Abandoned farmland identification method based on multi-source remote sensing data and time sequence correction

Abstract

The invention discloses a farmland abandoned land identification method based on multi-source remote sensing data and time sequence correction. The method comprises the following steps: 1, data acquisition and preprocessing; 2, extracting a key phenological period; 3, constructing a feature library; 4, training and predicting a random forest model; and 5, post-processing and precision evaluation are carried out. The method comprises the following steps: firstly, integrating multi-source remote sensing image data, and strictly preprocessing the multi-source remote sensing image data to ensure the time-space consistency of the data; and accurately extracting key phenological period information by deeply analyzing the minimum vegetative period and the vegetative peak period of the vegetation through NDVI and NDSI time sequences. In the feature construction stage, multiple spectral indexes are fused, principal component analysis and feature importance screening are applied, feature dimensions are optimized, and model performance is improved. A random forest model is adopted for training and prediction, and a high-precision abandoned land spatial distribution diagram is finally generated by combining an oversampling technology and a category weight adjustment technology.

Description

Technical Field

[0001] The present invention relates to the technology in the field of cultivated land abandonment identification, and in particular to a cultivated land abandonment identification method based on multi-source remote sensing data and time series correction. Background Art

[0002] As a core link in ensuring food security and ecological balance, the technological evolution of cultivated land abandonment monitoring has always faced multi-dimensional challenges. Traditional monitoring methods highly rely on manual statistics and field surveys, which are not only costly and time-limited, but also difficult to capture the spatio-temporal heterogeneity of cultivated land use. With the popularization of remote sensing technology, monitoring methods based on satellite images have gradually become the mainstream, but their application effectiveness is restricted by multiple technical bottlenecks. One is that existing research mostly relies on medium and low spatial resolution data. In hilly and mountainous areas where fragmented cultivated land is widely distributed, the mixed pixel effect leads to a significant reduction in the recognition accuracy of small-scale abandoned land plots. The second is that although traditional time series analysis methods can characterize the dynamics of land surface cover, in cloudy and rainy climate regions, the problem of cumulative classification errors caused by the lack of data continuity has not been effectively solved. The third is that the existing algorithms still lack sufficient exploration of phenological characteristics and fail to fully integrate the spectral-temporal response differences in key phenological periods, restricting the discrimination ability between abandoned land and fallow land. Fourth, the classification model driven by a single data source has poor generalization in complex land surface cover scenarios, especially prone to misjudgment in the vegetation succession transition zone.

[0003] The current remote sensing monitoring technology for cultivated land abandonment mainly focuses on the collaborative optimization of multi-source remote sensing data and algorithms. The specific technical paths can be divided into the following categories: 1. Spectral analysis method based on phenological characteristics. By extracting the difference features of NDVI and NDSI in key phenological periods and combining with time series thresholds (such as NDVI continuously being lower than 0.2), abandoned land is determined. Such methods perform well in plain continuous cultivated areas, but are limited by the mixed pixel effect of medium and low resolution images, resulting in a high recognition error rate for fragmented plots.

[0004] 2. Random forest classification model. By integrating multi-spectral bands, vegetation indices (EVI, LSWI) and texture features, the classification accuracy of complex land surface cover is significantly improved, but its dependence on the continuity of optical images limits its applicability in cloudy and rainy regions.

[0005] 3. Multi-source data fusion technology. By combining optical remote sensing data and radar data, the advantage of radar penetrating clouds is used to make up for the lack of optical data, but the low spatial resolution of radar data limits the refined identification of small-scale abandoned land.

[0006] 4. Time series correction algorithm. Isolated mutated pixels are removed through a multi-year long time series sliding window, and the results are optimized by combining prior rules (such as continuous fallow for ≥2 years). However, the static correction rules are difficult to adapt to the annual dynamic changes of phenological characteristics, resulting in long-term error accumulation.

[0007] Disadvantages of the above existing technologies: 1. Resolution and mixed pixel problems. It is difficult to accurately identify small-scale fragmented fallow plots commonly existing in hilly areas of southern China based on medium and low-resolution data such as MODIS data or Landsat, resulting in significant underestimation of the classified area.

[0008] 2. Data loss under interference. The effective observation frequency of traditional optical images in cloudy and rainy areas is insufficient, resulting in broken time series, and further causing abnormal fluctuations in vegetation indices such as NDVI / EVI, misjudging the data loss period as fallow.

[0009] 3. Cumulative effect. Existing time series correction algorithms only rely on static rules and do not consider the annual dynamic changes of phenological characteristics, resulting in the gradual amplification of classification errors in long-term monitoring.

[0010] 4. Insufficient generalization ability. Classification models based on a single data source, such as optical remote sensing or radar data, are difficult to handle complex surface cover scenarios (such as the succession from fallow land to shrub forest), and their classification accuracy drops significantly in the vegetation transition zone.

[0011] The above technical bottlenecks seriously restrict the accuracy, timeliness and large-scale promotion potential of cultivated land fallow monitoring. Summary of the Invention

[0012] In view of this, in view of the deficiencies of the existing technology, the main purpose of the present invention is to provide a cultivated land fallow identification method based on multi-source remote sensing data and time series correction, which aims to achieve efficient and accurate monitoring of the cultivated land fallow phenomenon through systematic data processing and analysis techniques.

[0013] To achieve the above purpose, the present invention adopts the following technical solutions: A cultivated land fallow identification method based on multi-source remote sensing data and time series correction, comprising the following steps: Step 1: Data acquisition and preprocessing, the specific operations are as follows: 1.1. Remote sensing data acquisition: Select multi-source remote sensing image data covering the growth season data of the target area; 1.2. Cloud masking and atmospheric correction: When using optical remote sensing data, first remove cloud interference through the cloud masking algorithm to ensure the clarity of the image; then, perform atmospheric correction to eliminate the influence of atmospheric scattering and absorption on the image and ensure the spectral consistency of the data; 1.3. Geometric registration and accuracy correction: Geometric registration is performed on all images to ensure spatial alignment between different images and guarantee spatio-temporal consistency. In addition, radiometric correction is also carried out on the images to eliminate sensor differences and atmospheric effects and ensure data accuracy. 1.4. Radar image processing: Radar data is denoised using the Refined Lee filter, followed by radiometric calibration and terrain correction. The backscatter coefficient is extracted from the processed radar images, which helps analyze different land cover types. 1.5. Data integration and spatio-temporal consistency: By fusing optical images and radar images, spatio-temporal consistency of different source data is ensured, providing reliable input data for subsequent analysis. Step 2: Extraction of key phenological periods: Based on the NDVI and NDSI time series, two sensitive periods, the minimum vegetative period and the peak vegetative period, are extracted through filtering, smoothing, and extreme value detection algorithms. The phenological period positioning error is controlled within 5 days to ensure the accuracy of time series analysis. Step 3: Construction of the feature library: Multisource remote sensing data and spectral indices are fused to construct a multi-dimensional feature set. Through principal component analysis and feature importance screening, the feature dimensions are optimized to improve the discrimination efficiency of the classification model. Step 4: Training and prediction of the random forest model: The random forest algorithm is used, and the multi-source feature set is input. The model parameters are optimized through cross-validation. For the problem of sample imbalance, oversampling and class weight adjustment are applied. The model outputs a probability map of abandoned land, and through threshold segmentation, a binary classification result is generated to achieve the mapping of the spatial distribution of abandoned land at the regional scale. Step 5: Post-processing and accuracy evaluation: Spatial filtering is performed on the classification results to remove noise, generating a refined abandoned land distribution map. Based on an independent test set, a confusion matrix is calculated to evaluate the overall accuracy, Kappa coefficient, and class-specificity indicators, analyze the main sources of misclassification, and propose time series persistence rules to optimize the results. The final results meet the requirements of high-resolution monitoring and support decision-making applications.

[0014] Preferably, in step 1.1, Landsat satellite images are used for remote sensing data acquisition, and multi-temporal optical remote sensing image data of Landsat-5, Landsat-7, and Landsat-8, Sentinel-2 optical images, and Sentinel-1 radar image data are obtained from Google Earth Engine.

[0015] Preferably, the specific operations of step 2 are as follows: 2.1. NDVI and NDSI time series analysis: NDVI and NDSI time series data are used, and key phenological periods are extracted through extreme value detection algorithms. NDVI reflects the growth of vegetation, and NDSI is highly sensitive to bare soil. 2.2. Extraction of sensitive periods: By analyzing the changes in NDVI and NDSI, the minimum vegetative period and the peak vegetative period are extracted; LVP is usually the stage with the least vegetation cover and significant bare soil, while PVP is the stage with saturated vegetation cover, which can reflect the maximum vegetation growth state of cultivated land; 2.3. Error control: When extracting phenological periods, smoothing filtering and extreme value detection algorithms are used to ensure that the phenological period positioning error is controlled within 5 days, guaranteeing the accuracy of time series analysis; 2.4. Phenological period correction: To accurately calibrate LVP and PVP, corrections are made according to the growth cycles of different crops to ensure that the extracted phenological periods adapt to the changes under different regional and climatic conditions.

[0016] Preferably, the specific operations of step 3 are as follows: 3.1. Fusion of spectral indices and remote sensing data: Combining multiple spectral indices such as NDVI, EVI, and BSI, and different types of remote sensing images to construct a multi-dimensional feature set; 3.2. Feature dimension optimization: Using principal component analysis and feature importance screening algorithms to remove redundant features and only retain the features that can best distinguish different land cover types; 3.3. Feature importance screening: Calculating the influence of different features on the classification results through algorithms, and selecting the most representative features. This process can significantly improve the discrimination ability of the classification model and reduce the influence of noise data on the classification results; 3.4. Feature library expansion and update: As the remote sensing data is continuously updated, the feature library will also be continuously expanded to ensure its matching with the latest remote sensing image data.

[0017] Preferably, the specific operations of step 4 are as follows: 4.1. Model training and cross-validation: Using the random forest algorithm for training, inputting a multi-source feature set, and optimizing the number and depth of decision trees through cross-validation; by adjusting the parameters of the decision trees, ensuring that the model can accurately distinguish between different land cover types; 4.2. Solving the problem of sample imbalance: Using oversampling and class weight adjustment techniques to ensure that different classes of cultivated land and non-cultivated land are fully represented in model training; 4.3. Model output and classification results: The trained RF model outputs a probability map of abandoned land, which is converted into a binary classification result by setting a threshold; based on this classification result, an accurate spatial distribution map of abandoned land can be generated; 4.4. Model evaluation and optimization: By comparing with an independent test set, evaluating the classification accuracy of the model; the accuracy indicators include overall accuracy, Kappa coefficient, and class accuracy. According to these indicators, further optimizing the model parameters to improve the classification accuracy.

[0018] Preferably, the specific operations in the 5th step are as follows: 5.1 Spatial filtering and noise removal: Perform spatial filtering on the classification results to remove noise and generate a refined distribution map of abandoned arable land; 5.2 Accuracy evaluation and error analysis: Use the confusion matrix method to evaluate the accuracy of the classification results, calculate the overall accuracy, Kappa coefficient, user accuracy and producer accuracy of each category; Through the analysis of different misclassification sources, propose improvement suggestions and correction schemes; 5.3 Accuracy improvement and model optimization: Optimize the model according to the accuracy evaluation results, especially further adjust the distinction between abandoned arable land and non-cultivated land; Correct the misclassification situation to further improve the robustness of the model; 5.4 Generation of final results: Based on the corrected classification results, generate a monitoring map of abandoned arable land and provide decision-making support data to help local governments and relevant departments formulate reasonable land management and ecological protection strategies.

[0019] Compared with the prior art, the present invention has obvious advantages and beneficial effects. Specifically, as can be seen from the above technical solutions: The present invention first integrates multi-source remote sensing image data and performs strict preprocessing on it to ensure the spatio-temporal consistency of the data. By deeply analyzing the NDVI and NDSI time series, key phenological period information is accurately extracted. In the feature construction stage, multiple spectral indices are fused, and principal component analysis and feature importance screening are used to optimize the feature dimension and improve the model performance. The random forest model is used for training and prediction, and combined with oversampling and class weight adjustment techniques to solve the problem of sample imbalance, and finally a high-precision spatial distribution map of abandoned arable land is generated. Through post-processing and accuracy evaluation, the reliability and accuracy of the classification results are ensured. This method effectively overcomes the deficiencies of traditional monitoring technologies in terms of science, efficiency and accuracy, and provides strong technical support for cultivated land protection and management. Brief Description of the Drawings

[0020] Figure 1 is the processing flow chart of the present invention. Detailed Embodiment

[0021] The present invention discloses a method for identifying cultivated land abandonment based on multi-source remote sensing data and time series correction, including the following steps: Step 1: Data acquisition and preprocessing. Data acquisition and preprocessing are the basis of the present invention, involving the integration, cleaning and correction of remote sensing image data. In order to accurately monitor abandoned arable land, appropriate remote sensing data must be selected and strict preprocessing must be performed on it. The specific operations are as follows: 1.1 Remote sensing data acquisition: Select multi-source remote sensing image data covering the growth season data of the target area; Landsat satellite images are used for remote sensing data acquisition, which provides data with a spatial resolution of 30 meters, can effectively capture surface features, and is suitable for monitoring abandoned farmland; and multi-temporal optical remote sensing image data such as Landsat-5, Landsat-7, and Landsat-8, Sentinel-2 optical images, and Sentinel-1 radar image data are obtained from Google Earth Engine (GEE), covering a long time scale (1989 - 2021).

[0022] 1.2 Cloud masking and atmospheric correction: When using optical remote sensing data, first remove cloud interference through the cloud masking algorithm to ensure the clarity of the image; then, perform atmospheric correction to eliminate the influence of atmospheric scattering and absorption on the image and ensure the spectral consistency of the data.

[0023] 1.3 Geometric registration and accuracy correction: Geometric registration is performed on all images to ensure spatial alignment between different images and guarantee spatio-temporal consistency; in addition, radiometric correction is also performed on the images to eliminate sensor differences and atmospheric effects and ensure the accuracy of the data.

[0024] 1.4 Radar image processing: Radar data is denoised using the Refined Lee filter, further radiometric calibration and terrain correction are performed, and the backscatter coefficient is extracted from the processed radar image, which helps analyze different land cover types.

[0025] 1.5 Data integration and spatio-temporal consistency: By fusing optical images and radar images, ensure the spatio-temporal consistency of different source data and provide reliable input data for subsequent analysis.

[0026] Step 2: Extraction of key phenological periods: Based on the NDVI and NDSI time series, two sensitive periods are extracted through filtering, smoothing, and extreme value detection algorithms: the least vegetative period (LVP, the lowest NDVI value reflects bare soil characteristics) and the peak vegetative period (PVP, the NDVI peak characterizes saturated vegetation cover). The phenological period positioning error is controlled within 5 days to ensure the accuracy of time series analysis. The specific operations are as follows: 2.1 NDVI and NDSI time series analysis: Use NDVI (Normalized Difference Vegetation Index) and NDSI (Bare Soil Index) time series data to extract key phenological periods through extreme value detection algorithms; NDVI reflects the growth of vegetation, and NDSI is highly sensitive to bare soil.

[0027] 2.2, Extraction of sensitive periods: By analyzing the changes in NDVI and NDSI, the least vegetative period (LVP) and the peak vegetative period (PVP) are extracted; LVP is usually the stage with the least vegetation cover and significant bare soil, while PVP is the stage with saturated vegetation cover, which can reflect the maximum vegetation growth state of cultivated land.

[0028] 2.3, Error control: When extracting phenological periods, smoothing filtering and extreme value detection algorithms are used to ensure that the phenological period positioning error is controlled within 5 days, guaranteeing the accuracy of time series analysis.

[0029] 2.4, Phenological period correction: To accurately calibrate LVP and PVP, corrections are made according to the growth cycles of different crops to ensure that the extracted phenological periods adapt to the changes under different regional and climatic conditions.

[0030] Step 3: Feature library construction: Integrate multi-source remote sensing data and spectral indices such as NDVI, EVI, and BSI to construct a multi-dimensional feature set. Through principal component analysis (PCA) and feature importance screening, optimize the feature dimensions and improve the discrimination efficiency of the classification model. The specific operations are as follows: 3.1, Fusion of spectral indices and remote sensing data: Combine multiple spectral indices such as NDVI, EVI, and BSI, as well as different types of remote sensing images (optical and radar data) to construct a multi-dimensional feature set. These features can describe different change patterns of the surface in detail, helping to improve the classification accuracy.

[0031] 3.2, Feature dimension optimization: Use principal component analysis and feature importance screening algorithms to remove redundant features and only retain the features that can best distinguish different land cover types to improve the efficiency of the classification model.

[0032] 3.3, Feature importance screening: Calculate the influence of different features on the classification results through algorithms and select the most representative features. This process can significantly improve the discrimination ability of the classification model and reduce the impact of noise data on the classification results.

[0033] 3.4, Feature library expansion and update: As remote sensing data is continuously updated, the feature library will also be continuously expanded to ensure its matching with the latest remote sensing image data, thereby improving the adaptability of the model and the reliability of long-term monitoring.

[0034] Step 4: Random forest model training and prediction: Use the random forest algorithm, input the multi-source feature set, and optimize the model parameters (number of decision trees, depth) through cross-validation; for the problem of sample imbalance, apply oversampling and class weight adjustment. The model outputs a probability map of abandoned land, and through threshold segmentation, a binary classification result is generated to achieve the mapping of the spatial distribution of abandoned land at the regional scale. The specific operations are as follows: 4.1. Model Training and Cross-Validation: The random forest algorithm is used for training. The multi-source feature set is input, and the number and depth of decision trees are optimized through cross-validation. By adjusting the parameters of the decision trees, the model can accurately distinguish between different land cover types.

[0035] 4.2. Solving the Problem of Sample Imbalance: Oversampling and class weight adjustment techniques are used to ensure that different classes of arable land and non-arable land are fully represented in model training.

[0036] 4.3. Model Output and Classification Results: The trained RF model outputs a probability map of abandoned land, which is converted into a binary classification result by setting a threshold. Based on this classification result, an accurate spatial distribution map of abandoned land can be generated.

[0037] 4.4. Model Evaluation and Optimization: By comparing with an independent test set, the classification accuracy of the model is evaluated. The accuracy metrics include overall accuracy, Kappa coefficient, and class accuracy, etc. According to these metrics, the model parameters are further optimized to improve the classification accuracy.

[0038] Step 5: Post-Processing and Accuracy Evaluation: The classification result is spatially filtered to remove noise, generating a refined distribution map of abandoned land. The confusion matrix is calculated based on the independent test set to evaluate the overall accuracy, Kappa coefficient, and class-specificity metrics, analyze the main sources of misclassification, and propose temporal persistence rules to optimize the results. The final results meet the requirements of high-resolution monitoring and support decision-making applications. The specific operations are as follows: 5.1. Spatial Filtering and Noise Removal: The classification result is spatially filtered to remove noise and generate a refined distribution map of abandoned land. Spatial filtering can effectively smooth the classification result, reduce the impact of misclassification, and ensure the clarity and accuracy of the final map.

[0039] 5.2. Accuracy Evaluation and Error Analysis: The confusion matrix method is used to evaluate the accuracy of the classification result, calculating the overall accuracy, Kappa coefficient, user accuracy, and producer accuracy of each class. By analyzing different sources of misclassification, improvement suggestions and correction schemes are proposed.

[0040] 5.3. Accuracy Improvement and Model Optimization: According to the accuracy evaluation results, the model is optimized, especially for further adjustment of the distinction between abandoned land and non-arable land. The misclassified cases (such as grassland or shrubland misclassified as abandoned land) are corrected to further improve the robustness of the model.

[0041] 5.4. Generation of Final Results: Based on the corrected classification result, a monitoring map of abandoned land is generated, and decision support data is provided to help local governments and relevant departments formulate reasonable land management and ecological protection strategies. Example

[0042] 1. Data acquisition: Acquire Landsat-7 / 8, Sentinel-2 optical imagery and Sentinel-1 radar imagery data for areas A and B from 2019 to 2024, and perform cloud detection and cloud removal on the optical imagery to ensure data integrity during key phenological periods.

[0043] 2. Extraction of key phenological periods: Based on the time series analysis of NDVI and NDSI, extract data during the LVP and PVP periods.

[0044] 3. Feature library construction: Combine multi-band data of optical imagery and radar imagery to construct spectral feature libraries for abandoned farmland and non-abandoned farmland. Introduce multiple spectral indices such as NDVI, EVI, GCVI, BSI, and LSWI as features.

[0045] 4. Model training and prediction: Use the random forest algorithm for model training. The input features include multi-band data of optical imagery, backscatter coefficients of radar imagery, and multiple spectral indices. Use the trained model to predict the cultivated land abandonment situation in 2024 and generate an abandoned farmland distribution map.

[0046] 5. Accuracy evaluation: Evaluate the classification accuracy of the model through the confusion matrix and Kappa coefficient. The results show that the overall accuracy of area A is 91.06% and the Kappa coefficient is 82.14%; the overall accuracy of area B is 93.96% and the Kappa coefficient is 81.75%.

[0047] The method of the present invention has the following advantages: 1. Significantly improved classification accuracy: Combine the phenological characteristics of optical imagery with the anti-interference ability of radar data to effectively distinguish abandoned farmland from fallow land, reduce the misjudgment risk of mixed pixels and vegetation transition zones, and achieve high-confidence spatial identification.

[0048] 2. Enhanced anti-environmental interference ability: Utilize the characteristic of radar data to penetrate clouds to make up for the data missing of optical imagery in cloudy and rainy areas, ensure the monitoring continuity under complex climate conditions, and at the same time the red edge band enhances the sensitive capture of the physiological state of vegetation.

[0049] 3. Optimized calculation efficiency and scalability: Based on the cloud parallel computing framework, achieve automated and rapid processing of large-scale areas, significantly shorten the analysis cycle, and support the requirements of high-frequency dynamic monitoring.

[0050] 4. Improved spatial resolution: Utilize high-resolution imagery to be able to more precisely capture small-scale and fragmented cultivated land abandonment phenomena.

[0051] The technical principle of the present invention has been described above in connection with specific embodiments. These descriptions are only for explaining the principle of the present invention and cannot be construed in any way as a limitation on the protection scope of the present invention. Based on the explanations herein, those skilled in the art can readily conceive of other specific embodiments of the present invention without creative efforts, and these embodiments will fall within the protection scope of the present invention.

Claims

1. A method for identifying abandoned cultivated land based on multi-source remote sensing data and time series correction, characterized in that it includes the following steps: Step 1: Data acquisition and preprocessing, the specific operations are as follows: 1.

1. Remote sensing data acquisition: Select multi-source remote sensing image data covering the growth season data of the target area; 1.

2. Cloud masking and atmospheric correction: When using optical remote sensing data, first remove cloud interference through the cloud masking algorithm to ensure the clarity of the image; then, perform atmospheric correction to eliminate the influence of atmospheric scattering and absorption on the image and ensure the spectral consistency of the data; 1.

3. Geometric registration and accuracy correction: Geometrically register all images to ensure spatial alignment between different images and guarantee spatio-temporal consistency; in addition, perform radiometric correction on the images to eliminate sensor differences and atmospheric effects and ensure the accuracy of the data; 1.

4. Radar image processing: The radar data is denoised using the Refined Lee filter, further radiometric calibration and terrain correction are carried out, and the backscattering coefficient is extracted from the processed radar image, which helps to analyze different land cover types; 1.

5. Data integration and spatio-temporal consistency: By fusing optical images and radar images, ensure the spatio-temporal consistency of different source data and provide reliable input data for subsequent analysis; Step 2: Extraction of key phenological periods: Based on the NDVI and NDSI time series, extract two sensitive periods, the minimum nutrition period and the peak nutrition period, through filtering, smoothing and extreme value detection algorithms. The phenological period positioning error is controlled within 5 days to ensure the accuracy of the time series analysis; Step 3: Feature library construction: Integrate multi-source remote sensing data and spectral indices to construct a multi-dimensional feature set. Optimize the feature dimensions through principal component analysis and feature importance screening to improve the discrimination efficiency of the classification model; Step 4: Random forest model training and prediction: Use the random forest algorithm, input the multi-source feature set, and optimize the model parameters through cross-validation; for the problem of sample imbalance, apply oversampling and class weight adjustment; the model outputs a probability map of abandoned land, and a binary classification result is generated through threshold segmentation to achieve the mapping of the spatial distribution of abandoned land at the regional scale; Step 5: Post-processing and accuracy evaluation: Perform spatial filtering on the classification results to remove noise and generate a refined abandoned land distribution map; calculate the confusion matrix based on the independent test set, evaluate the overall accuracy, Kappa coefficient and class specificity index, analyze the main sources of misclassification and propose time series persistence rules to optimize the results; the final results meet the requirements of high-resolution monitoring and support decision-making applications.

2. The cultivated land abandonment identification method based on multi-source remote sensing data and time series correction according to claim 1, wherein: In step 1.1, Landsat satellite images are used for remote sensing data acquisition, and multi-temporal optical remote sensing image data of Landsat-5, Landsat-7 and Landsat-8, as well as Sentinel-2 optical images and Sentinel-1 radar images are obtained from Google Earth Engine.

3. The cultivated land abandonment identification method based on multi-source remote sensing data and time series correction according to claim 1, characterized in that: The specific operations of the second step are as follows: 2.

1. NDVI and NDSI time series analysis: Use NDVI and NDSI time series data to extract key phenological periods through extreme value detection algorithms; NDVI reflects the growth of vegetation, and NDSI is sensitive to bare soil; 2.

2. Extraction of sensitive periods: By analyzing the changes in NDVI and NDSI, the minimum vegetative period and the peak vegetative period are extracted; LVP is usually the stage with the least vegetation cover and significant bare soil, while PVP is the stage with saturated vegetation cover, which can reflect the maximum vegetation growth state of cultivated land; 2.

3. Error control: When extracting phenological periods, a smoothing filter and an extreme value detection algorithm are used to ensure that the phenological period positioning error is controlled within 5 days, guaranteeing the accuracy of time series analysis; 2.

4. Phenological period correction: To accurately calibrate LVP and PVP, corrections are made according to the growth cycles of different crops to ensure that the extracted phenological periods adapt to the changes under different regional and climatic conditions.

4. The cultivated land abandonment recognition method based on multi-source remote sensing data and time series correction according to claim 1, wherein: The specific operations of the third step are as follows: 3.

1. Fusion of spectral indices and remote sensing data: Combine multiple spectral indices such as NDVI, EVI, and BSI, and different types of remote sensing images to construct a multi-dimensional feature set; 3.

2. Feature dimension optimization: Use principal component analysis and feature importance screening algorithms to remove redundant features and only retain the features that can best distinguish different land cover types; 3.

3. Feature importance screening: Calculate the influence of different features on the classification results through algorithms and select the most representative features; 3.

4. Feature library expansion and update: As the remote sensing data is continuously updated, the feature library will also be continuously expanded to ensure its matching with the latest remote sensing image data.

5. The cultivated land abandonment recognition method based on multi-source remote sensing data and time series correction according to claim 1, characterized in that: The specific operations of the fourth step are as follows: 4.

1. Model training and cross-validation: Use the random forest algorithm for training, input a multi-source feature set, and optimize the number and depth of decision trees through cross-validation; by adjusting the parameters of the decision trees, ensure that the model can accurately distinguish between different land cover types; 4.

2. Solving the problem of sample imbalance: Use oversampling and class weight adjustment techniques to ensure that different categories of cultivated land and non-cultivated land are fully represented in model training; 4.

3. Model output and classification results: The trained RF model outputs a probability map of abandoned land, which is converted into a binary classification result by setting a threshold; based on this classification result, an accurate spatial distribution map of abandoned land can be generated; 4.

4. Model evaluation and optimization: Evaluate the classification accuracy of the model by comparing it with an independent test set; the accuracy indicators include overall accuracy, Kappa coefficient, and class accuracy. According to these indicators, further optimize the model parameters to improve the classification accuracy.

6. The cultivated land abandonment recognition method based on multi-source remote sensing data and time series correction according to claim 1, wherein: The specific operations of the fifth step are as follows: 5.

1. Spatial filtering and noise removal: Perform spatial filtering on the classification results to remove noise and generate a refined distribution map of abandoned land; 5.

2. Accuracy evaluation and error analysis: Use the confusion matrix method to evaluate the accuracy of the classification results, calculate the overall accuracy, Kappa coefficient, and the user accuracy and producer accuracy of each category; through the analysis of different misclassification sources, put forward improvement suggestions and correction schemes; 5.

3. Accuracy improvement and model optimization: Optimize the model according to the accuracy evaluation results, especially make further adjustments to the distinction between abandoned land and non-cultivated land; correct the misclassified situations to further improve the robustness of the model; 5.

4. Final result generation: Based on the corrected classification results, generate a monitoring map of abandoned farmland and provide decision support data to assist in formulating reasonable land management and ecological protection strategies.

Images