compared to other image resampling algorithms, Lansos image resampling preserves more image details and reduces image artifacts, thus providing higher quality image resampling results.

[0140] Figure 5 This invention provides a map showing the perception accuracy achieved by the optimal feature subset at different spatial resolutions and the required number of features, such as... Figure 5 As shown.

[0141] Figure 6 This is a schematic diagram of the optimal feature subset at different spatial resolutions provided by the present invention, such as... Figure 6 As shown in the figure, the features corresponding to the green blocks represent one or more landscape features that correspond to the highest perception accuracy at this spatial resolution. The one or more landscape features that correspond to the highest perception accuracy at this spatial resolution constitute an optimal feature subset corresponding to this spatial resolution.

[0142] For example, at a spatial resolution of 10m, the landscape features corresponding to the highest perception accuracy include: the mean of the red-edge mean texture, the mean of the red-edge standard deviation texture, the standard deviation of the red-edge standard deviation texture, the mean of the near-infrared mean texture, the standard deviation of the near-infrared mean texture, the mean of the near-infrared standard deviation texture, the mean of the near-infrared anisotropy texture, the mean of the greenness index (GI), the mean of the normalized difference vegetation index (NDVI), the standard deviation of the normalized difference vegetation index (NDVI), the mean of the greening index (VIgreen), the standard deviation of the renormalized difference vegetation index (RDVI), the standard deviation of the soil-adjusted vegetation index (SAVI), the mean of the red-edge normalized difference vegetation index (RENDVI), the mean of the red-edge converted chlorophyll vegetation index (RETCARI), the mean of the red-edge triangular vegetation index (RETVI), the standard deviation of the red-edge triangular vegetation index (RETVI), the standard deviation of the blue band, the standard deviation of the red band, and the standard deviation of the red-edge band.

[0143] Figure 7 This is the landscape zoning result map of the experimental area provided by the present invention, such as... Figure 7 As shown. Within the experimental area, combined with the natural discontinuity method, the landscape was divided into three categories. Among them, the landscape units along the roads and near villages and towns had low landscape factor scores and were not easy to classify, and were defined as Zone C (complex zone). Within the cultivated land, the landscape factor scores also varied, with some landscape units being more homogeneous and defined as Zone A (simple zone), while others were more complex and defined as Zone B (medium zone).

[0144] Then, feature optimization and optimal classifier selection were performed within different partitions. The results of feature optimization are as follows:

[0145] Within the experimental area, without considering landscape complexity, the globally optimal classification features are: B2, B3, B11, GCVI, LSWI, NDVI, NRED2, NRED3, NIR_ASM, NIR_CONT, NIR_CORR, NIR_HOM, RE_ASM, RE_CONT, RE_CORR, RE_DISS.

[0146] The optimal classification features for region A are: B2, B3, B4, B11, GCVI, LSWI, NDVI, NRED2, NIR_ASM, NIR_CONT, NIR_CORR, RE_ASM, RE_CONT, RE_CORR.

[0147] The optimal classification features for region B are: B2, B3, B11, GCVI, LSWI, NIR_ASM, NIR_CONT, NIR_CORR, NIR_HOM, RE_ASM, RE_CONT, and RE_CORR.

[0148] The optimal classification features for region C are: B2, B3, B11, GCVI, LSWI, NDVI, NRED3, RETCARI, NIR_ASM, NIR_CONT, NIR_CORR, NIR_HOM, RE_ASM, RE_CONT, RE_CORR.

[0149] B2, B3, B4, and B11 represent the original band reflectance of satellite remote sensing data. Specifically, B2 is the blue light band reflectance, B3 is the green light band reflectance, B4 is the red light band reflectance, and B11 is the shortwave infrared band reflectance.

[0150] GCVI, LSWI, NDVI, NRED2, NRED3, and RETCARI are different vegetation indices;

[0151] Green chlorophyll vegetation index (GCVI), vegetation moisture index (LSWI), normalized difference vegetation index (NDVI), normalized difference red edge index (NDRE1,2,3) and red edge converted chlorophyll spectral absorption index (RETCARI).

[0152] NIR_ASM, NIR_CONT, NIR_CORR, NIR_HOM, RE_ASM, RE_CONT, RE_CORR, and RE_DISS are all texture indices. In these abbreviations, NIR represents the near-infrared band, ASM is the second-order moment texture, CONT is the contrast texture, CORR is the correlation texture, HOM is the local stationary texture, RE represents the red-edge band, and DISS is the dissimilarity texture.

[0153] Specifically, NIR_ASM is the near-infrared angular second-moment texture, NIR_CONT is the near-infrared contrast texture, NIR_CORR is the near-infrared band correlation texture, NIR_HOM is the near-infrared band local stationary texture, RE_ASM is the red-edge band angular second-moment texture, RE_CONT is the red-edge band contrast texture, RE_CORR is the red-edge band correlation texture, and RE_DISS is the red-edge band dissimilarity texture. These are used to define the combination of texture indices.

[0154] Figure 8 This is an evaluation chart of the classifier performance for different landscape zones provided by the present invention, such as... Figure 8 As shown. Random forests and decision trees perform similarly, both achieving good classification results, with random forests outperforming decision trees. The minimum distance classifier performs the worst. Support vector machines (SVMs) have an overall classification performance between tree models and minimum distance models, but achieve the best accuracy in region B, outperforming other models.

[0155] The random forest model performs best in regions A and C, while the support vector machine model performs best in region B. The random forest model achieves overall classification accuracy of 97% and 94% in regions A and C, respectively, while the support vector machine achieves the best accuracy of 93% in region B. Therefore, the random forest model can be chosen to classify crops in regions A and C, and the support vector machine can be chosen to classify crops in region B. Finally, the classification results from regions A, B, and C are concatenated to obtain the crop distribution map with the highest classification accuracy. Figure 9 This invention provides a crop distribution map of an experimental area based on farmland landscape zoning, such as... Figure 9 As shown.

[0156] Figure 10 This is a comparison chart of crop classification accuracy provided by the present invention. The blue line represents the accuracy result of crop classification using traditional methods, while the orange line represents the accuracy result of crop classification using the crop classification method based on a landscape complexity remote sensing model provided by the present invention. Figure 10 As shown, after utilizing the landscape complexity zoning classification strategy proposed in this invention, the overall classification accuracy of the experimental area reached 95%, which is 2% higher than the 93% of the traditional method, and the Kappa coefficient increased from 0.91 to 0.93.

[0157] This invention provides a crop classification method based on a landscape complexity remote sensing model. It acquires satellite remote sensing image data of farmland landscapes, obtains the optimal feature set for remote sensing of farmland landscape complexity based on this data, and then uses the landscape complexity remote sensing model to accurately partition the farmland landscape, obtaining multiple sensing partitions. Finally, based on these multiple sensing partitions, crops within each partition are classified, thus improving the accuracy of crop classification.

[0158] The farmland landscape complexity perception model established in this invention can directly assess farmland landscape complexity based on remote sensing images of the experimental area, without relying on historical crop distribution data of the experimental area, thus giving the model good transferability. By constructing a landscape complexity remote sensing perception model, farmland landscape complexity zoning is achieved without relying on prior knowledge. The influence of landscape factors on crop remote sensing classification under different landscape complexity zoning is analyzed, the indicative role of farmland landscape information on crop remote sensing classification is summarized, and a crop remote sensing classification method based on farmland landscape complexity zoning is established. Based on farmland landscape zoning, crops are identified and classified, improving the accuracy of crop classification.

[0159] The crop classification device based on the landscape complexity remote sensing model provided by the present invention will be described below. The crop classification device based on the landscape complexity remote sensing model described below can be referred to in correspondence with the crop classification method based on the landscape complexity remote sensing model described above.

[0160] Based on any of the above embodiments Figure 11 This is a schematic diagram of the structure of a crop classification device based on a landscape complexity remote sensing model provided in an embodiment of the present invention, as shown below. Figure 11 As shown, this embodiment of the invention provides a crop classification device based on a landscape complexity remote sensing model, including an acquisition module 1101, a feature module 1102, a sensing module 1103, a partitioning module 1104, and a classification module 1105, wherein:

[0161] The acquisition module 1101 is used to acquire satellite remote sensing image data of farmland landscape; the feature module 1102 is used to obtain the optimal feature set for remote sensing perception of farmland landscape complexity based on the satellite remote sensing image data of farmland landscape; the perception module 1103 is used to input the optimal feature set for remote sensing perception of farmland landscape complexity into the remote sensing perception model of landscape complexity to obtain the farmland landscape perception grid output by the remote sensing perception model of landscape complexity; the partitioning module 1104 is used to obtain multiple perception partitions of farmland landscape based on the farmland landscape perception grid; the classification module 1105 is used to classify the crops in the multiple perception partitions of farmland landscape to obtain the spatial distribution map of the crops.

[0162] The crop classification device based on a landscape complexity remote sensing model provided in this invention acquires satellite remote sensing image data of farmland landscapes, obtains the optimal feature set for remote sensing of farmland landscape complexity based on the satellite remote sensing image data of the farmland landscape, and then uses the landscape complexity remote sensing model to achieve complex zoning of farmland landscapes without relying on prior knowledge. This makes the landscape complexity remote sensing model have good transferability and obtains multiple sensing zones of farmland landscapes. Then, based on the multiple sensing zones of farmland landscapes, the crops in the multiple sensing zones are classified, thereby improving the accuracy of crop classification.

[0163] Figure 12 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 12 As shown, the electronic device may include: a processor 1210, a communications interface 1220, a memory 1230, and a communication bus 1240, wherein the processor 1210, the communications interface 1220, and the memory 1230 communicate with each other via the communication bus 1240. The processor 1210 can call logical instructions from the memory 1230 to execute a crop classification method based on a landscape complexity remote sensing model. This method includes:

[0164] Acquire satellite remote sensing image data of farmland landscapes;

[0165] Based on the satellite remote sensing image data of the farmland landscape, the optimal feature set for remote sensing perception of the complexity of the farmland landscape is obtained;

[0166] The optimal feature set of the farmland landscape complexity remote sensing is input into the landscape complexity remote sensing model to obtain the farmland landscape sensing raster output by the landscape complexity remote sensing model.

[0167] Based on the farmland landscape perception grid, multiple perception zones of the farmland landscape are obtained;

[0168] Based on multiple sensing zones of the farmland landscape, the crops in the multiple sensing zones are classified to obtain a spatial distribution map of the crops.

[0169] Furthermore, the logical instructions in the aforementioned memory 1230 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0170] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the crop classification method based on the landscape complexity remote sensing perception model provided by the above methods, the method comprising:

[0171] Acquire satellite remote sensing image data of farmland landscapes;

[0172] Based on the satellite remote sensing image data of the farmland landscape, the optimal feature set for remote sensing perception of the complexity of the farmland landscape is obtained;

[0173] The optimal feature set of the farmland landscape complexity remote sensing is input into the landscape complexity remote sensing model to obtain the farmland landscape sensing raster output by the landscape complexity remote sensing model.

[0174] Based on the farmland landscape perception grid, multiple perception zones of the farmland landscape are obtained;

[0175] Based on multiple sensing zones of the farmland landscape, the crops in the multiple sensing zones are classified to obtain a spatial distribution map of the crops.

[0176] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the crop classification method based on a landscape complexity remote sensing model provided by the above methods, the method comprising:

[0177] Acquire satellite remote sensing image data of farmland landscapes;

[0178] Based on the satellite remote sensing image data of the farmland landscape, the optimal feature set for remote sensing perception of the complexity of the farmland landscape is obtained;

[0179] The optimal feature set of the farmland landscape complexity remote sensing is input into the landscape complexity remote sensing model to obtain the farmland landscape sensing raster output by the landscape complexity remote sensing model.

[0180] Based on the farmland landscape perception grid, multiple perception zones of the farmland landscape are obtained;

[0181] Based on multiple sensing zones of the farmland landscape, the crops in the multiple sensing zones are classified to obtain a spatial distribution map of the crops.

[0182] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A crop classification method based on a landscape complexity remote sensing model, characterized in that, include: Acquire satellite remote sensing image data of farmland landscapes; Based on the satellite remote sensing image data of the farmland landscape, the optimal feature set for remote sensing perception of the complexity of the farmland landscape is obtained; The optimal feature set of the farmland landscape complexity remote sensing is input into the landscape complexity remote sensing model to obtain the farmland landscape sensing raster output by the landscape complexity remote sensing model. Based on the farmland landscape perception grid, multiple perception zones of the farmland landscape are obtained; Based on multiple sensing zones of the farmland landscape, the crops in the multiple sensing zones are classified to obtain a spatial distribution map of the crops; The steps for establishing the landscape complexity remote sensing model are as follows: Acquire drone remote sensing image data of farmland landscape; Landscape features are extracted from the UAV remote sensing image data. These landscape features include band reflectance features, vegetation index features, and texture features, along with the standard deviation and mean of these features within the landscape unit grid of the farmland landscape. The landscape unit grid consists of regular units of a preset size within the farmland landscape. Based on the UAV remote sensing image data and the landscape features, multiple decision trees are constructed; Based on the aforementioned decision trees, a landscape complexity remote sensing model is established using an ensemble learning method. The satellite remote sensing image data based on the farmland landscape is used to obtain the optimal feature set for remote sensing perception of farmland landscape complexity, including: Based on the satellite remote sensing image data of the farmland landscape, the landscape features of the farmland landscape are constructed; One or more landscape features of the farmland landscape are input into the landscape complexity remote sensing model to obtain different sensing accuracy results corresponding to one or more landscape features of the farmland landscape output by the landscape complexity remote sensing model. By selecting one or more landscape features corresponding to the highest perception accuracy results, the optimal feature set for remote sensing perception of farmland landscape complexity is obtained. The elements in the optimal feature set include one or more optimal feature subsets; the elements in the optimal feature subset include one or more landscape features. The process of classifying crops in multiple sensing zones based on the farmland landscape to obtain a spatial distribution map of the crops includes: Based on the multiple sensing zones of the farmland landscape, the optimal feature set within the multiple sensing zones of the farmland landscape is selected to obtain the optimal feature set corresponding to each sensing zone. The optimal feature set corresponding to each perceptual partition is input into multiple classifiers to obtain the classification result output by each classifier. By comparing the classification results output by the multiple classifiers, the optimal classification result corresponding to each perceptual partition is obtained; By stitching together the optimal classification results corresponding to each sensing partition, a spatial distribution map of the crop is obtained; The process involves selecting the optimal feature set within each of the multiple sensing zones of the farmland landscape, based on the farmland landscape, to obtain the optimal feature set corresponding to each sensing zone. This includes: Based on multiple perceptual partitions of the farmland landscape, the optimal feature set within the multiple perceptual partitions of the farmland landscape is obtained; Based on the optimal feature set within multiple perceptual zones of the farmland landscape, the importance of each feature in the optimal feature set is evaluated by repeatedly training one or more models; wherein, the higher the importance, the greater the contribution of the feature to the model performance, and vice versa. Based on the degree of importance, features that contribute little to the model performance in the best feature set of multiple perception zones of the farmland landscape are gradually removed until the number of remaining features in the best feature set reaches a set threshold. The best feature set within multiple sensing zones of the farmland landscape, where the number of features reaches a set threshold, is taken as the optimal feature set for each sensing zone.

2. The crop classification method based on a landscape complexity remote sensing model according to claim 1, characterized in that, The classifier includes one or more of the following: decision tree classifier, random forest classifier, support vector machine, and minimum distance classifier.

3. A crop classification device based on a landscape complexity remote sensing model, characterized in that, include: The acquisition module is used to acquire satellite remote sensing image data of farmland landscapes; The feature module is used to obtain the optimal feature set for remote sensing perception of the complexity of the farmland landscape based on the satellite remote sensing image data of the farmland landscape. The perception module is used to input the optimal feature set of the farmland landscape complexity remote sensing perception into the landscape complexity remote sensing perception model to obtain the farmland landscape perception raster output by the landscape complexity remote sensing perception model. The partitioning module is used to obtain multiple perceptual partitions of the farmland landscape based on the farmland landscape perception grid. The classification module is used to classify the crops in the multiple sensing zones of the farmland landscape based on the multiple sensing zones, and obtain the spatial distribution map of the crops; The steps for establishing the landscape complexity remote sensing model are as follows: Acquire drone remote sensing image data of farmland landscape; Landscape features are extracted from the UAV remote sensing image data. These landscape features include band reflectance features, vegetation index features, and texture features, along with the standard deviation and mean of these features within the landscape unit grid of the farmland landscape. The landscape unit grid consists of regular units of a preset size within the farmland landscape. Based on the UAV remote sensing image data and the landscape features, multiple decision trees are constructed; Based on the aforementioned decision trees, a landscape complexity remote sensing model is established using an ensemble learning method. The satellite remote sensing image data based on the farmland landscape is used to obtain the optimal feature set for remote sensing perception of farmland landscape complexity, including: Based on the satellite remote sensing image data of the farmland landscape, the landscape features of the farmland landscape are constructed; One or more landscape features of the farmland landscape are input into the landscape complexity remote sensing model to obtain different sensing accuracy results corresponding to one or more landscape features of the farmland landscape output by the landscape complexity remote sensing model. By selecting one or more landscape features corresponding to the highest perception accuracy results, the optimal feature set for remote sensing perception of farmland landscape complexity is obtained. The elements in the optimal feature set include one or more optimal feature subsets; the elements in the optimal feature subset include one or more landscape features. The process of classifying crops in multiple sensing zones based on the farmland landscape to obtain a spatial distribution map of the crops includes: Based on the multiple sensing zones of the farmland landscape, the optimal feature set within the multiple sensing zones of the farmland landscape is selected to obtain the optimal feature set corresponding to each sensing zone. The optimal feature set corresponding to each perceptual partition is input into multiple classifiers to obtain the classification result output by each classifier. By comparing the classification results output by the multiple classifiers, the optimal classification result corresponding to each perceptual partition is obtained; By stitching together the optimal classification results corresponding to each sensing partition, a spatial distribution map of the crop is obtained; The process involves selecting the optimal feature set within each of the multiple sensing zones of the farmland landscape, based on the farmland landscape, to obtain the optimal feature set corresponding to each sensing zone. This includes: Based on multiple perceptual partitions of the farmland landscape, the optimal feature set within the multiple perceptual partitions of the farmland landscape is obtained; Based on the optimal feature set within multiple perceptual zones of the farmland landscape, the importance of each feature in the optimal feature set is evaluated by repeatedly training one or more models; wherein, the higher the importance, the greater the contribution of the feature to the model performance, and vice versa. Based on the degree of importance, features that contribute little to the model performance in the best feature set of multiple perception zones of the farmland landscape are gradually removed until the number of remaining features in the best feature set reaches a set threshold. The best feature set within multiple sensing zones of the farmland landscape, where the number of features reaches a set threshold, is taken as the optimal feature set for each sensing zone.

4. 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 classification method based on the landscape complexity remote sensing model as described in any one of claims 1 or 2.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the crop classification method based on the landscape complexity remote sensing model as described in any one of claims 1 or 2.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the crop classification method based on the landscape complexity remote sensing model as described in any one of claims 1 or 2.

Images