Land cover change detection method and device based on texture feature enhancement, equipment, medium and product

Abstract

The application provides a land cover change detection method, device, equipment, medium and product based on texture feature enhancement, the method comprises the following steps: acquiring remote sensing image data set of a monitoring area; based on the remote sensing image data set, extracting original spectral band and calculating vegetation index band; based on the gray level co-occurrence matrix algorithm, extracting the contrast texture feature band of the vegetation index band; based on the original spectral band, the vegetation index band and the contrast texture feature band, constructing a multi-dimensional input data set, and performing continuous change detection on the multi-dimensional input data set to obtain a time series model parameter; based on the multi-dimensional input data set and the time series model parameter, outputting the annual land cover classification result through the land cover change detection model. The application can significantly improve the distinguishability of easily confused ground objects in complex geographical environment.

Description

Technical Field

[0001] This invention relates to the field of land detection technology, and in particular to a method, apparatus, equipment, medium and product for detecting land cover change based on texture feature enhancement. Background Technology

[0002] Currently, existing land cover change monitoring schemes are mainly divided into two categories, but both have limitations. The first category is the traditional CCDC method, which focuses on the temporal evolution of a single pixel and detects changes by fitting time series curves. However, it relies solely on the spectral characteristics of the pixel itself and rarely considers the spatial structural relationship between the pixel and its surrounding environment. In complex geographical environments, the phenomenon of different objects with the same spectrum or different spectra of the same object is quite serious, and the model is prone to confusion when distinguishing land features with similar spectral characteristics. The second category is monitoring methods based on multi-temporal discrete comparison and texture feature fusion, such as the scheme disclosed in CN 120125885 A. Although it introduces texture features, it belongs to discrete multi-temporal detection, only comparing isolated time points, and placing texture fusion after classification prediction. It does not incorporate spatial structural features into the long-term continuous evolution trajectory modeling. It is evident that existing monitoring schemes either focus only on the temporal evolution of single pixels while ignoring spatial structural relationships, or only perform discrete temporal comparisons and separate spatial texture from long-term continuous modeling. This makes it difficult to distinguish spectral confusion features with the same spectrum or the same spectrum with different spectra in complex geographical environments, thus affecting the accuracy of land cover classification. Summary of the Invention

[0003] This invention provides a land cover change detection method, device, equipment, medium, and product based on texture feature enhancement, which solves the technical problem of existing monitoring schemes being unable to distinguish spectral confusion features of the same spectrum or the same spectrum of different features in complex geographical environments.

[0004] This invention provides a land cover change detection method based on texture feature enhancement, comprising the following steps: Obtain the remote sensing image dataset of the area to be monitored; Based on the remote sensing image dataset, the original spectral bands are extracted, and the vegetation index bands are calculated. Based on the gray-level co-occurrence matrix algorithm, the contrast texture feature bands of the vegetation index bands are extracted; Based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, a multidimensional input dataset is constructed, and continuous change detection is performed on the multidimensional input dataset to obtain time series model parameters. Based on the multidimensional input dataset and the time-series model parameters, the land cover change detection model outputs annual land cover classification results.

[0005] According to the land cover change detection method based on texture feature enhancement provided by the present invention, the step of acquiring a remote sensing image dataset of the area to be monitored includes: Obtain the initial remote sensing image dataset of the area to be monitored; The boundary range of the area to be monitored is determined in the form of an expanded buffer zone; Based on the boundary range, the initial remote sensing image dataset is preprocessed by cloud removal using the CFMAS masking algorithm to generate a new remote sensing image dataset.

[0006] According to the present invention, a land cover change detection method based on texture feature enhancement is provided, wherein the method for extracting contrast texture feature bands of the vegetation index bands based on the gray-level co-occurrence matrix algorithm includes: The floating-point pixel values ​​of the vegetation index band are linearly mapped to a preset gray level range to obtain the target vegetation index band. A sliding window is set up and traversed pixel by pixel on the target vegetation index band to generate multiple local sub-windows; For each local sub-window, a multi-directional gray-level co-occurrence matrix is ​​calculated using the gray-level co-occurrence matrix algorithm, and a contrast feature texture value is calculated based on the multi-directional gray-level co-occurrence matrix. Based on the contrast feature texture values ​​of each local sub-window, the contrast texture feature band of the vegetation index band is determined.

[0007] According to the land cover change detection method based on texture feature enhancement provided by the present invention, the step of performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters includes: The multidimensional input dataset is input into the continuous change detection algorithm to fit the pixel time series trajectory; The intercept term, trend term, sine and cosine harmonic coefficients of each order, and root mean square error are extracted from the pixel time series trajectory and used as parameters of the time series model.

[0008] According to the present invention, a land cover change detection method based on texture feature enhancement is provided, wherein the method outputs annual land cover classification results through a land cover change detection model based on the multidimensional input dataset and the time-series model parameters, including: A land cover change detection model was constructed using the random forest algorithm; Obtain the auxiliary environmental factors of the area to be monitored; The time series model parameters, the multidimensional input dataset, and the auxiliary environmental factors are fused using multidimensional features to obtain fused features; The fused features are input into the land cover change detection model, and the annual land cover classification results are output.

[0009] A land cover change detection method based on texture feature enhancement according to the present invention further includes: Obtain the actual land cover data of the area to be monitored; The annual land cover classification results are compared with the actual land cover data to calculate the overall accuracy and Kappa coefficient; The classification accuracy of the annual land cover classification results is evaluated based on the overall accuracy and the Kappa coefficient.

[0010] The present invention also provides a land cover change detection device based on texture feature enhancement, comprising the following modules: The acquisition module is used to acquire remote sensing image datasets of the area to be monitored. The calculation module is used to extract the original spectral bands and calculate the vegetation index bands based on the remote sensing image dataset. The extraction module is used to extract the contrast texture feature bands of the vegetation index bands based on the gray-level co-occurrence matrix algorithm. The construction module is used to construct a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, and to perform continuous change detection on the multidimensional input dataset to obtain time series model parameters. The output module is used to output annual land cover classification results based on the multidimensional input dataset and the time series model parameters through the land cover change detection model.

[0011] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the land cover change detection method based on texture feature enhancement as described above.

[0012] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the land cover change detection method based on texture feature enhancement as described above.

[0013] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the land cover change detection method based on texture feature enhancement as described above.

[0014] This invention provides a land cover change detection method, apparatus, device, medium, and product based on texture feature enhancement. The method involves acquiring a remote sensing image dataset of the area to be monitored; extracting original spectral bands and calculating vegetation index bands based on the remote sensing image dataset; extracting contrast texture feature bands from the vegetation index bands using a gray-level co-occurrence matrix algorithm; constructing a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands; performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters; and outputting annual land cover classification results through a land cover change detection model based on the multidimensional input dataset and the time-series model parameters. This solves the technical problem of existing monitoring schemes struggling to distinguish spectrally confused features (either similar or dissimilar spectra) in complex geographical environments. Compared to existing technologies, this invention introduces contrast texture features from vegetation index bands, which, together with the original spectral bands and vegetation index bands, construct a multi-dimensional input dataset for continuous change detection. This allows the time-series model to explicitly incorporate spatial structure information within the pixel neighborhood while fitting long-term spectral evolution trajectories. This overcomes the shortcomings of traditional CCDC methods that rely solely on single spectral features, upgrading pixel-by-pixel spectral modeling to spatiotemporal correlation modeling. This effectively enhances the ability to distinguish between different objects with the same spectrum or different spectral features with the same spectrum, reducing the risk of false positives and false negatives due to spectral confusion. Furthermore, by incorporating texture features into the time-series modeling beforehand, rather than post-processing them in classification decisions, the model can capture spatial pattern evolution during gradual, seasonal, and other non-abrupt change processes. This improves the temporal consistency and spatial continuity of land cover classification, ultimately achieving higher classification accuracy and change detection reliability in complex geographical environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is one of the flowcharts of the land cover change detection method based on texture feature enhancement provided by the present invention.

[0017] Figure 2 This is a remote sensing image of the land cover change detection method based on texture feature enhancement provided by the present invention.

[0018] Figure 3 This is the second flowchart of the land cover change detection method based on texture feature enhancement provided by the present invention.

[0019] Figure 4 This is a land cover classification sample distribution map of the land cover change detection method based on texture feature enhancement provided by the present invention.

[0020] Figure 5 This is a comparison chart of land cover classification results for the land cover change detection method based on texture feature enhancement provided by this invention.

[0021] Figure 6 This is a schematic diagram of the land cover change detection device based on texture feature enhancement provided by the present invention.

[0022] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] The following is combined Figure 1 and Figure 2 The present invention describes a land cover change detection method based on texture feature enhancement. This method is applicable to any land cover change detection based on texture feature enhancement. The subject executing this method can be an electronic device or a land cover change detection device based on texture feature enhancement installed in the electronic device. The land cover change detection device based on texture feature enhancement can be implemented by software, hardware, or a combination of both.

[0025] Figure 1 This is one of the flowcharts of the land cover change detection method based on texture feature enhancement provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Obtain the remote sensing image dataset of the area to be monitored; It should be noted that the area to be monitored refers to the geographic spatial range where land cover change detection is required, which can be the boundary of a nature reserve, the scope of an administrative division, or the boundary of a watershed. A remote sensing image dataset refers to a collection of multi-temporal surface observation images covering the area to be monitored, acquired through satellite or airborne remote sensing sensors. Each temporal image typically contains multiple spectral bands, such as blue, green, red, near-infrared, and shortwave infrared bands.

[0026] In practical implementation, the geographical boundaries of the area to be monitored are first determined. Then, multi-temporal images covering the area are retrieved from a remote sensing data platform, such as Landsat, Sentinel-2, or HLS30 datasets. Images with cloud cover below 30% and continuous imaging time are selected to ensure sufficient observations every year within the monitoring period. Finally, the blue, green, red, near-infrared, and shortwave infrared bands of each image are acquired to form a remote sensing image dataset, such as... Figure 2 The image shown is a satellite remote sensing image of the Ningba Mountain Nature Reserve and its surrounding area.

[0027] Step 102: Based on the remote sensing image dataset, extract the original spectral bands and calculate the vegetation index bands; It should be noted that the original spectral bands refer to the reflected energy values ​​of different wavelength ranges on the Earth's surface directly recorded by remote sensing image sensors. Each band corresponds to a specific wavelength range, such as the blue band (BLUE), green band (GREEN), red band (RED), near-infrared band (NIR), and short-wave infrared band (SWIR1), and is stored in the form of a two-dimensional matrix, where each element represents the reflectance value of a pixel. Vegetation index bands are derived bands generated by mathematically combining the original spectral bands. They are used to highlight the coverage and growth status of surface vegetation. Common vegetation indices include the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI).

[0028] Step 103: Based on the gray-level co-occurrence matrix algorithm, extract the contrast texture feature bands of the vegetation index bands; It should be noted that the gray-level co-occurrence matrix (GLCM) is a second-order statistical method used to analyze image texture features. It constructs a square co-occurrence matrix by statistically analyzing the joint probability distribution of gray-level value pairs between two pixels with specific spatial relationships, such as fixed distance and fixed orientation. Contrast texture features are one of the statistical quantities derived from the GLCM, used to measure the degree of local variation and contrast intensity of gray-level values ​​within a local area of ​​an image. A higher contrast value indicates deeper texture grooves and more drastic local variations.

[0029] Understandably, extracting contrast texture features from vegetation index bands using the gray-level co-occurrence matrix algorithm extends vegetation information from the spectral dimension to the spatial texture dimension. Different land cover types often exhibit different structural patterns in space; for example, farmland displays regular striped textures, forests display irregular patchy textures, and bare land displays relatively homogeneous texture features. This spatial structural information is difficult to fully express in simple spectral features, while contrast texture features can quantify this spatial heterogeneity. By introducing texture features, land cover types with similar spectra but different spatial structures can be distinguished in the texture dimension, thereby enhancing the discriminative ability of change detection.

[0030] Step 104: Based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, construct a multidimensional input dataset, and perform continuous change detection on the multidimensional input dataset to obtain time series model parameters; It should be noted that the multidimensional input dataset is a composite data cube formed by stitching together the original spectral bands, vegetation index bands, and contrast texture feature bands along the channel dimension. Each time point image corresponds to a layer containing multiple feature bands. Continuous change detection is a time series analysis method that primarily captures the evolution trajectory of ground cover spectral and texture features over time by fitting a parametric model to the feature value sequence of each pixel on the time axis. The time series model parameters are the fitted model coefficients used to quantitatively describe the statistical characteristics of the pixel time series.

[0031] Understandably, the specific parameter configurations for the continuous change detection algorithm include: using millisecond-level Unix timestamps as the date format to improve the processing accuracy of time series data; setting the regularization parameter lambda to 0.002, and setting the minimum number of observations required for single-segment fitting and the chi-square probability threshold to balance the flexibility and robustness of model fitting.

[0032] Step 105: Based on the multidimensional input dataset and the time series model parameters, output the annual land cover classification results through the land cover change detection model.

[0033] It should be noted that the land cover change detection model refers to a machine learning or statistical model that can map temporal features (model parameters) and original features (multi-band values) into land cover category labels. The annual land cover classification results refer to the digital map generated for each year within the monitoring period, which contains the spatial distribution of various land cover categories (such as forests, grasslands, farmland, water bodies, bare land, etc.) for that year.

[0034] In practical implementation, the original multi-band observations for each year, the corresponding time-series model parameters, and possible environmental auxiliary data are combined into a feature vector for each pixel. This vector is then input into a pre-trained classification model, which assigns the most probable land cover category to the pixel. This step integrates the static spectral texture features and dynamic time-series model parameters obtained earlier into the final classification decision, achieving a complete closed loop from "change detection" to "classification after change." This ensures that the output not only indicates the time of the change but also clearly identifies the land cover type before and after the change. Specifically, auxiliary environmental factors (including elevation, slope, aspect, annual average rainfall, and annual average temperature) for the area to be monitored are first acquired and resampled to the same spatial resolution as the remote sensing image. For each pixel, its 10-dimensional original input data for each year, the corresponding time-series model parameters for that year (intercept, trend term, harmonic coefficients, root mean square error), and the environmental factors at that location are extracted and concatenated into a multi-dimensional feature vector. These feature vectors are then input into a pre-trained random forest classification model. The model uses decision tree voting based on the feature values ​​to assign the pixel to a specific land cover category. Perform this operation on all pixels and all years to obtain the classification result layer for each year.

[0035] The land cover change detection method based on texture feature enhancement provided in this invention introduces contrast texture features from vegetation index bands. These features, along with the original spectral bands and vegetation index bands, are used to construct a multi-dimensional input dataset for continuous change detection. This allows the time-series model to explicitly incorporate spatial structure information within the pixel neighborhood while fitting long-term spectral evolution trajectories. This overcomes the shortcomings of traditional CCDC methods that rely solely on single spectral features, upgrading pixel-by-pixel spectral modeling to spatiotemporal correlation modeling. This effectively enhances the ability to distinguish between different objects with the same spectrum or different spectral features with the same object, reducing the risk of false positives and false negatives due to spectral confusion. Furthermore, by incorporating texture features into the time-series modeling beforehand, rather than post-processing them in classification decisions, the model can capture spatial pattern evolution during gradual, seasonal, and other non-abrupt change processes. This improves the temporal consistency and spatial continuity of land cover classification, ultimately achieving higher classification accuracy and change detection reliability in complex geographical environments.

[0036] Based on any of the above embodiments, obtaining the remote sensing image dataset of the area to be monitored includes: Obtain the initial remote sensing image dataset of the area to be monitored; The boundary range of the area to be monitored is determined in the form of an expanded buffer zone; Based on the boundary range, the initial remote sensing image dataset is preprocessed by cloud removal using the CFMAS masking algorithm to generate a new remote sensing image dataset.

[0037] It should be noted that the initial remote sensing image dataset refers to the raw image collection directly acquired from the remote sensing data source. These images have not yet undergone boundary cropping and cloud / snow removal processing for the area to be monitored, and may contain redundant areas beyond the monitored area, or be contaminated by atmospheric and weather factors such as clouds, cloud shadows, and snow. The extended buffer zone refers to the new boundary area formed by extending a certain distance outward from the original vector boundary or minimum bounding rectangle boundary of the monitored area. The boundary extent refers to the new geographic spatial range enclosed by this extension operation.

[0038] Understandably, the CFMAS masking algorithm is a decision tree-based automatic detection algorithm for clouds, cloud shadows, snow, and water bodies in remote sensing images. Its input consists of reflectance data from various spectral bands and brightness and temperature data from the thermal infrared band. The output is a quality label band, where each pixel is labeled as one of the following categories: clear land, clear water, cloud shadow, thin cloud, thick cloud, or snow. Cloud removal preprocessing refers to excluding pixels labeled as clouds, cloud shadows, or snow—invalid observations—based on the quality labels output by CFMAS, in subsequent processing, retaining only clear pixels for calculation.

[0039] In the specific implementation, for each image in the initial remote sensing image dataset, reflectance data for all spectral bands and brightness and temperature data for the thermal infrared bands are read. The band data is organized and input according to the format and order required by the CFMAS algorithm. The algorithm executes an internal decision tree classification process, sequentially determining whether each pixel meets the spectral and temperature threshold conditions for clouds, cloud shadows, snow, and water bodies, and finally outputs an integer mask array of the same size as the image, where different values ​​represent different quality categories. Based on this mask array, pixel-by-pixel processing is performed on the original reflectance bands: for pixels whose mask values ​​indicate non-clear skies, such as thick clouds, cloud shadows, or snow, their reflectance values ​​in all spectral bands are set to invalid labels, such as negative maxima or null values. For pixels whose mask values ​​indicate clear land or clear water bodies, their original reflectance values ​​are retained. After scene-by-scene processing, all images are accompanied by valid observation labels for each pixel. Based on the boundary range, spatial cropping is performed on all images, retaining only the pixels within the boundary range, forming the final high-quality remote sensing image dataset (i.e., the remote sensing image dataset).

[0040] The land cover change detection method based on texture feature enhancement provided in this invention eliminates the edge effect generated at the original region boundary during the calculation of the gray-level co-occurrence matrix sliding window by determining the boundary range in the form of an outer buffer, and avoids texture feature calculation errors caused by the lack of neighborhood information of boundary pixels. At the same time, the CFMAS masking algorithm is used for cloud removal preprocessing to remove contaminated pixels covered by clouds and cloud shadows, ensuring the purity of effective observations in the remote sensing image dataset, and directly improving the extraction accuracy of contrast texture feature bands and the quality of the dataset.

[0041] Based on any of the above embodiments, the step of performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters includes: The multidimensional input dataset is input into the continuous change detection algorithm to fit the pixel time series trajectory; The intercept term, trend term, sine and cosine harmonic coefficients of each order, and root mean square error are extracted from the pixel time series trajectory and used as parameters of the time series model.

[0042] It should be noted that continuous change detection algorithms refer to a class of methods based on time series fitting and monitoring abrupt changes, such as the CCDC algorithm, which can perform piecewise linear or harmonic fitting on multi-temporal observations of each pixel (using Fourier harmonic functions to fit the intercept, trend, and harmonic coefficients of the time series). A pixel time series trajectory refers to treating the sequence of multi-band observations of a pixel at all time points as one or more curves that change with time. The intercept term refers to the baseline value of the fitted function at the zero point of time. The trend term refers to the rate coefficient of linear change with time, used to characterize a long-term increasing or decreasing trend. The sine and cosine harmonic coefficients refer to the amplitude parameters when fitting periodic fluctuations using Fourier series; the first harmonic corresponds to a fluctuation with a one-year cycle, and the second harmonic corresponds to a fluctuation with a six-month cycle. The root mean square error refers to the standard deviation of the deviation between the fitted value and the actual observed value, used to measure the quality of the fit.

[0043] In practical implementation, multidimensional input data (each time point corresponds to a 10-dimensional feature vector) can be input into a continuous time series model. The algorithm automatically fits a mathematical expression describing the long-term trend and seasonal variation of each pixel in each band. The intercept, trend term, harmonic coefficient, and root mean square error are extracted from the fitting results as a set of highly compressed but information-rich parameters. These parameters retain all the key dynamic features of the original time series while eliminating noise and redundancy from period-by-period observations, becoming efficient input features for subsequent classification tasks.

[0044] The land cover change detection method based on texture feature enhancement provided in this invention inputs a multi-dimensional input dataset containing contrast texture feature bands into a continuous change detection algorithm. This algorithm fits parameters such as the intercept, trend, and harmonic coefficients of the time series data, enabling the model to not only possess spatial perception capabilities but also fully preserve the seasonal fluctuations and long-term gradual evolution trajectories of land cover. The intercept and trend terms capture the long-term state and rate of change of land cover, while the harmonic coefficients characterize the seasonal fluctuation patterns. Compared to existing methods that rely solely on spectral time series parameters, the parameters extracted in this scheme simultaneously contain information on spectral evolution and spatial heterogeneity. This effectively distinguishes spectrally similar but structurally different land cover features, accurately identifies gradual and seasonal disturbances, and significantly improves the accuracy of capturing complex dynamic changes.

[0045] Based on any of the above embodiments, the step of outputting annual land cover classification results through a land cover change detection model based on the multidimensional input dataset and the time-series model parameters includes: A land cover change detection model was constructed using the random forest algorithm; Obtain the auxiliary environmental factors of the area to be monitored; The time series model parameters, the multidimensional input dataset, and the auxiliary environmental factors are fused using multidimensional features to obtain fused features; The fused features are input into the land cover change detection model, and the annual land cover classification results are output.

[0046] It's important to note that the Random Forest algorithm is an ensemble learning algorithm. It uses a bootstrap sampling method to extract multiple subsets of samples with replacement from the original training sample set, and independently trains a decision tree on each subset. When splitting at a node, each decision tree randomly selects a subset of candidate features from all features to choose the optimal splitting feature. Finally, the prediction result is determined by the votes or average of all decision trees during ensemble analysis. The land cover change detection model is a Random Forest classifier trained on this algorithm. Auxiliary environmental factors refer to geographic environmental variables independent of remote sensing imagery that control or correlate with the distribution of land cover types, such as elevation, slope, aspect, rainfall, and temperature. Multidimensional feature fusion refers to concatenating feature vectors from different data sources or different physical attributes along the sample dimension to form a longer feature vector for simultaneous use by the classification model.

[0047] In practice, auxiliary environmental factors include topographic data (such as the SRTM digital elevation model) and temperature and precipitation data (such as WorldClim bioclimatic variable data).

[0048] In the specific implementation, a random forest classifier with good robustness and anti-overfitting ability is first selected as the classification model. Then, auxiliary environmental factors independent of remote sensing time-series information are introduced to supplement prior knowledge of topography and climate. Next, the previously extracted time-series model parameters (dynamic features), the original multidimensional input data (static or annual spectral texture features), and the auxiliary environmental factors (spatial stationary features) are concatenated to form a complementary fused feature vector. Finally, this fused feature is input into the random forest model, which outputs the most likely land cover category for each pixel in each year, thus obtaining an annual classification map. Specifically, the hyperparameters of the random forest classifier are first set: the number of decision trees is set to 200, and other parameters are kept at default (such as the number of features randomly selected when splitting nodes is the square root of the total number of features). Then, the auxiliary environmental factor dataset of the area to be monitored is obtained, including elevation, slope, and aspect from the digital elevation model, as well as annual average rainfall and annual average temperature from meteorological interpolation data. All factors are resampled to the same 30-meter spatial resolution as the remote sensing image. For each pixel and each year, a fused feature vector is constructed: the time-series model parameters (e.g., intercept, trend term, first sine, first cosine, and root mean square error for each band, totaling 10 × 5 = 50 dimensions), the multi-dimensional input data for that year (10 dimensions), and the five auxiliary environmental factors (elevation, slope, aspect, rainfall, and temperature) at that pixel location are sequentially concatenated to obtain a 65-dimensional fused feature vector. These fused feature vectors and their corresponding real land cover labels (from manually interpreted or ground survey data) are then input into a random forest model for training. After training, for the pixel and year to be predicted, a 65-dimensional feature vector is also constructed and input into the model. The 200 decision trees within the model independently predict the category, and the category with the most votes is taken as the final land cover classification result. All pixels and years are traversed to generate a yearly land cover classification map.

[0049] The land cover change detection method based on texture feature enhancement provided in this invention employs a random forest algorithm to construct a land cover change detection model and fuses temporal model parameters, multidimensional input datasets, and auxiliary environmental factors into multidimensional features. This fusion strategy simultaneously introduces temporal parameters reflecting the long-term evolution of land cover, original features reflecting spectral and spatial heterogeneity, and environmental factors reflecting topographic and climatic conditions. This enables the random forest model to comprehensively utilize dynamic and static, remote sensing and non-remote sensing information for decision-making, thereby significantly improving the accuracy and spatial consistency of annual land cover classification results.

[0050] Based on any of the above embodiments, it further includes: Obtain the actual land cover data of the area to be monitored; The annual land cover classification results are compared with the actual land cover data to calculate the overall accuracy and Kappa coefficient; The classification accuracy of the annual land cover classification results is evaluated based on the overall accuracy and the Kappa coefficient.

[0051] It should be noted that true land cover data can be obtained through manual interpretation of high-resolution imagery, field surveys, or third-party high-precision products, and is considered to closely approximate the actual ground conditions. Overall accuracy refers to the proportion of correctly classified pixels out of all pixels involved in the verification, reflecting the overall classification accuracy rate. The Kappa coefficient is a statistical indicator used to measure the consistency between the classification results and the true reference data, to exclude the influence of random consistency. Its value typically ranges from -1 to 1, with higher values ​​indicating stronger consistency.

[0052] In the implementation, firstly, highly reliable validation data, independent of the training samples, is acquired as a benchmark to measure the accuracy of the classification results. Then, pixel-by-pixel, the classification results for each year are compared with the corresponding real data for that year, and the number of correctly classified pixels is counted to calculate the overall accuracy. Simultaneously, a confusion matrix is ​​constructed, and the Kappa coefficient is calculated based on the difference between observational consistency and random expected consistency. Finally, the reliability of the classification results is quantitatively evaluated based on the overall accuracy and the Kappa coefficient. When the overall accuracy is higher than 70% and the Kappa coefficient is higher than 0.6, the classification results can be considered to have good consistency.

[0053] Specifically, verification points are first randomly selected from the monitored area according to spatial stratification, with each point corresponding to a spatial location. For each location, the true land cover category for each year is determined by professionals through comprehensive interpretation of high spatial resolution imagery (such as WorldView, QuickBird) or multi-temporal imagery, forming an annual true land cover dataset. Then, the annual land cover classification results are spatially overlaid with the aforementioned annual true land cover dataset. For each year, a confusion matrix of size C×C is constructed (C is the number of categories, such as forest, grassland, farmland, bare land, water, etc.), where the rows of the matrix represent the true category and the columns represent the classification category. Based on the confusion matrix, the overall accuracy and Kappa coefficient are calculated. (1) Overall accuracy = (Total number of correctly classified pixels / Total number of verified pixels) × 100%; The total number of correctly classified pixels is the sum of the elements on the diagonal of the confusion matrix, and the total number of verified pixels is the cumulative sum of all rows or all columns. (2) Kappa coefficient = (observational consistency - stochastic expected consistency) / (1 - stochastic expected consistency); Among them, observational consistency is the proportional form of overall accuracy (the number of correctly classified pixels divided by the total number of validated pixels), and random expected consistency = (the sum of the products of the true total number in each row and the total number of classified pixels in each column) / (the square of the total number of validated pixels).

[0054] Finally, the overall accuracy and Kappa coefficient for each year are calculated, and the average value is taken as the evaluation index of the overall performance of the method.

[0055] The land cover change detection method based on texture feature enhancement provided in this invention acquires real land cover data of the area to be monitored, compares the annual land cover classification results with the real data, and calculates the overall accuracy and Kappa coefficient. Overall accuracy directly reflects the degree of consistency between the classification results and the actual land cover types, while the Kappa coefficient eliminates the interference of random consistency and objectively measures the degree of agreement between the two. Quantitatively evaluating the annual classification results based on these two indicators allows for precise quantification and horizontal comparison of the method's classification performance, providing verifiable objective evidence for the method's effectiveness.

[0056] Figure 3 This is the second flowchart of the land cover change detection method based on texture feature enhancement provided by the present invention, as shown in Figure 3. Step 103 further includes steps 1031 to 1042: Step 1031: Map the floating-point pixel values ​​of the vegetation index band to a preset gray level range to obtain the target vegetation index band. It should be noted that the floating-point pixel value of the vegetation index band refers to a continuous value calculated through the spectral band, such as NDVI or EVI, and its value range is usually [-1, 1]. The preset gray level range refers to a range of discrete integer values ​​set by the user, such as [0, 100].

[0057] Step 1032: Set a sliding window and slide it pixel by pixel on the target vegetation index band to generate multiple local sub-windows; It's important to note that a sliding window refers to a fixed-size square pixel region, typically with odd-numbered side lengths such as 3, 5, 7, or 9 pixels, to ensure the window has a clearly defined center pixel. Pixel-by-pixel sliding traversal involves sequentially aligning the window's center with each pixel in the image, moving from left to right and from top to bottom with a step size of 1. Each time the window moves to a new position, the pixel region currently covered by the window becomes a local sub-window.

[0058] In the specific implementation, the window size is first set to N×N (N is an odd number). Then, the center pixel of the window is taken as the current processing pixel, and pixel-by-pixel sliding is performed across the entire spatial range of the target vegetation index band. At each sliding position, all grayscale values ​​within that N×N region are extracted to form a local sub-window. This decomposes the entire grayscale image into a large number of overlapping local neighborhoods, each neighborhood corresponding to a center pixel, thus providing spatial constraints for subsequent pixel-by-pixel texture feature extraction.

[0059] Step 1033: For each local sub-window, calculate the multi-directional gray-level co-occurrence matrix using the gray-level co-occurrence matrix algorithm, and calculate the contrast feature texture value based on the multi-directional gray-level co-occurrence matrix; It should be noted that the gray-level co-occurrence matrix (GLCM) is a statistical matrix used to describe the joint occurrence frequency of different gray-level pixel pairs within a local sub-window at a specific direction and distance. Specifically, it refers to multiple co-occurrence matrices formed by counting the occurrence frequency of gray-level pairs at different angles (typically 0°, 45°, 90°, and 135°) for the same local sub-window. The contrast feature texture value is a derived statistic of the GLCM, and the formula for calculating the contrast feature texture value is: Where i and j are gray levels, and p(i,j) is the joint probability of gray levels i and j in the gray-level co-occurrence matrix. This value reflects the degree of drastic change in gray level within the local window.

[0060] In the specific implementation, for each local sub-window, the gray-level co-occurrence matrix in four directions is first calculated, and the gray-level combination frequency of adjacent pixel pairs (distanced by 1 pixel) within the window is counted in each matrix. Then, the average value of the corresponding positions in the four matrices is taken to obtain the average gray-level co-occurrence matrix. Finally, based on this average matrix, a scalar value is calculated according to the contrast formula, which is used as the contrast feature texture value of the center pixel of the local sub-window. This step quantifies the spatial structure information within the local window into a numerical value; the greater the gray-level difference and the coarser the texture, the higher the contrast value.

[0061] Step 1034: Based on the contrast feature texture values ​​of each local sub-window, determine the contrast texture feature band of the vegetation index band.

[0062] It should be noted that the contrast texture feature band refers to a single-band image that has the same spatial resolution as the original image, but each pixel value represents the local contrast at that location.

[0063] In the specific implementation, the contrast feature texture value calculated for the center cell of each local sub-window can be filled according to the row and column positions of that cell in the original space. After the above calculation is completed for all cell positions, a two-dimensional matrix with the same spatial size as the input vegetation index band is formed. This matrix is ​​the contrast texture feature band. Each value in this band represents the degree of spatial heterogeneity in the local neighborhood of the corresponding location and can be used as an independent feature layer to participate in the subsequent construction of the multidimensional input dataset.

[0064] The land cover change detection method based on texture feature enhancement provided in this invention linearly maps the floating-point pixel values ​​of vegetation index bands to a preset gray-level range to obtain the target vegetation index band; a sliding window is set to traverse pixel by pixel to generate multiple local sub-windows; a multi-directional gray-level co-occurrence matrix is ​​calculated for each local sub-window and contrast feature texture values ​​are extracted; finally, the contrast texture feature band is determined. This process transforms the continuous floating-point value vegetation index into discrete gray-level spatial structure features. The contrast value directly quantifies the intensity of pixel gray-level differences within a local window, enabling it to keenly capture the spatial heterogeneity and roughness information of the land surface.

[0065] The embodiments of the present invention described above will now be described with reference to the accompanying drawings: Figure 4 This is a land cover classification sample distribution map of the Ningba Mountain Nature Reserve provided according to an embodiment of the present invention, wherein (a) is a training sample set and (b) is a multi-temporal test sample set. Figure 4 It can be seen that the training sample set and the multi-temporal test sample set are evenly distributed in the protected area and surrounding areas, covering five land cover types: forest, shrub, grassland, building and green space, and water body. The sample distribution has strong representativeness and spatial balance, providing a reliable data foundation for the training and validation of the subsequent land cover change detection model.

[0066] Figure 5 This is a comparative map of land cover classification results for the Ningba Mountain Nature Reserve over multiple years (2013, 2016, 2019, and 2023). From... Figure 5 It can be seen that the forest cover area within the reserve exhibits certain dynamic changes over the years, with some areas transitioning from forest to shrubs or grassland, while buildings and green spaces are mainly distributed on the periphery of the reserve. This classification result is largely consistent with the actual situation, verifying the effectiveness and accuracy of the method of this invention in detecting annual land cover changes in complex geographical environments.

[0067] The land cover change detection device based on texture feature enhancement provided by the present invention is described below. The land cover change detection device based on texture feature enhancement described below can be referred to in correspondence with the land cover change detection method based on texture feature enhancement described above. Figure 6 As shown, the land cover change detection device based on texture feature enhancement includes: Module 10 is used to acquire remote sensing image datasets of the area to be monitored; The calculation module 20 is used to extract the original spectral bands and calculate the vegetation index bands based on the remote sensing image dataset. Extraction module 30 is used to extract the contrast texture feature bands of the vegetation index bands based on the gray-level co-occurrence matrix algorithm; The construction module 40 is used to construct a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, and to perform continuous change detection on the multidimensional input dataset to obtain time series model parameters. The output module 50 is used to output annual land cover classification results based on the multidimensional input dataset and the time series model parameters through the land cover change detection model.

[0068] Optionally, the acquisition module 10 is configured to: Obtain the initial remote sensing image dataset of the area to be monitored; The boundary range of the area to be monitored is determined in the form of an expanded buffer zone; Based on the boundary range, the initial remote sensing image dataset is preprocessed by cloud removal using the CFMAS masking algorithm to generate a new remote sensing image dataset.

[0069] Optionally, the extraction module 30 is used for: The floating-point pixel values ​​of the vegetation index band are linearly mapped to a preset gray level range to obtain the target vegetation index band. A sliding window is set up and traversed pixel by pixel on the target vegetation index band to generate multiple local sub-windows; For each local sub-window, a multi-directional gray-level co-occurrence matrix is ​​calculated using the gray-level co-occurrence matrix algorithm, and a contrast feature texture value is calculated based on the multi-directional gray-level co-occurrence matrix. Based on the contrast feature texture values ​​of each local sub-window, the contrast texture feature band of the vegetation index band is determined.

[0070] Optionally, the building module 40 is configured to: The step of performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters includes: The multidimensional input dataset is input into the continuous change detection algorithm to fit the pixel time series trajectory; The intercept term, trend term, sine and cosine harmonic coefficients of each order, and root mean square error are extracted from the pixel time series trajectory and used as parameters of the time series model.

[0071] Optionally, the output module 50 is used for: A land cover change detection model was constructed using the random forest algorithm; Obtain the auxiliary environmental factors of the area to be monitored; The time series model parameters, the multidimensional input dataset, and the auxiliary environmental factors are fused using multidimensional features to obtain fused features; The fused features are input into the land cover change detection model, and the annual land cover classification results are output.

[0072] Optionally, the output module 50 is used for: Obtain the actual land cover data of the area to be monitored; The annual land cover classification results are compared with the actual land cover data to calculate the overall accuracy and Kappa coefficient; The classification accuracy of the annual land cover classification results is evaluated based on the overall accuracy and the Kappa coefficient.

[0073] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communications interface 720, a memory 730, and a communication bus 740, wherein the processor 710, communications interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a land cover change detection method based on texture feature enhancement. This method includes: acquiring a remote sensing image dataset of the area to be monitored; extracting original spectral bands and calculating vegetation index bands based on the remote sensing image dataset; extracting contrast texture feature bands of the vegetation index bands based on a gray-level co-occurrence matrix algorithm; constructing a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, and performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters; and outputting annual land cover classification results through a land cover change detection model based on the multidimensional input dataset and the time-series model parameters.

[0074] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0075] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the land cover change detection method based on texture feature enhancement provided by the above methods. The method includes: acquiring a remote sensing image dataset of the area to be monitored; extracting original spectral bands and calculating vegetation index bands based on the remote sensing image dataset; extracting contrast texture feature bands of the vegetation index bands based on a gray-level co-occurrence matrix algorithm; constructing a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, and performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters; and outputting annual land cover classification results through a land cover change detection model based on the multidimensional input dataset and the time-series model parameters.

[0076] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the land cover change detection method based on texture feature enhancement provided by the above methods. The method includes: acquiring a remote sensing image dataset of the area to be monitored; extracting original spectral bands and calculating vegetation index bands based on the remote sensing image dataset; extracting contrast texture feature bands of the vegetation index bands based on a gray-level co-occurrence matrix algorithm; constructing a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, and performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters; and outputting annual land cover classification results through a land cover change detection model based on the multidimensional input dataset and the time-series model parameters.

[0077] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0078] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A land cover change detection method based on texture feature enhancement, characterized in that, include: Obtain the remote sensing image dataset of the area to be monitored; Based on the remote sensing image dataset, the original spectral bands are extracted, and the vegetation index bands are calculated. Based on the gray-level co-occurrence matrix algorithm, the contrast texture feature bands of the vegetation index bands are extracted; Based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, a multidimensional input dataset is constructed, and continuous change detection is performed on the multidimensional input dataset to obtain time series model parameters. Based on the multidimensional input dataset and the time-series model parameters, the land cover change detection model outputs annual land cover classification results.

2. The land cover change detection method based on texture feature enhancement according to claim 1, characterized in that, The acquisition of the remote sensing image dataset of the area to be monitored includes: Obtain the initial remote sensing image dataset of the area to be monitored; The boundary range of the area to be monitored is determined in the form of an expanded buffer zone; Based on the boundary range, the initial remote sensing image dataset is preprocessed by cloud removal using the CFMAS masking algorithm to generate a new remote sensing image dataset.

3. The land cover change detection method based on texture feature enhancement according to claim 1, characterized in that, The gray-level co-occurrence matrix algorithm extracts the contrast texture feature bands of the vegetation index bands, including: The floating-point pixel values ​​of the vegetation index band are linearly mapped to a preset gray level range to obtain the target vegetation index band. A sliding window is set up and traversed pixel by pixel on the target vegetation index band to generate multiple local sub-windows; For each local sub-window, a multi-directional gray-level co-occurrence matrix is ​​calculated using the gray-level co-occurrence matrix algorithm, and a contrast feature texture value is calculated based on the multi-directional gray-level co-occurrence matrix. Based on the contrast feature texture values ​​of each local sub-window, the contrast texture feature band of the vegetation index band is determined.

4. The land cover change detection method based on texture feature enhancement according to claim 1, characterized in that, The step of performing continuous change detection on the multidimensional input dataset to obtain time-series model parameters includes: The multidimensional input dataset is input into the continuous change detection algorithm to fit the pixel time series trajectory; The intercept term, trend term, sine and cosine harmonic coefficients of each order, and root mean square error are extracted from the pixel time series trajectory and used as parameters of the time series model.

5. The land cover change detection method based on texture feature enhancement according to claim 1, characterized in that, The process of outputting annual land cover classification results through the land cover change detection model based on the multidimensional input dataset and the time-series model parameters includes: A land cover change detection model was constructed using the random forest algorithm; Obtain the auxiliary environmental factors of the area to be monitored; The time series model parameters, the multidimensional input dataset, and the auxiliary environmental factors are fused using multidimensional features to obtain fused features; The fused features are input into the land cover change detection model, and the annual land cover classification results are output.

6. The land cover change detection method based on texture feature enhancement according to claim 1, characterized in that, Also includes: Obtain the actual land cover data of the area to be monitored; The annual land cover classification results are compared with the actual land cover data to calculate the overall accuracy and Kappa coefficient; The classification accuracy of the annual land cover classification results is evaluated based on the overall accuracy and the Kappa coefficient.

7. A land cover change detection device based on texture feature enhancement, characterized in that, include: The acquisition module is used to acquire remote sensing image datasets of the area to be monitored. The calculation module is used to extract the original spectral bands and calculate the vegetation index bands based on the remote sensing image dataset. The extraction module is used to extract the contrast texture feature bands of the vegetation index bands based on the gray-level co-occurrence matrix algorithm. The construction module is used to construct a multidimensional input dataset based on the original spectral bands, the vegetation index bands, and the contrast texture feature bands, and to perform continuous change detection on the multidimensional input dataset to obtain time series model parameters. The output module is used to output annual land cover classification results based on the multidimensional input dataset and the time series model parameters through the land cover change detection model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the land cover change detection method based on texture feature enhancement as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the land cover change detection method based on texture feature enhancement as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the land cover change detection method based on texture feature enhancement as described in any one of claims 1 to 6.

Images