Corn planting area corn identification method based on spatial-temporal feature selection

Abstract

The invention provides a corn planting area corn identification method based on spatial-temporal feature selection, and belongs to the technical field of remote sensing. The method comprises the following steps: acquiring a remote sensing image of a corn planting area, determining an NDVI characteristic curve of each pixel in the corn planting area, and screening out corn pixels conforming to key growth period planting characteristics as corn sample points; based on the NDVI characteristic curves of the corn sample points and the non-corn sample points, determining the importance of different vegetation characteristics to corn identification; removing redundant features by integrating importance and feature correlation to obtain classification features; calculating the time sequence-spectrum joint similarity between the to-be-classified pixels and the corn sample points and the non-corn sample points; and classifying the to-be-classified pixels into a category corresponding to the minimum hour sequence-spectrum joint similarity. By using the method, the problems of insufficient separability and feature redundancy in corn identification of a traditional classification method and the problem of agricultural statistical data threshold dependence of a TWDTW method can be solved.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing technology, and in particular relates to a corn identification method for corn planting areas based on spatiotemporal feature selection. Background Art

[0002] Corn is one of the world's most important food and cash crops, serving as a staple food for humans and livestock feed, as well as biofuels and industrial starch feed. Therefore, it's crucial to accurately understand the spatial distribution and production dynamics of corn cultivation.

[0003] In recent years, breakthroughs in satellite remote sensing technology have provided a new paradigm for crop identification. Compared to traditional methods that rely on manual surveys, this approach significantly reduces labor costs and improves monitoring efficiency. Acquiring continuous spatiotemporal observation data of vegetation surfaces based on remote sensing imagery has become a core technology for identifying crop planting areas.

[0004] Existing crop identification methods primarily include supervised learning-based classification and phenological curve similarity methods. Supervised learning methods utilize labeled samples to train models, extracting features from remote sensing images and generating crop distribution maps. Machine learning methods such as random forests (RF) and support vector machines (SVM), as well as supervised learning methods such as convolutional neural networks (CNN) and long short-term memory networks (LSTM), have been widely used in time series remote sensing crop classification tasks and have demonstrated significant success in identifying typical crops such as wheat and rice. However, these supervised learning methods are highly dependent on the number of labeled samples, making them costly for large-scale crop mapping. Phenological analysis methods based on curve similarity offer a novel approach. The weighted dynamic time warping (TWDTW) method quantifies the morphological differences in spectral trajectories across the crop growth cycle and embeds a phenological time weighting matrix, effectively enabling precise crop identification in complex agricultural environments with diverse climatic conditions. Existing research has shown that TWDTW exhibits superior classification performance compared to RF when training data is insufficient. However, in small sample data scenarios, the statistical representation extraction bias of the curve shape can easily lead to the expansion of the confidence interval, which in turn affects the classification results.

[0005] Traditional TWDTW research often relies solely on single spectral features for crop identification. This presents limitations in areas with complex crop structures and high spectral similarity. This is particularly true during the peak growing season, when the NDVI time series curves of corn are highly similar to those of soybeans, sorghum, and peanuts during the same growing season. Relying solely on single vegetation indices such as NDVI or EVI for identification can easily lead to confusion. In recent years, a growing number of studies have combined multi-source remote sensing features to identify multiple crops. Using the TWDTW method for corn identification, a study found that a two-band combination can effectively improve corn identification accuracy. While the use of multi-source features can improve crop classification accuracy to a certain extent, the simplistic superposition and blind addition of features can lead to information redundancy, increased computational burden, and reduced computational efficiency.

[0006] Existing TWDTW recognition methods, when distinguishing target from non-target crops, mostly rely on preset DTW distance thresholds based on regional agricultural statistical data, combined with the area of ​​the corn-growing region, to identify corn pixels. This method first uses the calculated DTW distance to find similarities between pixel points, and then spatializes the overall number based on the statistical data. This method has achieved high accuracy in mapping crops such as corn, rice, wheat, and sugarcane. However, statistical data are typically regional and macroscopic, making it difficult to accurately reflect the complexity and heterogeneity of crop spatial distribution. Using these as hard constraints for pixel-level classification or as a basis for threshold derivation can introduce uncertainty in crop spatial distribution. Furthermore, the acquisition of agricultural statistical data often has a certain lag, limiting the timeliness of classification results.

[0007] In summary, existing crop recognition methods have the following shortcomings:

[0008] 1. Although the method based on phenological curve similarity shows classification advantages under small sample conditions, insufficient samples can easily lead to deviations in the extraction of curve morphological statistical features, thereby widening the confidence interval of the results and ultimately affecting classification accuracy;

[0009] 2. Traditional corn recognition methods are limited in their ability to distinguish corn from summer crops with similar phenology. When using multi-source features for classification, there is often information redundancy, which restricts the accuracy and efficiency of recognition.

[0010] 3. Traditional TWDTW recognition methods mostly rely on DTW distance thresholds preset by regional agricultural statistical data when distinguishing target and non-target crops. This makes it difficult to accurately reflect the spatial distribution complexity of crops at the pixel scale and the heterogeneity within the plot.

[0011] In summary, these inherent limitations restrict the performance of corn recognition technology in practical applications, especially when faced with small sample sizes and in areas with complex crop planting structures, which often leads to recognition accuracy that is difficult to meet requirements and overall computational efficiency is low. Summary of the Invention

[0012] In view of this, the present invention provides a corn identification method for corn-growing areas based on spatiotemporal feature selection, which can solve the problems of insufficient separability and feature redundancy in corn identification of traditional classification methods and the threshold dependence problem of agricultural statistical data of TWDTW method.

[0013] In order to solve the above technical problems, the present invention is implemented as follows.

[0014] A corn identification method for corn planting areas based on spatiotemporal feature selection, comprising:

[0015] Step 1: Obtain remote sensing images of corn-growing areas;

[0016] Step 2: Determine corn sample points: Based on remote sensing images of the corn-growing area, determine the NDVI characteristic curve of each pixel in the corn-growing area. Based on the NDVI characteristic curve, select corn pixels that meet the planting characteristics of the key growth period as corn sample points;

[0017] Step 3: Classification feature selection: For corn sample points and non-corn sample points, calculate vegetation feature time series curves of N feature categories; based on the vegetation feature time series curves of corn sample points and non-corn sample points, use the Relief algorithm to determine the importance of different vegetation features for corn identification; based on the vegetation feature time series curves of corn sample points, determine the correlation between vegetation features; select a group of M vegetation features with the highest importance and mutual correlation no higher than the set correlation threshold as classification features;

[0018] The vegetation characteristic time series curves of M classification characteristics of each corn sample point and non-corn sample point constitute the sample set;

[0019] Step 4: Classification: Calculate the temporal-spectral joint similarity between the pixel to be classified and the corn sample points and non-corn sample points in the sample set; the temporal-spectral joint similarity between two pixels is expressed using the comprehensive TWDTW distance of the vegetation feature time series curves of M classification features; classify the pixel to be classified into the category corresponding to the minimum temporal-spectral joint similarity.

[0020] Preferably, in step 1, the remote sensing image of the corn-growing area is obtained by:

[0021] When the location of corn planting areas in the current year in the study area is known, the data of corn planting areas in the measured remote sensing images are directly used;

[0022] In the case that the location of the corn planting areas in the study area for the current year is unknown, the existing national corn dataset is used to determine the corn planting areas in the study area for the last three consecutive years as the predicted corn planting areas for the current year; the data of the corn planting areas for the current year predicted from the measured remote sensing images are used.

[0023] Preferably, the method further comprises preprocessing the acquired remote sensing image of the corn-growing area, including:

[0024] Decloud remote sensing images and retain images with cloud coverage below a set ratio;

[0025] Resample the bands of remote sensing images;

[0026] Perform linear interpolation of remote sensing images based on a selected time window;

[0027] Savitzky-Golay filtering is used to smooth remote sensing images.

[0028] Preferably, the method further comprises: using the cultivated land distribution provided by the land cover dataset CLCD as a mask file to mask the remote sensing image and exclude non-cultivated land pixels in the remote sensing image.

[0029] Preferably, in step 2, selecting corn pixels that meet the planting characteristics of the key growth period according to the NDVI characteristic curve as corn sample points includes:

[0030] Determine the critical growth period: Calculate the NDVI characteristic curve for each corn pixel based on remote sensing images of the corn-growing area. Average the data at the same time point for each NDVI characteristic curve to obtain the NDVI characteristic mean curve. Obtain the amplitude variation f between the maximum and minimum values ​​in the NDVI characteristic mean curve. The time interval between the time when the NDVI amplitude begins to exceed 20% of f and the time when it begins to fall below 20% of f is defined as the critical growth period.

[0031] Determine the key phenological fluctuation range: calculate the mean μ and standard deviation σ of the data at the same time point of each NDVI characteristic curve, and use [μ-σ, μ+σ] as the allowable range of NDVI characteristic curve fluctuation, that is, the key phenological fluctuation range;

[0032] For the NDVI characteristic curve of each pixel in the corn planting area, the NDVI characteristic curve segment of the key growth period is intercepted using the key growth period; for each pixel, it is determined whether the fluctuation amplitude of the intercepted NDVI characteristic curve segment is all within the key phenological fluctuation range. If so, the current pixel is identified as a corn sample point.

[0033] Preferably, during classification feature selection and classification, the key growth period is used to intercept the vegetation characteristic time series curves of corn sample points and pixels to be classified.

[0034] Preferably, in step 3, the correlation between vegetation characteristics is determined based on the vegetation characteristic time series curve of the corn sample point as follows:

[0035] For each vegetation characteristic, the vegetation characteristic time series mean curve of all corn sample points was calculated; for the vegetation characteristic time series mean curves of N vegetation characteristics, the Pearson correlation coefficient was calculated pairwise as the correlation between vegetation characteristics.

[0036] Preferably, in step 3, a group of M vegetation features with a higher importance and a mutual correlation not higher than a set correlation threshold are selected as classification features:

[0037] The N vegetation features are sorted by importance and selected one by one in descending order of importance; each time a vegetation feature is selected, it is determined whether the Pearson correlation coefficient between the vegetation feature and the determined classification feature is greater than a set correlation threshold; if it is greater, it is determined that the correlation is too large and the currently selected vegetation feature is not used as a classification feature; if it is less than or equal to the set correlation threshold, the currently selected vegetation feature is determined as a classification feature; until M classification features are selected.

[0038] Preferably, the time series-spectrum joint similarity is obtained in the following manner:

[0039] For the pixel to be classified and a corn sample point or non-corn sample point in the sample set, the TWDTW distance of each classification feature is calculated respectively, and the TWDTW distances of M classification features are calculated comprehensively and expressed as the comprehensive TWDTW distance.

[0040] Preferably, the N feature categories include vegetation index features and red edge vegetation index features:

[0041] The vegetation index features include: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Difference Vegetation Index (DVI), Optimized Soil Adjusted Vegetation Index (OSAVI), Spectral Ratio Index (SR), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Greenness Index (NDGI), Green Leaf Cover Index (GCVI), Ratio Vegetation Index (RVI), Normalized Difference Pollution Index (NDPI), Land Surface Water Index (LSWI), Photochemical Vegetation Reflectance Index (PSRI);

[0042] The red edge vegetation index features include: a first normalized red edge difference index NREDI1, a second normalized red edge difference index NREDI2, a third normalized red edge difference index NREDI3, a near infrared / red edge ratio index NIR / RE, a red band / red edge ratio index RED / RE, and a red edge position index REP.

[0043] Beneficial effects:

[0044] (1) Improving sample representativeness and constructing a standard growth curve: The present invention selects sample points that meet the planting characteristics of the key growth period based on morphological characteristics, and can construct a more representative corn standard growth curve, providing a more reliable basis for subsequent TWDTW matching, making TWDTW matching based on the minimum distance possible.

[0045] (2) Screening for separability features of corn to overcome reliance on single or pre-set features: This paper constructs a relief filtering feature importance assessment method to calculate the importance of different features for corn identification. It then comprehensively considers feature correlations to eliminate redundant features, and constructs a corn identification feature set that balances both feature importance and independence. This significantly reduces the omissions common in single NDVI methods, effectively improves the separability between corn and other crops, and enhances adaptability and robustness in complex terrain and planting structures.

[0046] (3) Overcoming the dependence and timeliness limitations of traditional TWDTW agricultural statistical data threshold setting: The present invention quantifies the time-series-spectral joint similarity between the pixels to be classified and the standard curve through DTW distance calculation, and constructs a corn recognition model based on minimum distance classification based on the comprehensive feature distance. The present invention does not require prior statistical area data as a constraint, avoiding the uncertainty caused by subjective threshold setting. It effectively suppresses salt and pepper noise and improves the homogeneity and spatial continuity within the image patch. The identified corn plot boundaries are clearer and more accurate. Since statistical area data is not required, the lag in obtaining statistical area is avoided.

[0047] (4) The present invention is an organic combination of various steps: the existing technology needs to determine the TWDTW threshold in combination with the corn planting area. Due to the small sample size and unreasonable selection of classification features, the TWDTW threshold may be difficult to determine and inaccurate. To solve this problem, the present invention combines the morphological sample point screening of key growth periods and the classification feature selection strategy of importance + correlation, making the similarity matching scheme based on minimum distance possible. Through the systematic optimization of the input sample quality and the spectral-temporal feature set, not only the accuracy and robustness of corn recognition are significantly improved, but also the dependence of traditional methods on statistical prior knowledge and subjective thresholds is overcome. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1This is a flow chart of the corn identification method for corn planting areas based on spatiotemporal feature selection according to the present invention.

[0049] Figure 2(a) shows the extraction of the key growth period of maize phenology in Region 1.

[0050] Figure 2(b) shows the extraction of the key growth period of maize phenology in region 2.

[0051] Figure 2(c) shows the extraction of the key growth period of maize phenology in region 3.

[0052] Figure 3(a) shows the separability graph of different features in region 1.

[0053] Figure 3(b) shows the separability graph of different features in region 2.

[0054] Figure 3(c) shows the separability graph of different features in region 3.

[0055] Figure 4(a) shows the correlation diagram between the time series in region 1.

[0056] Figure 4(b) shows the correlation diagram between the time series in region 2.

[0057] Figure 4(c) shows the correlation diagram between the time series in region 3.

[0058] Figure 5(a) shows the corn recognition result in area 1.

[0059] Figure 5(b) shows the corn recognition result in area 2.

[0060] Figure 5(c) shows the corn recognition result in area 3. DETAILED DESCRIPTION

[0061] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0062] The present invention provides a corn identification method for corn planting areas based on spatiotemporal feature selection. Figure 1 As shown, the method includes the following steps:

[0063] Step 1: Obtain remote sensing images of corn-growing areas.

[0064] In this step, two situations may occur. One is to obtain a remote sensing image of the study area and know the location of the corn planting area that year, that is, to know which pixels in the remote sensing image are corn pixels and which are non-corn pixels. Then, the data of the corn planting area in the measured remote sensing image can be directly used.

[0065] In the second case, if the current year's corn-growing areas in the study area are unknown, this embodiment uses historical data-based prediction. Using an existing national corn dataset, we determine the corn-growing areas in the study area for three consecutive years. This is the union of the three most recent years' corn-growing areas, which we use as the predicted corn-growing areas for the current year. The predicted locations of the corn-growing areas for the current year are then extracted from the measured remote sensing imagery for subsequent processing steps.

[0066] In a preferred solution, in order to eliminate the impact of non-arable land pixels on crop classification surveys and research, the cultivated land distribution provided by the historical China 30-meter Annual Land Cover Dataset (CLCD) can be used as a mask file to mask the remote sensing image to obtain the overall distribution data of farmland vegetation, thereby excluding non-arable land pixels and conducting subsequent steps on this basis.

[0067] In practice, acquired remote sensing images may have issues with noise and resolution. Therefore, preprocessing can be performed. First, remove the cloud from the remote sensing images, retaining images with cloud cover below 20%. Next, resample the image bands, resampling bands with a spatial resolution of 20 meters to 10 meters to achieve a uniform spatial resolution. Next, perform linear interpolation on the images using a 10-day time window. Finally, use the Savitzky-Golay filter for image smoothing.

[0068] In this specific embodiment, the remote sensing data may be Sentinel-2 images acquired in Google Earth Engine.

[0069] Step 2: Determine the corn sample points.

[0070] This method uses phenological curve similarity (vegetation characteristic time-series curves) for classification. However, insufficient samples can easily lead to bias in the extraction of statistical features of curve morphology, which in turn widens the confidence interval of the results and ultimately affects classification accuracy. Therefore, by selecting sample points based on morphological characteristics, a more representative standard growth curve for corn can be constructed, providing a more reliable basis for subsequent TWDTW matching.

[0071] Therefore, this step determines the NDVI characteristic curve of each pixel in the corn-growing area based on the remote sensing image of the corn-growing area; and screens out corn pixels that meet the planting characteristics of the key growth period according to the NDVI characteristic curve as corn sample points.

[0072] The present invention selects corn sample points from two aspects: one is the time selection of the key growth period; the other is the limitation of the key phenological fluctuation range that shows the phenological spectral characteristics; through the limitation of time and phenological spectral characteristics, corn pixels that meet the planting characteristics of the key growth period are selected as standard sample points.

[0073] Among them, the method for determining the critical growth period is: based on the remote sensing image of the corn planting area, calculate the NDVI characteristic curve of each corn pixel; average the data at the same time point of each NDVI characteristic curve to obtain the NDVI characteristic mean curve; obtain the amplitude change f between the maximum and minimum values ​​in the NDVI characteristic mean curve, and determine the time point t1 when the amplitude of the NDVI characteristic curve begins to be greater than 20%f, and the time point t2 when the amplitude begins to be less than 20%f. The interval [t1, t2] is the critical growth period.

[0074] When selecting and classifying classification features in the following steps, the preferred solution is to intercept the vegetation characteristic time series curves of corn sample points and pixels to be classified using the key growth period to ensure that the characteristic curves used for feature selection and similarity comparison can accurately represent the planting characteristics of the key growth period.

[0075] Among them, the method for determining the key phenological fluctuation range is: calculate the mean μ and standard deviation σ of the data at the same time point of each NDVI characteristic curve. The mean μ and standard deviation σ are both curves. [μ-σ, μ+σ] is taken as the allowable range of fluctuation of the NDVI characteristic curve. The mean curve and the standard deviation curve are superimposed and calculated in the form of μ-σ and μ+σ to obtain two curves. The curve between the two curves is the allowable range of fluctuation of the NDVI characteristic curve, that is, the key phenological fluctuation range.

[0076] When screening corn sample points, the NDVI characteristic curve of each pixel in the corn planting area is used to intercept the NDVI characteristic curve segment of the key growth period using the key growth period; for each pixel, it is determined whether there is a part of the NDVI characteristic curve segment that exceeds the key phenological fluctuation range. If there is, the pixel is removed and not identified as a corn sample point. Through this operation, only the NDVI characteristic curve of the key growth period with a fluctuation amplitude within the key phenological fluctuation range is retained, and the corresponding pixel is the corn sample point.

[0077] Step 3: Classification feature selection.

[0078] In this step, M species (M < N) are selected from the specified N vegetation features as classification features. By constructing a Relief filtering feature importance evaluation method, the importance of different features for corn recognition is calculated, redundant features are removed by comprehensively considering feature correlation, and a corn recognition feature set that takes into account both feature importance and independence is constructed. This solution significantly reduces the common misclassification phenomenon in the single NDVI method, effectively improves the separability between corn and other summer crops, and enhances the adaptability and robustness under complex terrain and planting structures.

[0079] In this step, for corn sample points and non-corn sample points, the temporal curves of vegetation features of N feature categories are calculated; based on the temporal curves of vegetation features of corn sample points and non-corn sample points, the Relief algorithm is used to determine the importance of different vegetation features for corn recognition; based on the temporal curves of vegetation features of corn sample points, the correlation between vegetation features is determined; a set of M vegetation features with the highest importance and mutual correlation not higher than the set correlation threshold are selected as classification features.

[0080] Based on the selected classification features, the temporal curves of vegetation features of the M classification features of each corn sample point and each non-corn sample point are obtained to form a sample set.

[0081] In one example, M can be selected as M = 5.

[0082] In a preferred solution, the specific implementation solution of this step 3 includes the following sub-steps:

[0083] Step 301: For corn sample points and non-corn sample points, spectral features are calculated, and then 34 vegetation features are calculated. The present invention selects a total of 34 vegetation index features and red-edge vegetation index features. Table 1 shows the calculation methods of the 34 vegetation features selected in the embodiments of the present invention.

[0084] Table 1 Spectral indices used

[0085]

[0086] Step 302: For corn sample points and non-corn sample points, the Relief algorithm is used to calculate the feature importance between sample points, and the classification importance of the selected 34 features for corn and other crops at the key growth stages is obtained.

[0087] Among them, the calculation formula of the Relief algorithm is:

[0088] In the Relief model, any two sample points (corn sample point) and (non-corn sample point) on the vegetation feature The distance representation on, represents the Vegetation features:

[0089]

[0090] Where, yes exist The value on yes exist The value on Indicates The maximum value on Indicates The minimum value on . The iterative calculation formula of Relief feature importance is expressed as:

[0091]

[0092] In the formula, NH and NM represent the nearest neighbor samples of the same type and the nearest neighbor samples of different types, respectively. Indicates that x and NH are The distance on Indicates that x and NM are in f i The distance on express The iterative result of feature weights.

[0093] After the Relief algorithm, each vegetation feature has a feature weight, which represents the feature importance, and the features are sorted according to the feature weight.

[0094] Step 303: For the corn sample points, the correlation between vegetation features is determined based on the vegetation feature time series curve.

[0095] The present invention adds a screening perspective of feature correlation, increases the independence of the screened classification features, solves the information redundancy problem that often exists when using multi-source features for classification, helps to construct a corn identification feature set that takes into account both feature importance and independence, effectively improves the separability between corn and other crops, and improves the accuracy and efficiency of identification.

[0096] Specifically, this step calculates the vegetation characteristic time series mean curve for all corn sample points for each vegetation characteristic. For each of the N vegetation characteristic time series mean curves, the Pearson correlation coefficient is calculated pairwise to represent the correlation between the vegetation characteristics. In this example, the Pearson correlation coefficients for 34 characteristics are calculated, resulting in a 34×34 Pearson correlation coefficient matrix.

[0097] Step 304: Select a group of M vegetation features with the highest importance and mutual correlation not higher than a set threshold as classification features.

[0098] This step can be specifically as follows: 34 vegetation features are sorted by importance, and selected one by one in descending order of importance; each time a vegetation feature is selected, whether the Pearson correlation coefficient between it and the determined classification feature is greater than the set correlation threshold; if it is greater, it is determined that the correlation is too large, and the currently selected vegetation feature is not used as a classification feature; if it is less than or equal to the set correlation threshold, the currently selected vegetation feature is determined as a classification feature; until 5 classification features are selected.

[0099] In a preferred embodiment, the correlation threshold may be selected as 0.9.

[0100] Step 305: Obtain vegetation characteristic time series curves for the five classification features of each corn sample point, as well as vegetation characteristic time series curves for the five classification features of each non-corn sample point, to form a multi-dimensional feature time series of the samples. These together constitute a standard sample, which is stored in a sample set to form a reference library. This reference library provides a standard set of characteristic time variation patterns for both corn and non-corn categories.

[0101] Step 4: Classify the pixels to be classified.

[0102] The two differences in this step are that, first, the comprehensive TWDTW distance is used to express similarity, and second, the traditional solution of using the set TWDTW threshold and corn planting area to jointly determine the pixel classification is abandoned, avoiding the use of the corn planting area that needs to be obtained with a lag from the yearbook data, and solving the problem of inaccurate TWDTW threshold determined in combination with the corn planting area. The existing technology needs to determine the TWDTW threshold in combination with the corn planting area because the small sample size and unreasonable feature selection lead to the difficulty and inaccuracy of the TWDTW threshold. The present invention combines the morphological sample point screening of the key growth period and the classification feature selection strategy of importance + correlation to make the similarity matching scheme based on the minimum distance possible.

[0103] In this step, the temporal-spectral joint similarity between the pixel to be classified and the corn sample points and non-corn sample points in the sample set is calculated; the temporal-spectral joint similarity between the two pixels is expressed using the comprehensive TWDTW distance of the vegetation feature time series curves of M classification features; the pixel to be classified is classified into the category corresponding to the minimum temporal-spectral joint similarity.

[0104] This step uses the Time-Weighted Dynamic Time Warping (TWDTW) algorithm, which is based on logistic function weighting and demonstrates significant advantages in crop classification accuracy compared to linear weighting methods. In the TWDTW method, the logistic weights for the steepness α and the midpoint β are determined.

[0105]

[0106] in is the date in the time series and The time elapsed in the days between them, α and β represent the logistic weights of the steepness and midpoint, responsible for shape matching and time alignment, respectively. The logistic weighting function used in this step constructs a nonlinear time penalty mechanism by setting α = -0.1 and β = 50, giving a lower penalty for time warps less than 50 days and a higher penalty for larger time warps.

[0107] The crop classification process based on TWDTW includes the following steps:

[0108] (1) Calculate the multidimensional weighted dynamic time warping distance.

[0109] For the pixels to be classified and the sample points of each category, the TWDTW distance of each classification feature is calculated separately, that is, the similarity of the characteristic curve under the classification feature is calculated; the TWDTW distances of the five classification features are added or weighted to express the comprehensive TWDTW distance, that is, the multidimensional weighted dynamic time warping distance.

[0110] Among them, each category of sample points includes corn and non-corn sample points.

[0111] For each remote sensing image pixel to be classified within the study area, the same method used to construct the reference library was used to extract the pixel's multidimensional feature time series across all selected classification features. For each multidimensional feature time series of the pixel to be classified, the weighted dynamic warping distance between it and the maize and non-maize categories in the reference library was calculated to obtain a set of maize and non-maize distance images.

[0112] (2) Category determination is performed based on the minimum distance principle.

[0113] For each pixel in the study area, the TWDTW distance values ​​to the corn sample point and the non-corn sample point are compared, and the pixel is assigned to the category with the closest distance, that is, the category corresponding to the minimum TWDTW distance value.

[0114] The effects of the present invention are verified below.

[0115] To evaluate the effectiveness of the research method, this example uses overall accuracy (OA), Kappa coefficient, producer accuracy (PA), and user accuracy (UA) indicators for accuracy verification.

[0116] Three typical corn-growing regions with significant climate and topographical differences were selected as research subjects. The proposed method was used to screen samples from the experimental regions for characteristics during the critical growth period of corn in the experimental regions. The top five characteristics identified were then added to the classification model.

[0117] Figures 2(a), 2(b), and 2(c) are schematic diagrams of maize growth period acquisition in the three study areas. The green curve is a smoothed NDVI time series curve. Green dots indicate the start of the phenological period, and red dots indicate the end of the phenological period. Figures 3(a), 3(b), and 3(c) present the results of feature importance analysis for the three study areas, with darker colors indicating higher importance. Figures 4(a), 4(b), and 4(c) present feature correlation analysis, with darker colors indicating stronger correlations. Figures 5(a), 5(b), and 5(c) show maize identification results for the three study areas. A1-A3 represent three local areas in Region 1, B1-B3 represent three local areas in Region 2, and C1-C3 represent three local areas in Region 3. In the second row of figures, orange indicates maize-growing areas, and white indicates non-maize areas.

[0118] The results show that the classification accuracy of multiple feature combinations is generally better than that of a single feature. Due to differences in topographic conditions and crop planting structures across the study areas, the number and combination of features required to achieve optimal classification accuracy vary. The recognition results for different feature combinations are shown in Table 2.

[0119] Table 2 Corn mapping accuracy assessment

[0120]

[0121] The method of the present invention not only significantly improves the accuracy and robustness of corn identification through systematic optimization of input sample quality and spectral-temporal feature sets, but also overcomes the traditional method's reliance on statistical prior knowledge and subjective thresholds.

[0122] The above specific embodiments merely illustrate the design principles of the present invention. The shapes and names of the components described herein may vary and are not limiting. Therefore, those skilled in the art may modify or substitute equivalents for the technical solutions described in the above embodiments. Such modifications and substitutions, without departing from the inventive spirit and technical solutions of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A corn identification method for corn planting areas based on spatiotemporal feature selection, characterized in that: include: Step 1: Obtain remote sensing images of corn-growing areas; Step 2: Determine corn sample points: Based on the remote sensing image of the corn-growing area, determine the NDVI characteristic curve of each pixel in the corn-growing area; Corn pixels that meet the planting characteristics of the key growth period are selected according to the NDVI characteristic curve as corn sample points; Step 3: Classification feature selection: For corn sample points and non-corn sample points, calculate the vegetation feature time series curves of N feature categories; Based on the vegetation characteristic time series curves of corn sample points and non-corn sample points, the Relief algorithm is used to determine the importance of different vegetation characteristics for corn identification. Based on the vegetation characteristic time series curves of corn sample points, the correlation between vegetation characteristics is determined. A group of M vegetation characteristics with the highest importance and mutual correlation no higher than the set correlation threshold is selected as classification features. The vegetation characteristic time series curves of M classification characteristics of each corn sample point and non-corn sample point constitute the sample set; Step 4: Classification: Calculate the temporal-spectral joint similarity between the pixel to be classified and the corn sample points and non-corn sample points in the sample set; the temporal-spectral joint similarity between two pixels is expressed using the comprehensive TWDTW distance of the vegetation feature time series curves of M classification features; classify the pixel to be classified into the category corresponding to the minimum temporal-spectral joint similarity.

2. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 1, characterized in that: In step 1, the remote sensing image of the corn planting area is obtained as follows: When the location of corn planting areas in the current year in the study area is known, the data of corn planting areas in the measured remote sensing images are directly used; In the case that the location of the corn planting areas in the study area for the current year is unknown, the existing national corn dataset is used to determine the corn planting areas in the study area for the last three consecutive years as the predicted corn planting areas for the current year; the data of the corn planting areas for the current year predicted from the measured remote sensing images are used.

3. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 1, characterized in that: The method further includes preprocessing the acquired remote sensing image of the corn-growing area, including: Decloud remote sensing images and retain images with cloud coverage below a set ratio; Resample the bands of remote sensing images; Perform linear interpolation of remote sensing images based on a selected time window; Savitzky-Golay filtering is used to smooth remote sensing images.

4. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 1, 2 or 3, characterized in that: The method further includes: using the cultivated land distribution provided by the land cover dataset CLCD as a mask file to mask the remote sensing image and exclude non-cultivated land pixels in the remote sensing image.

5. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 1, characterized in that: In step 2, corn pixels that meet the planting characteristics of the key growth period are screened out according to the NDVI characteristic curve as corn sample points, including: Determine the critical growth period: Calculate the NDVI characteristic curve for each corn pixel based on remote sensing images of the corn-growing area. Average the data at the same time point for each NDVI characteristic curve to obtain the NDVI characteristic mean curve. Obtain the amplitude variation f between the maximum and minimum values ​​in the NDVI characteristic mean curve. The time interval between the time when the NDVI amplitude begins to exceed 20% of f and the time when it begins to fall below 20% of f is defined as the critical growth period. Determine the key phenological fluctuation range: calculate the mean μ and standard deviation σ of the data at the same time point of each NDVI characteristic curve, and use [μ-σ, μ+σ] as the allowable range of NDVI characteristic curve fluctuation, that is, the key phenological fluctuation range; For the NDVI characteristic curve of each pixel in the corn planting area, the NDVI characteristic curve segment of the key growth period is intercepted using the key growth period; for each pixel, it is determined whether the fluctuation amplitude of the intercepted NDVI characteristic curve segment is all within the key phenological fluctuation range. If so, the current pixel is identified as a corn sample point.

6. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 5, characterized in that: In the classification feature selection and classification, the key growth period is used to intercept the vegetation characteristic time series curves of corn sample points and pixels to be classified.

7. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 1, characterized in that: In step 3, the correlation between vegetation characteristics is determined based on the vegetation characteristic time series curve of the corn sample point: For each vegetation characteristic, the vegetation characteristic time series mean curve of all corn sample points was calculated; for the vegetation characteristic time series mean curves of N vegetation characteristics, the Pearson correlation coefficient was calculated pairwise as the correlation between vegetation characteristics.

8. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 7, characterized in that: In step 3, a group of M vegetation features with the highest importance and mutual correlation not higher than a set correlation threshold are selected as classification features: Using the importance, N vegetation features are sorted and selected one by one in descending order of importance. For each selected vegetation feature, it is determined whether the Pearson correlation coefficient between the selected vegetation feature and the determined classification feature is greater than a set correlation threshold. If so, it is determined that the correlation is too large and the currently selected vegetation feature is not used as a classification feature. If it is less than or equal to the set correlation threshold, the currently selected vegetation feature is determined as a classification feature; Until M classification features are selected.

9. The corn identification method for corn planting areas based on spatiotemporal feature selection according to claim 1, characterized in that: The method for obtaining the time series-spectrum joint similarity is: For the pixel to be classified and a sample point in the sample set, the TWDTW distance of each classification feature is calculated separately, and the TWDTW distances of M classification features are calculated comprehensively to express them as the comprehensive TWDTW distance.

10. The corn identification method based on spatiotemporal feature selection in corn planting areas according to claim 1, characterized in that: The N feature categories include vegetation index features and red edge vegetation index features: The vegetation index features include: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Difference Vegetation Index (DVI), Optimized Soil Adjusted Vegetation Index (OSAVI), Spectral Ratio Index (SR), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Greenness Index (NDGI), Green Leaf Cover Index (GCVI), Ratio Vegetation Index (RVI), Normalized Difference Pollution Index (NDPI), Land Surface Water Index (LSWI), Photochemical Vegetation Reflectance Index (PSRI); The red edge vegetation index features include: a first normalized red edge difference index NREDI1, a second normalized red edge difference index NREDI2, a third normalized red edge difference index NREDI3, a near infrared / red edge ratio index NIR / RE, a red band / red edge ratio index RED / RE, and a red edge position index REP.

Images