Similar crop classification system and method based on multi-temporal remote sensing and phenology fusion

Abstract

The application discloses a kind of similar crop classification systems and methods based on multi-temporal remote sensing and phenology fusion, belong to agricultural remote sensing monitoring and crop identification technical field. Including the following steps: S1, obtain the multi-temporal multispectral remote sensing image of whole growing season in study area and complete pretreatment and time sequence reconstruction;S2, extract crop phenology characteristics from the continuous vegetation index time sequence sequence obtained by pretreatment and analysis;S3, determine key phase based on phenology period and category separability analysis, extract corresponding spectral characteristics and index characteristics;S4, standardization fusion and optimization are carried out to spectral characteristics and phenology characteristics, and low-redundancy high-discriminative fusion feature space is constructed;S5, train supervised classifier based on fusion feature space, complete crop classification, precision evaluation and post-processing, and generate crop type distribution map. The system and method of the application can maintain robustness in cloud rain interference scene, and are suitable for large-area planting structure identification and business monitoring.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing monitoring and crop identification technology, and in particular to a classification system and method for similar crops based on multi-temporal remote sensing and phenological fusion. Background Technology

[0002] Crop planting structure identification is a core foundation for agricultural monitoring, food security assessment, agricultural insurance, and precision agriculture management. Remote sensing technology, with its advantages of wide coverage, high timeliness, and relatively low cost, has become the mainstream technology for large-scale crop identification.

[0003] Existing crop classification methods mainly include single-phase spectral classification, multi-phase spectral overlay, vegetation index time-series analysis, and machine learning classification, but they have obvious technical shortcomings: similar crops with similar spectral characteristics are prone to classification confusion when relying solely on spectral information; the selection of time phases lacks specificity, and blindly overlaying data from the entire growing season can easily lead to an explosion of feature dimensions, increasing computational costs and reducing the generalization ability of the classifier; severe weather such as clouds and rain can interfere with the quality of remote sensing images, resulting in a decrease in the accuracy of phenological feature extraction; and the degree of automation in the technical process is insufficient, making it difficult to adapt to the batch processing deployment requirements of large-scale operational monitoring. Summary of the Invention

[0004] The purpose of this invention is to provide a similar crop classification system and method based on multi-temporal remote sensing and phenological fusion, thereby solving the above-mentioned technical problems.

[0005] To achieve the above objectives, this invention provides a method for classifying similar crops based on multi-temporal remote sensing and phenological fusion, comprising the following steps: S1. Acquire multi-temporal, multispectral remote sensing images of the study area throughout the entire growing season and complete preprocessing and time-series reconstruction. S2. Extract and analyze crop phenological characteristics from the preprocessed continuous vegetation index time series; S3. Based on the separability analysis of phenological periods and categories, key time phases are identified, and corresponding spectral and index features are extracted. S4. Standardize, fuse, and optimize spectral and phenological features to construct a low-redundancy, high-discrimination fusion feature space. S5. Train a supervised classifier based on the fusion feature space to complete crop classification, accuracy evaluation and post-processing, and generate a crop type distribution map.

[0006] The system for classifying similar crops based on multi-temporal remote sensing and phenological fusion includes a data acquisition and preprocessing module, a temporal reconstruction module, a phenological feature extraction module, a key temporal phase selection module, a spectral feature extraction module, a feature fusion and optimization module, a classification decision module, an accuracy evaluation module, and a post-processing and mapping module.

[0007] Therefore, the present invention employs the above-mentioned method for classifying similar crops based on multi-temporal remote sensing and phenological fusion, which has the following beneficial effects: 1. Fully explore the differences in growth rhythms of different crops to efficiently distinguish similar crops.

[0008] 2. By combining phenological periods with separability analysis, key time phases are screened to reduce feature redundancy and improve computational efficiency.

[0009] 3. Temporal reconstruction combined with cloud removal processing improves the quality of phenological extraction and adapts to cloud and rain interference scenarios.

[0010] 4. Establish an end-to-end automated processing workflow to support large-scale operational monitoring and batch processing of crop planting structures.

[0011] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall technical solution of the present invention; Figure 2 This is a schematic diagram of the time-series purification and phenological parameter extraction of the present invention; Figure 3 This is a schematic diagram of the preferred key timing phase of the present invention; Figure 4 This is a schematic diagram of the spectral-phenological feature fusion and dimensionality reduction optimization of the present invention; Figure 5 This is a schematic diagram of the classification and accuracy evaluation process of the present invention; Figure 6 This is a schematic diagram illustrating the post-processing of the classification results and land parcel consistency in this invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0014] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0015] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0016] like Figures 1-6 As shown, the present invention provides a method for classifying similar crops based on multi-temporal remote sensing and phenological fusion, comprising the following steps: S1. Acquire multi-temporal, multispectral remote sensing images of the study area throughout the entire growing season and complete preprocessing and time-series reconstruction; S2. Extract and analyze crop phenological features from the preprocessed continuous vegetation index time-series; S3. Determine key temporal phases based on the separability analysis of phenological periods and categories, and extract corresponding spectral and index features; S4. Standardize, fuse, and optimize the spectral and phenological features to construct a low-redundancy, high-discrimination fusion feature space; S5. Train a supervised classifier based on the fusion feature space, complete crop classification, accuracy evaluation, and post-processing, and generate a crop type distribution map.

[0017] S1 specifically includes: S11, acquiring Sentinel-2 or Landsat series multispectral remote sensing images covering the study area for 6-12 months, with a temporal resolution of 5-16 days; performing preprocessing operations such as radiometric or atmospheric correction and geometric registration on the remote sensing images; and using cloud detection algorithms (such as Fmask, Sentinel-2 SCL band cloud detection, etc.) to identify anomalous pixels such as clouds, cloud shadows, and snow, and constructing cloud / cloud shadow / snow masks. For example, in the scenario of similar crop classification based on Sentinel-2 full growing season images, acquiring Sentinel-2 multi-temporal images covering the entire growing season of the study area, selecting blue, green, red, near-infrared, and shortwave infrared bands, and performing atmospheric correction and geometric registration on the images; S12, calculating based on the preprocessed remote sensing images. and Two vegetation indices were used to obtain a continuous time series of vegetation indices, where: ; ; in, For near-infrared reflectivity, For red light band reflectivity, S13. Using a combination of Savitzky-Golay filtering and dual-logic function fitting, the continuous vegetation index time series is reconstructed to obtain a continuous and reliable time series dataset. The Savitzky-Golay filtering satisfies the following conditions: ; in, For the current moment, for Smoothing after time-major filtering value, These are the Savitzky-Golay filter coefficients. This is the time offset. The width of the filter window. for Original moments value; The fit of the two logic functions satisfies: ; in, for After time fitting value, This is the baseline value of the vegetation index. This represents the vegetation index saturation value. For the rapid growth period of vegetation, The rate of vegetation decay. This refers to the time corresponding to the inflection point of the rapid growth period. This refers to the time corresponding to the inflection point of the decay period. It is a natural constant.

[0018] S2 is the core step in extracting crop phenological characteristics, specifically including: S21, using the annual maximum... Using a threshold of 10%-30%, the Start of the Growing Season (SOS) and End of the Growing Season (EOS) were extracted from the continuous vegetation index time-series data using the threshold method. The Length of the Growing Season (LOS) was also calculated. The Rapid Upward Point (RUP) and Rapid Downward Point (RDN) of the rapid growth period were extracted using the first-order difference method. Simultaneously, the Peak of the Growing Season (POS) was extracted. ; ; ; in, for The first derivative of the time series curve, argmax represents the time node corresponding to the maximum value of the first derivative, and argmin represents the time node corresponding to the minimum value of the first derivative; S22, calculate the peak value, baseline, growth amplitude, rise / fall rate, curve integral and asymmetry morphological parameters, and combine them with the basic phenological features extracted in S21 to form a 10 to 15-dimensional phenological feature vector. For the planting pattern of multiple crops or replanting in the study area, multi-peak detection (such as peak detection method, sliding window method, etc.) is also required to extract the position, height and width parameters of the multi-peaks to adapt to the needs of crop phenological feature analysis under the replanting pattern.

[0019] S3 specifically includes: S31, defining key temporal windows based on phenological periods, introducing Jeffries-Matusita distance to conduct category separability analysis, selecting the 5-8 temporal windows with the largest Jeffries-Matusita distance to form a key temporal set, avoiding blind overlay of full-temporal data; S32, extracting multi-band reflectance, vegetation index, and texture features from remote sensing images of key temporal phases as spectral and index features, and constructing multi-temporal spectral feature vectors based on the above features, wherein the vegetation index includes... , , , The formula for adjusting the vegetation index for soil is: ; in, The soil adjustment coefficient ranges from 0 to 1 and can be flexibly adjusted according to the soil type of the study area. In addition, key time phases include ±15 days before and after SOS, RUP, POS, RDN, and ±15 days before and after EOS. A ±15-day window is established with the key time phases as the center to effectively distinguish the spectral characteristics of similar crops.

[0020] S4 is the core step in the fusion and optimization of spectral and phenological features, specifically including: S41, performing normalization or Z-score standardization on the spectral and phenological features, and completing the feature fusion using any one of the following methods: direct stitching, weighted fusion, or hierarchical fusion. Direct stitching satisfies the following conditions: ; in, The fused feature vector For spectral eigenvectors, This is a phenological feature vector; Weighted fusion satisfies: ; in, These are the spectral feature weighting coefficients. The weighting coefficient for phenological characteristics. , Assign values ​​based on feature importance, and S42. Weighted fusion is performed by setting weights based on the importance of random forest features; Recursive Feature Elimination (RFE) is used for feature selection, or Principal Component Analysis (PCA) or Relief-F algorithm can be used for feature selection. One or a combination of the above methods is selected. For PCA, principal components with a cumulative contribution rate of not less than 80% are retained. For RFE and Relief-F algorithms, Top-K important features are selected to reduce feature redundancy and noise, and ensure high discriminability of the fused feature space, thereby constructing a low-redundancy, high-discriminability fused feature space.

[0021] S5 specifically includes: S51, constructing training samples (such as random sampling, stratified sampling, etc.) and training a supervised classifier using random forest, support vector machine, or gradient boosting tree. Hyperparameter optimization of the classifier is achieved through cross-validation to realize accurate identification of similar crops. Specifically, samples from each crop category are collected to construct a training sample set, which is then divided into a training set and a validation set in a 7:3 ratio. S52, selecting confusion matrix, overall accuracy (OA), Kappa coefficient, producer accuracy (PA), user accuracy (UA), and F1-score as accuracy evaluation metrics to conduct classification accuracy assessment. S53, employing spatial filtering (such as median filtering, Gaussian filtering, etc.), morphological denoising (such as opening and closing operations, etc.), and rule constraints (such as plot area thresholds, neighborhood similarity rules, boundary smoothing rules, etc.) for post-processing to suppress salt-and-pepper noise and improve plot scale. Figure 1 Consistency, generating crop type distribution maps.

[0022] A system for classifying similar crops based on multi-temporal remote sensing and phenological fusion is proposed. This system enables end-to-end automated processing, enhances robustness under cloud and rain interference scenarios, and supports batch deployment for large-area crop planting structure identification and operational monitoring. The system includes modules for data acquisition and preprocessing, temporal reconstruction, phenological feature extraction, key temporal phase selection, spectral feature extraction, feature fusion and optimization, classification decision-making, accuracy evaluation, and post-processing and mapping. The data acquisition and preprocessing module acquires multispectral remote sensing images throughout the growing season and performs preprocessing... The system includes the following modules: a time-series reconstruction module for filtering and fitting vegetation index sequences for reconstruction; a phenological feature extraction module for extracting phenological features and morphological parameters; a key time phase selection module for selecting a set of key time phases; a spectral feature extraction module for extracting spectral, index, and texture features of key time phases; a feature fusion and optimization module for constructing a low-redundancy fusion feature space; a classification decision module for training a classifier and outputting preliminary crop classification results; a precision evaluation module for evaluating classification precision using multiple indicators and verifying the effect; and a post-processing and mapping module for post-processing denoising and rule constraints to generate crop type distribution maps.

[0023] The similar crop classification system based on multi-temporal remote sensing and phenological fusion of the present invention is implemented on a computer-readable storage medium, on which a computer program is stored, and the collaborative work of each module is realized by running the computer program.

[0024] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for classifying similar crops based on multi-temporal remote sensing and phenological fusion, characterized in that, Includes the following steps: S1. Acquire multi-temporal, multispectral remote sensing images of the study area throughout the entire growing season and complete preprocessing and time-series reconstruction. S2. Extract and analyze crop phenological characteristics from the preprocessed continuous vegetation index time series; S3. Based on the separability analysis of phenological periods and categories, key time phases are identified, and corresponding spectral and index features are extracted. S4. Standardize, fuse, and optimize spectral and phenological features to construct a low-redundancy, high-discrimination fusion feature space. S5. Train a supervised classifier based on the fusion feature space to complete crop classification, accuracy evaluation and post-processing, and generate a crop type distribution map.

2. The method for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 1, characterized in that: S1 specifically includes: S11. Acquire Sentinel-2 or Landsat series multispectral remote sensing images covering the study area for 6-12 months, with a temporal resolution of 5-16 days. Perform preprocessing operations such as radiometric or atmospheric correction and geometric registration on the remote sensing images, and use cloud detection algorithms to identify anomalous pixels and construct cloud / cloud shadow / snow masks. S12, Calculation based on preprocessed remote sensing images and Two vegetation indices were used to obtain a continuous time series of vegetation indices, where: ； ； in, For near-infrared reflectivity, For red light band reflectivity, Reflectivity in the blue light band; S13. A time-series reconstruction of continuous vegetation index time series is performed using a combination of Savitzky-Golay filtering and dual-logic function fitting. The Savitzky-Golay filtering satisfies the following: ； in, For the current moment, for Smoothing after time-major filtering value, These are the Savitzky-Golay filter coefficients. This is the time offset. The width of the filter window. for Original moments value; The fit of the two logic functions satisfies: ； in, for After time fitting value, This is the baseline value of the vegetation index. This represents the vegetation index saturation value. For the rapid growth period of vegetation, The rate of vegetation decay. This refers to the time corresponding to the inflection point of the rapid growth period. This refers to the time corresponding to the inflection point of the decay period. It is a natural constant.

3. The method for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 2, characterized in that: S2 specifically includes: S21, Using annual maximum Using a threshold of 10%-30%, SOS and EOS are extracted from continuous vegetation index time-series data using the threshold method, and LOS is calculated. RUP and RDN are extracted using the first-order difference method, and POS is also extracted. ； ； ； in, for The first derivative of the time series curve, argmax represents the time node corresponding to the maximum value of the first derivative, and argmin represents the time node corresponding to the minimum value of the first derivative. S22. Calculate the morphological parameters of peak value, baseline, growth amplitude, rising / falling rate, curve integral and asymmetry, and combine them with the basic phenological features extracted in S21 to form a 10 to 15-dimensional phenological feature vector.

4. The method for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 3, characterized in that: S3 specifically includes: S31. Based on the phenological period, key time phase windows are delineated, and Jeffries-Matusita distance is introduced to conduct category separability analysis. The time phase combination with the largest JM distance is selected as the key time phase set. S32. Extract multi-band reflectance, vegetation index, and texture features from key temporal remote sensing images as spectral and index features, where the vegetation index includes... , , , The formula for adjusting the vegetation index for soil is: ； in, This is the soil adjustment coefficient, with a value ranging from 0 to 1.

5. The method for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 4, characterized in that: Key time phases include ±15 days before and after SOS, the RUP period, the POS period, the RDN period, and ±15 days before and after EOS.

6. The method for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 5, characterized in that: S4 specifically includes: S41. Perform normalization or Z-score standardization on the spectral and phenological features, and complete the feature fusion using any one of the following methods: direct stitching, weighted fusion, or hierarchical fusion. Direct stitching satisfies the following condition: ； in, The fused feature vector For spectral eigenvectors, This is a phenological feature vector; Weighted fusion satisfies: ； in, These are the spectral feature weighting coefficients. is the weighting coefficient for phenological characteristics, and ; S42. Feature optimization can be performed using RFE, or PCA or Relief-F algorithm can be used. One or a combination of the above methods can be used. For PCA, the principal components with a cumulative contribution rate of not less than 80% are retained. For RFE and Relief-F algorithms, the Top-K important features are selected to construct a low-redundancy, high-discrimination fusion feature space.

7. The method for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 6, characterized in that: S5 specifically includes: S51. Construct training samples and train a supervised classifier using random forest, support vector machine, or gradient boosting tree. Optimize the hyperparameters of the classifier through cross-validation. S52. Select confusion matrix, OA, Kappa coefficient, PA, UA and F1-score as accuracy evaluation indicators to conduct classification accuracy evaluation; S53. Post-processing is performed using spatial filtering, morphological denoising, and rule constraints to suppress salt-and-pepper noise and improve the consistency of plot-scale mapping, ultimately generating a crop type distribution map.

8. The system for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to any one of claims 1-7, characterized in that: It includes modules for data acquisition and preprocessing, time series reconstruction, phenological feature extraction, key time phase selection, spectral feature extraction, feature fusion and optimization, classification decision, accuracy evaluation, and post-processing and mapping.

9. The system for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 8, characterized in that: The data acquisition and preprocessing module is used to acquire multispectral remote sensing images of the entire growing season and perform preprocessing operations; The time-series reconstruction module is used for filtering, fitting, and reconstructing vegetation index sequences; The phenological feature extraction module is used to extract phenological features and morphological parameters; The key time phase selection module is used to select the set of key time phases; The spectral feature extraction module is used to extract spectral, index, and texture features of key time phases; The feature fusion and optimization module is used to construct a low-redundancy fusion feature space; The classification decision module is used to train the classifier and output preliminary crop classification results; The accuracy evaluation module is used to evaluate classification accuracy using multiple indicators and to verify the results. The post-processing and mapping module is used for post-processing noise reduction and rule constraints to generate crop type distribution maps.

10. The system for classifying similar crops based on multi-temporal remote sensing and phenological fusion according to claim 9, characterized in that: The system is implemented based on a computer-readable storage medium, on which a computer program is stored. The collaborative operation of each module is achieved by running the computer program.

Images