Land resource dynamic monitoring analysis system and method

Abstract

The invention relates to the technical field of land monitoring, in particular to a land resource dynamic monitoring analysis system and method, and the method comprises the steps: determining the space boundary and time range of a monitoring region, and obtaining the image information of the monitoring region through a satellite; a mixed strategy of'deep learning automatic classification + manual accurate verification 'is adopted, land properties of obtained image information are classified, differential monitoring periods are implemented for land types with different change characteristics, meaningless high-frequency monitoring on low-change-rate areas such as cultivated land fallow areas and mature forest land is avoided, and the land quality is improved. The redundant workload of satellite image acquisition, data processing and manual verification is reduced; meanwhile, monitoring resources are preferentially concentrated in high-change-rate areas such as a construction land development area and an ecologically fragile area, and the overall monitoring cost is reduced on the premise that monitoring full coverage is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of land monitoring technology, and more specifically, to a dynamic monitoring and analysis system and method for land resources. Background Technology

[0002] Land resources, as a core component of national land space, are the fundamental carrier for ensuring food security, maintaining ecological balance, and supporting urban and rural construction and sustainable economic and social development.

[0003] Traditional land resource monitoring methods mainly rely on manual on-site surveys, low-altitude photogrammetry, and conventional remote sensing technology. These methods have many inherent limitations and are difficult to adapt to the needs of modern monitoring and management. In terms of monitoring methods, manual surveys rely on on-site verification and mapping by staff, which not only has limited coverage and high costs in terms of manpower and resources, but also has a long monitoring cycle (usually on an annual or semi-annual basis). This makes it impossible to capture short-term land use changes in a timely manner. Furthermore, there are monitoring blind spots in remote mountainous areas, ecological protection areas, and other areas with inconvenient transportation, which can easily lead to the untimely detection of problems such as illegal land use and ecological damage.

[0004] To address the aforementioned issues, there is an urgent need for dynamic monitoring and analysis methods for land resources. Summary of the Invention

[0005] The purpose of this invention is to provide a method for dynamic monitoring and analysis of land resources to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, a method for dynamic monitoring and analysis of land resources is provided, including the following steps: S1. Determine the spatial boundaries and time range of the monitoring area, and acquire image information of the monitoring area via satellite; S2. A hybrid strategy of "automatic classification by deep learning + precise manual verification" is adopted to classify the land properties of the acquired image information; S3. Intelligent differentiated monitoring cycle setting: Based on the land type change rate, a dynamic cycle adjustment model is established to achieve adaptive switching between routine monitoring, key monitoring and emergency monitoring. S4. Set a threshold for the rate of change of land type, acquire satellite imagery data of the monitored area in real time, and combine it with drone aerial photography. If the rate of change of land type exceeds the set threshold, activate the emergency monitoring mode to quickly capture the scope of disaster impact and the damage to land use, and provide data support for emergency response and post-disaster recovery. S5. Multi-dimensional dynamic change detection and analysis: Through classification and comparison combined with spectral difference verification, land use change information is accurately extracted, and change attribution analysis and trend prediction are completed.

[0007] As a further improvement to this technical solution, the acquisition of monitoring area image information via satellite in step S1 includes: S1.1 The spatial boundary of the monitoring area is precisely delineated using latitude and longitude coordinates and positioned using the WGS-84 coordinate system; S1.2 Select at least two types of satellite data sources with different resolutions to construct a collaborative dataset. Perform geometric fine correction, image cropping and stitching, and adaptive enhancement processing on the acquired satellite images in sequence to obtain standardized preprocessed images.

[0008] As a further improvement to this technical solution, step S1, which involves acquiring image information of the monitoring area via satellite, includes: S1.1 The spatial boundary of the monitoring area is precisely delineated using latitude and longitude coordinates and positioned using the WGS-84 coordinate system; S1.2 Select at least two types of satellite data sources with different resolutions to construct a collaborative dataset. Perform geometric fine correction, image cropping and stitching, and adaptive enhancement processing on the acquired satellite images in sequence to obtain standardized preprocessed images.

[0009] As a further improvement to this technical solution, step S2 involves classifying the land properties of the acquired image information, including: S2.1 Based on the national standard "Classification of Current Land Use", and combined with monitoring needs, we divide land into primary and secondary land types and construct a hierarchical classification system; S2.2 Feature extraction: Extract spectral features, texture features, spatial features, and temporal features from standardized preprocessed images to construct a comprehensive feature set for accurate identification of land resources in the monitoring area; S2.3. The improved U-Net model is used to train and classify the comprehensive feature set, and the preliminary land classification results are output. S2.4. Precise manual verification: The classification results are optimized through precision verification and manual correction to obtain the final land classification map.

[0010] As a further improvement to this technical solution, section 2.3 employs an improved U-Net model to train and classify the comprehensive feature set, including: Sample preprocessing involves dividing the jointly constructed training samples into training, validation, and test sets in a 7:2:1 ratio. Data augmentation processing is performed on the samples by random cropping, flipping, and rotation. The sample size is uniformly adjusted to 256×256 pixels to meet the model input requirements. The model initialization process involves an improved U-Net model comprising an encoder, a decoder, and an attention fusion module. The encoder uses four convolutional blocks to extract features, each containing two 3×3 convolutional layers, one batch normalization layer, and a ReLU activation function, along with a 2×2 max pooling layer for downsampling. The decoder uses four deconvolutional blocks for upsampling, each containing one 2×2 deconvolutional layer, a feature concatenation operation, and two 3×3 convolutional layers. For model training, an adaptive learning rate optimization algorithm was used. The initial learning rate was set to 0.001, the weight decay coefficient was 0.0001, and the loss function was a weighted sum of cross-entropy loss and Dice loss (with weight ratios of 0.6 and 0.4, respectively). The number of training iterations was set to 200. Every 50 iterations, the learning rate was decayed to 1 / 10 of its original value. Every 10 iterations, the model performance was evaluated using a validation set, and the optimal model with the minimum validation set loss was saved. The classification reasoning process inputs the preprocessed comprehensive feature set into the optimal model, uses the softmax function to output probability maps of various land features, and uses a probability threshold of 0.5 as the judgment standard. If the probability is ≥0.5, it is judged as the corresponding land feature type, and if it is <0.5, it is classified into the undetermined area, and the preliminary land classification results are output. The classification results are then precisely verified manually, and the final land classification map is obtained through accuracy verification and manual correction.

[0011] As a further improvement to this technical solution, the intelligent differentiated monitoring cycle setting in S3 includes: S3.1 Based on the land use classification results completed in S2, clarify the functional attributes, control requirements and historical change characteristics of each land type; S3.2 Collect multi-source data such as satellite imagery data, land use change ledgers, and field survey records of the monitoring area for the past 5 years. Calculate the average annual change rate, quarterly fluctuation range, and frequency of abrupt changes for different land types using time-series change analysis algorithms, and establish a baseline database of change rates for each land type. S3.3 Parameter Construction and Weight Assignment for the Dynamic Periodic Adjustment Model: Using the baseline of change rate as a reference, a negative correlation mapping relationship is established between the monitoring period and the change rate. That is, the higher the land type change rate, the shorter the monitoring period; the lower the change rate, the longer the monitoring period. An example of the function expression is as follows: In the formula: The monitoring period is for the i-th type of land; This serves as the baseline monitoring period for this type of land; Let be the baseline rate of change for the i-th type of land; Let be the real-time rate of change of the i-th type of land; Let be the weight of the j-th auxiliary parameter; Let be the score value of the i-th type of land under the j-th auxiliary parameter; As a further improvement to this technical solution, based on the calculation results of the dynamic periodic adjustment model and combined with the actual needs of land management, three types of monitoring modes are divided into regular monitoring mode, key monitoring mode and emergency monitoring preparation mode, and corresponding periodic threshold ranges are set.

[0012] As a further improvement to this technical solution, the emergency monitoring mode is activated in S4, including: S4.1 Based on the baseline database of land change rates for each land type established in S3, and combined with the land control level of the monitoring area, the results of ecological vulnerability assessment, and historical disaster case data, emergency threshold intervals are divided, and the response levels corresponding to different thresholds are clarified. S4.2. The latest satellite image data of the monitoring area is acquired in real time through the satellite remote sensing data receiving system, and combined with the drone aerial image data to form a real-time monitoring dataset with multi-source data fusion. S4.3. Call the land type classification algorithm of "deep learning automatic classification + manual accurate verification" in S2 to quickly classify the land type of the real-time monitoring dataset, calculate the current land type change rate through time series difference method, and generate a real-time change rate monitoring report. S4.4 Establish an automatic threshold comparison engine to match the real-time change rate with the hierarchical threshold system one by one to determine whether the emergency triggering conditions have been met.

[0013] As a further improvement to this technical solution, S5 accurately extracts land use change information and completes change attribution analysis and trend prediction, including: S5.1 Collects multi-period satellite imagery data during the monitoring period, and dynamically monitors topographic data, meteorological data, and human activity records during the monitoring period; S5.2 Spatially overlay the land use classification maps of the baseline period and the monitoring period, extract the pixels where the classification categories have changed, and form a preliminary land use change map; S5.3 Select the spectral bands of land type characteristics, construct a multi-band spectral difference model, calculate the spectral difference of corresponding pixels between the baseline period and the monitoring period, and determine the spectral difference threshold range of each change type based on the differences in spectral characteristics of different land types through statistical analysis methods. S5.4 Quantify the correlation between each driving factor and the type and magnitude of land use change, identify the dominant driving factors of land change in different regions, and combine GIS spatial analysis technology to overlay the dominant driving factors with land use change patches to generate a thematic map of change attribution. S5.5 uses multi-period land use change data and driving factor data to select a time series prediction model. It trains the model using historical monitoring data, meteorological data, and human activity records. The calibrated model is then used to predict future land use change trends and output land use prediction maps for each scenario.

[0014] The second objective of this invention is to provide a dynamic monitoring and analysis system for land resources, including any one of the land resource dynamic monitoring and analysis methods described above, comprising an image acquisition unit, a classification unit, a dynamic monitoring unit, an emergency response unit, and a prediction unit; The image acquisition unit is used to determine the spatial boundary and time range of the monitoring area and to acquire image information of the monitoring area via satellite. The classification unit adopts a hybrid strategy of "deep learning automatic classification + manual precise verification" to classify the land properties of the acquired image information; The dynamic monitoring unit intelligently sets a differentiated monitoring cycle, establishes a dynamic cycle adjustment model based on the land type change rate, and realizes adaptive switching between routine monitoring, key monitoring and emergency monitoring. The emergency unit sets a land type change rate threshold, dynamically acquires satellite image data of the monitoring area in real time, and combines it with drone aerial photography. If the land type change rate exceeds the set threshold, the emergency monitoring mode is activated. The prediction unit accurately extracts land use change information by detecting and analyzing multi-dimensional dynamic changes, and by combining classification comparison with spectral difference verification, thereby completing change attribution analysis and trend prediction.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this dynamic monitoring and analysis method for land resources, differentiated monitoring cycles are implemented for land types with different change characteristics. This avoids meaningless high-frequency monitoring of low-change-rate areas such as fallow farmland and mature forest land, reducing redundant workload in satellite image acquisition, data processing, and manual verification. At the same time, monitoring resources are prioritized for high-change-rate areas such as construction land development zones and ecologically fragile areas, thereby reducing overall monitoring costs while ensuring full monitoring coverage.

[0016] 2. This dynamic monitoring and analysis method for land resources establishes a negative correlation between the monitoring cycle and the rate of change, significantly increasing the monitoring frequency in areas with high rates of change. This allows for real-time capture of the dynamic evolution of land use and timely detection of abnormal changes such as illegal development and ecological damage. Compared to fixed-cycle monitoring, it can advance the detection time of abnormal changes, providing accurate time-series data support for land law enforcement and ecological management, and effectively avoiding the problem of "change omissions" caused by excessively long monitoring cycles. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the land resource dynamic monitoring and analysis method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, a method for dynamic monitoring and analysis of land resources is provided, including the following steps: S1. Determine the spatial boundaries and time range of the monitoring area, and acquire image information of the monitoring area via satellite; In the process of dynamic analysis of land resources, it is necessary to acquire real-time land images to perform dynamic analysis of the land. Therefore, the acquisition of monitoring area image information via satellite in step S1 includes: S1.1 The spatial boundary of the monitoring area is precisely delineated using latitude and longitude coordinates and positioned using the WGS-84 coordinate system; S1.2 Select at least two types of satellite data sources with different resolutions to construct a collaborative dataset. Perform geometric fine correction, image cropping and stitching, and adaptive enhancement processing on the acquired satellite images in sequence to obtain standardized preprocessed images.

[0020] The spatial boundary of the monitoring area is precisely delineated using latitude and longitude coordinates and located using the WGS-84 coordinate system. This precise boundary and time definition allows for targeted locking of monitoring areas, avoiding interference from redundant data, reducing subsequent data processing volume, and improving overall monitoring efficiency. Furthermore, the clearly defined time dimension lays the foundation for constructing time-series datasets and achieving dynamic change monitoring. The acquired satellite imagery undergoes preprocessing to remove blurred details. The satellite-acquired imagery must meet the requirements of cloud cover control within 8%, absence of severe noise, and data incompleteness rate ≤5%. Furthermore, the radiometric consistency and spatial alignment accuracy of the standardized preprocessed images meet the needs of subsequent classification and change detection. Satellite acquisition breaks through the limitations of traditional manual surveys, enabling large-scale, blind-spot-free coverage. It is particularly suitable for remote mountainous areas, ecological reserves, and other areas with inconvenient transportation, ensuring the integrity of the monitoring range. At the same time, by combining information acquired from multi-resolution satellite data sources, it takes into account both macroscopic coverage and microscopic details, meeting the overall monitoring needs of large-scale areas while capturing image information of small plots and fine features, providing comprehensive data support for subsequent classification and change detection.

[0021] S2. A hybrid strategy of "automatic classification by deep learning + precise manual verification" is adopted to classify the land properties of the acquired image information; Considering the different land properties, when acquiring real-time land images and performing dynamic analysis, in order to facilitate the acquisition of information on changes in land resources, step S2 categorizes the land properties of the acquired image information, including: S2.1 Based on the national standard "Classification of Current Land Use", and in combination with monitoring needs, a hierarchical classification system is constructed by dividing land into primary and secondary types. The hierarchical classification system is constructed by dividing land into primary and secondary types based on the national standard "Classification of Current Land Use" and in combination with monitoring needs, and clarifying the definition, spectral characteristics and judgment criteria of each type of land. S2.2 Feature extraction: Extract spectral features, texture features, spatial features, and temporal features from standardized preprocessed images to construct a comprehensive feature set for accurate identification of land resources in the monitoring area; S2.3. An improved U-Net model is used to train and classify the comprehensive feature set, and preliminary land classification results are output. Through the hybrid strategy of "automatic classification by deep learning + accurate manual verification", large-scale multi-source images can be processed quickly, greatly reducing the workload and time cost of manual classification, and solving the problems of low efficiency and difficulty in adapting to large-scale monitoring in traditional manual classification.

[0022] Section 2.3 employs an improved U-Net model for training and classification of the comprehensive feature set, including: Sample preprocessing involves dividing the jointly constructed training samples into training, validation, and test sets in a 7:2:1 ratio. Data augmentation processing is performed on the samples by random cropping, flipping, and rotation. The sample size is uniformly adjusted to 256×256 pixels to meet the model input requirements. The model initialization process involves an improved U-Net model comprising an encoder, a decoder, and an attention fusion module. The encoder uses four convolutional blocks to extract features, each containing two 3×3 convolutional layers, one batch normalization layer, and a ReLU activation function, along with a 2×2 max pooling layer for downsampling. The decoder uses four deconvolutional blocks for upsampling, each containing one 2×2 deconvolutional layer, a feature concatenation operation, and two 3×3 convolutional layers. For model training, the Adaptive Learning Rate Optimization (AdamW) algorithm was used. The initial learning rate was set to 0.001, the weight decay coefficient was 0.0001, and the loss function was a weighted sum of cross-entropy loss and Dice loss (with weight ratios of 0.6 and 0.4, respectively). The number of training iterations was set to 200. Every 50 iterations, the learning rate was decayed to 1 / 10 of its original value. Every 10 iterations, the model performance was evaluated using a validation set, and the optimal model with the minimum validation set loss was saved. The classification reasoning process inputs the preprocessed comprehensive feature set into the optimal model, uses the softmax function to output probability maps of various land features, and uses a probability threshold of 0.5 as the judgment standard. If the probability is ≥0.5, it is judged as the corresponding land feature type, and if it is <0.5, it is classified into the undetermined area, and the preliminary land classification results are output. By combining the improved U-Net model with a multi-dimensional feature set, compared with traditional machine learning algorithms, it can accurately capture subtle differences in the edges, textures, and spectra of land features, effectively distinguishing between spectrally similar land features such as dry land and bare land, and urban residential land and industrial land.

[0023] S2.4. Precise manual verification: The classification results are optimized through precision verification and manual correction to obtain the final land classification map.

[0024] Human verification compensates for the limitations of the algorithm. In areas of ambiguity such as the boundary between land categories and areas where the algorithm makes a mistake, human intervention is used to correct these errors, ensuring the accuracy and reliability of the classification results and avoiding systematic errors caused by a single algorithm.

[0025] S3. Intelligent differentiated monitoring cycle setting: Based on the land type change rate, a dynamic cycle adjustment model is established to achieve adaptive switching between routine monitoring, key monitoring and emergency monitoring. In the process of monitoring land resources, due to the different rates of change in land types, it is necessary to monitor different land types. Therefore, the intelligent differentiated monitoring cycle settings in S3 include: S3.1 Based on the land use classification results completed in S2, the land use types (such as cultivated land, forest land, construction land, ecological protection area, industrial and mining land, etc.) are divided into layers to clarify the functional attributes, control requirements and historical change characteristics of each land type; Differentiated monitoring cycles should be implemented for land types with different change characteristics to avoid meaningless high-frequency monitoring of low-change-rate areas such as fallow farmland and mature forest land, thereby reducing redundant workload in satellite image acquisition, data processing, and manual verification. At the same time, monitoring resources should be prioritized for high-change-rate areas such as construction land development zones and ecologically fragile areas, thereby reducing overall monitoring costs while ensuring full monitoring coverage.

[0026] S3.2 Collect multi-source data such as satellite imagery data, land use change ledgers, and field survey records of the monitoring area for the past 5 years. Calculate the average annual change rate, quarterly fluctuation range, and frequency of abrupt changes for different land types using time-series change analysis algorithms, and establish a baseline database of change rates for each land type. Establishing a negative correlation between monitoring cycle and rate of change significantly increases the monitoring frequency in areas with high rates of change, enabling real-time capture of the dynamic evolution of land use and timely detection of abnormal changes such as illegal development and ecological damage. Compared to fixed-cycle monitoring, it can advance the detection time of abnormal changes, providing accurate time-series data support for land law enforcement and ecological management, and effectively avoiding the problems of "change omission" or "delayed handling" caused by excessively long monitoring cycles.

[0027] Among them, the rate of change baseline includes two core indicators: the baseline rate of change (the rate of natural or human-induced alteration under normal conditions) and the critical rate of change (the rate threshold that may trigger ecological or management risks). The rate of change of land type was selected as the core driving parameter, while auxiliary influencing parameters were also included, including the ecological sensitivity of land type (e.g., high sensitivity of ecological red line areas and basic farmland protection areas), regional development planning (e.g., high development intensity of new urban construction areas and industrial parks), geographical vulnerability (e.g., weak resistance to interference in landslide-prone areas and wetland degradation areas), and intensity of human activities. The weights of each input parameter were assigned based on the analytic hierarchy process.

[0028] S3.3 Parameter Construction and Weight Assignment for the Dynamic Periodic Adjustment Model: Using the baseline of change rate as a reference, a negative correlation mapping relationship is established between the monitoring period and the change rate. That is, the higher the land type change rate, the shorter the monitoring period; the lower the change rate, the longer the monitoring period. An example of the function expression is as follows: In the formula: The monitoring period is for the i-th type of land; This serves as the baseline monitoring period for this type of land; Let be the baseline rate of change for the i-th type of land; Let be the real-time rate of change of the i-th type of land; Let be the weight of the j-th auxiliary parameter; Let be the score value of the i-th type of land under the j-th auxiliary parameter; Based on the calculation results of the dynamic cycle adjustment model and combined with the actual needs of land management, three types of monitoring modes are divided into regular monitoring mode, key monitoring mode and emergency monitoring preparation mode, and corresponding cycle threshold ranges are set. Conventional monitoring mode: applicable to areas where the rate of change of land type is ≤ the baseline rate of change, with a monitoring cycle of T-2T. For example, stable areas such as fallow farmland and mature forest land can be monitored quarterly or semi-annually to save monitoring costs.

[0029] Key monitoring mode: Applicable to areas where the baseline rate of change is less than the rate of change of land type and less than or equal to the critical rate of change. The monitoring period is T / 2-T. For example, active areas such as construction land within the boundary of new town development and forest land in ecological restoration areas are monitored monthly or bi-monthly to track the trend of change in real time.

[0030] Emergency monitoring preparation mode: Applicable to areas where the rate of land type change is greater than the critical rate of change, the monitoring cycle is shortened to T / 4-T, such as risk warning areas such as industrial and mining land expansion areas and wetland degradation areas. High-frequency monitoring is initiated and warning information is pushed out simultaneously to prepare for switching to emergency monitoring mode; Based on the rate of change baseline database and dynamic adjustment model, the system can smoothly switch between regular, key, and emergency monitoring modes without manual intervention: when the land change rate is within the baseline range, it automatically maintains the regular monitoring mode; when the rate exceeds the baseline threshold, it automatically upgrades to key monitoring; and when the rate approaches the critical threshold, it automatically switches to emergency monitoring standby mode, enhancing the system's flexibility.

[0031] S4. Set a threshold for the rate of land type change, acquire satellite imagery data of the monitoring area in real time, and combine it with drone aerial photography. If the rate of land type change exceeds the set threshold, activate the emergency monitoring mode to quickly capture the scope of disaster impact and land use damage, and provide data support for emergency response and post-disaster recovery. When monitoring land resources, in the event of illegal mining or natural disasters, it is necessary to quickly capture the scope of the disaster's impact and the extent of land use damage to provide data support for emergency response and post-disaster recovery. Therefore, activating the emergency monitoring mode in S4 includes: S4.1 Based on the baseline database of land change rates for each land type established in S3, and combined with the land control level, ecological vulnerability assessment results, and historical disaster case data of the monitoring area, emergency threshold intervals are delineated, and the response levels corresponding to different thresholds are clarified, including: Level I threshold: The rate of change in land type reaches 1.2-1.5 times the critical rate of change, corresponding to a potential risk warning, and a mechanism to increase the monitoring frequency needs to be activated; Level II threshold: The rate of change reaches 1.5-2.0 times the critical rate of change, corresponding to a moderate risk event (such as small-scale illegal development or precursors to local geological disasters), and the emergency monitoring mode needs to be officially activated. Level III threshold: The rate of change exceeds 2.0 times the critical rate of change, corresponding to major risk events (such as large-scale landslides, floods, and illegal land occupation and development), requiring the activation of the highest priority emergency response and the coordination of data sharing among multiple departments; S4.2. The latest satellite image data of the monitoring area is acquired in real time through the satellite remote sensing data receiving system, and combined with the drone aerial image data to form a real-time monitoring dataset with multi-source data fusion. S4.3. Call the land type classification algorithm of "deep learning automatic classification + manual accurate verification" in S2 to quickly classify the land type of the real-time monitoring dataset, calculate the current land type change rate through time series difference method, and generate a real-time change rate monitoring report. S4.4 Establish an automatic threshold comparison engine to match the real-time change rate with the hierarchical threshold system one by one to determine whether the emergency triggering conditions have been met.

[0032] S5. Multi-dimensional dynamic change detection and analysis: Through classification and comparison combined with spectral difference verification, land use change information is accurately extracted, and change attribution analysis and trend prediction are completed. The multi-source data fusion strategy of "satellite remote sensing + UAV aerial photography" compensates for the shortcomings of single satellite remote sensing data in terms of monitoring frequency and spatial resolution. Satellite remote sensing achieves full coverage of the monitoring area, while UAV aerial photography focuses on areas exceeding the threshold, providing centimeter-level high-precision image data. The real-time monitoring dataset formed by the combination of the two can quickly capture the spatial distribution characteristics of land damage after an emergency occurs. Compared with traditional single-source monitoring, data acquisition efficiency is improved, buying valuable time for emergency response decision-making.

[0033] By using a satellite remote sensing data receiving system, the latest satellite imagery data of the monitored area is acquired in real time. This data is then combined with high-frequency data collected by drone aerial photography and ground-based IoT sensors (such as displacement sensors and soil erosion monitors) to form a real-time monitoring dataset that integrates multiple data sources. A baseline database of change rates is constructed, and multi-level emergency threshold ranges are defined based on land control levels, ecological vulnerability, and historical disaster data. This approach overcomes the limitations of traditional monitoring's "indiscriminate triggering." For sensitive areas such as ecological red line zones and basic farmland protection zones, risk can be detected in advance by adjusting the thresholds. Differentiated response levels are matched to different change rate thresholds, avoiding resource waste caused by "over-response" and risk spread caused by "under-response," thus ensuring the scientific and accurate initiation of emergency monitoring.

[0034] S5 accurately extracts land use change information and completes change attribution analysis and trend prediction, including: After obtaining information on changes in land resources, in order to facilitate land prediction and future planning, S5.1 collects multi-period satellite imagery data during the monitoring period, as well as topographic data, meteorological data, and human activity records during the dynamic monitoring period. S5.2 Spatially overlay the land use classification maps of the baseline period and the monitoring period, extract the pixels whose classification categories have changed, and form a preliminary land use change map. The pixel-by-pixel overlay analysis algorithm is used to spatially overlay the land use classification maps of the baseline period and the monitoring period, extract the pixels whose classification categories have changed, and form a preliminary land use change map. S5.3 Select the spectral bands of land type characteristics, construct a multi-band spectral difference model, calculate the spectral difference of corresponding pixels between the baseline period and the monitoring period, and determine the spectral difference threshold range of each change type based on the differences in spectral characteristics of different land types through statistical analysis methods. By generating land use forecast maps for the next 5-10 years, it is possible to predict potential ecological risk areas (such as wetland degradation and farmland loss) and development potential areas (such as industrial park expansion and ecological restoration), helping land management departments shift from "passive response" to "proactive management" and achieve sustainable use of land resources.

[0035] S5.4 Quantify the correlation between each driving factor and the type and magnitude of land use change, identify the dominant driving factors of land change in different regions, and combine GIS spatial analysis technology to overlay the dominant driving factors with land use change patches to generate a thematic map of change attribution. S5.5 uses multi-period land use change data and driving factor data to select a time series prediction model. It trains the model using historical monitoring data, meteorological data, and human activity records. The calibrated model is then used to predict future land use change trends and output land use prediction maps for each scenario.

[0036] By combining GIS spatial analysis technology, high-precision land use change information maps, change attribution thematic maps, and trend prediction maps are generated, transforming abstract data analysis results into intuitive visualizations. These thematic maps clearly show the spatial distribution of change patches, the influence range of driving factors, and future change trends, which not only facilitates land management departments to quickly grasp the dynamics of regional land resources but also improves the collaborative efficiency of land resource management.

[0037] The second objective of this invention is to provide a dynamic monitoring and analysis system for land resources, including any one of the land resource dynamic monitoring and analysis methods described above, comprising an image acquisition unit, a classification unit, a dynamic monitoring unit, an emergency response unit, and a prediction unit; The image acquisition unit is used to determine the spatial boundary and time range of the monitoring area and to acquire image information of the monitoring area via satellite. The classification unit employs a hybrid strategy of "automatic classification using deep learning + precise manual verification" to classify the land properties of the acquired image information. The dynamic monitoring unit intelligently sets a differentiated monitoring cycle. Based on the land type change rate, a dynamic cycle adjustment model is established to achieve adaptive switching between routine monitoring, key monitoring and emergency monitoring. The emergency unit sets a threshold for the rate of land type change, dynamically acquires satellite imagery data of the monitored area in real time, and combines it with drone aerial photography. If the rate of land type change exceeds the set threshold, the emergency monitoring mode is activated. The prediction unit accurately extracts land use change information by detecting and analyzing multi-dimensional dynamic changes, and by combining classification comparison with spectral difference verification, thus completing change attribution analysis and trend prediction.

[0038] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic monitoring and analysis of land resources, characterized in that: Includes the following steps: S1. Determine the spatial boundaries and time range of the monitoring area, and acquire image information of the monitoring area via satellite; S2. A hybrid strategy of "automatic classification by deep learning + precise manual verification" is adopted to classify the land properties of the acquired image information; S3. Intelligent differentiated monitoring cycle setting: Based on the land type change rate, a dynamic cycle adjustment model is established to achieve adaptive switching between routine monitoring, key monitoring and emergency monitoring. S4. Set a land type change rate threshold, dynamically acquire satellite imagery data of the monitoring area in real time, and combine it with drone aerial photography. If the land type change rate exceeds the set threshold, activate the emergency monitoring mode. S5. Multi-dimensional dynamic change detection and analysis: Through classification and comparison combined with spectral difference verification, land use change information is accurately extracted, and change attribution analysis and trend prediction are completed.

2. The method for dynamic monitoring and analysis of land resources according to claim 1, characterized in that: The step S1, which involves acquiring image information of the monitored area via satellite, includes: S1.1 The spatial boundary of the monitoring area is precisely delineated using latitude and longitude coordinates and positioned using the WGS-84 coordinate system; S1.2 Select at least two types of satellite data sources with different resolutions to construct a collaborative dataset. Perform geometric fine correction, image cropping and stitching, and adaptive enhancement processing on the acquired satellite images in sequence to obtain standardized preprocessed images.

3. The method for dynamic monitoring and analysis of land resources according to claim 1, characterized in that: The S2 step involves classifying the land properties of the acquired image information, including: S2.1 Based on the national standard "Classification of Current Land Use", and combined with monitoring needs, we divide land into primary and secondary land types and construct a hierarchical classification system; S2.2 Feature extraction: Extract spectral features, texture features, spatial features, and temporal features from standardized preprocessed images to construct a comprehensive feature set for accurate identification of land resources in the monitoring area; S2.

3. The improved U-Net model is used to train and classify the comprehensive feature set, and the preliminary land classification results are output. S2.

4. Precise manual verification: The classification results are optimized through precision verification and manual correction to obtain the final land classification map.

4. The land resource dynamic monitoring and analysis method according to claim 3, characterized in that: Section 2.3 employs an improved U-Net model for training and classification of the comprehensive feature set, including: Sample preprocessing involves dividing the jointly constructed training samples into training, validation, and test sets in a 7:2:1 ratio. Data augmentation processing is performed on the samples by random cropping, flipping, and rotation. The sample size is uniformly adjusted to 256×256 pixels to meet the model input requirements. The model initialization process involves an improved U-Net model comprising an encoder, a decoder, and an attention fusion module. The encoder uses four convolutional blocks to extract features, each containing two 3×3 convolutional layers, one batch normalization layer, and a ReLU activation function, along with a 2×2 max pooling layer for downsampling. The decoder uses four deconvolutional blocks for upsampling, each containing one 2×2 deconvolutional layer, a feature concatenation operation, and two 3×3 convolutional layers. For model training, the Adaptive Learning Rate Optimization (AdamW) algorithm was used. The initial learning rate was set to 0.001, the weight decay coefficient was 0.0001, and the loss function was a weighted sum of cross-entropy loss and Dice loss (with weight ratios of 0.6 and 0.4, respectively). The number of training iterations was set to 200. Every 50 iterations, the learning rate was decayed to 1 / 10 of its original value. Every 10 iterations, the model performance was evaluated using a validation set, and the optimal model with the minimum validation set loss was saved. The classification reasoning process inputs the preprocessed comprehensive feature set into the optimal model, uses the softmax function to output probability maps of various land features, and uses a probability threshold of 0.5 as the judgment standard. If the probability is ≥0.5, it is judged as the corresponding land feature type, and if it is <0.5, it is classified into the undetermined area, and the preliminary land classification results are output. The classification results are then precisely verified manually, and the final land classification map is obtained through accuracy verification and manual correction.

5. The method for dynamic monitoring and analysis of land resources according to claim 1, characterized in that: The intelligent differentiated monitoring cycle setting in S3 includes: S3.1 Based on the land use classification results completed in S2, clarify the functional attributes, control requirements and historical change characteristics of each land type; S3.2 Collect multi-source data such as satellite imagery data, land use change ledgers, and field survey records of the monitoring area for the past 5 years. Calculate the average annual change rate, quarterly fluctuation range, and frequency of abrupt changes for different land types using time-series change analysis algorithms, and establish a baseline database of change rates for each land type. S3.3 Parameter Construction and Weight Assignment for the Dynamic Periodic Adjustment Model: Using the baseline of change rate as a reference, a negative correlation mapping relationship is established between the monitoring period and the change rate. That is, the higher the land type change rate, the shorter the monitoring period; the lower the change rate, the longer the monitoring period. An example of the function expression is as follows:

6. In the formula: The monitoring period is for land type i. This serves as the baseline monitoring period for this type of land; Let be the baseline rate of change for the i-th type of land; Let be the real-time rate of change of the i-th type of land; Let be the weight of the j-th auxiliary parameter; Let be the score value of the i-th type of land under the j-th auxiliary parameter.

7. The method for dynamic monitoring and analysis of land resources according to claim 5, characterized in that: Based on the calculation results of the dynamic periodic adjustment model and combined with the actual needs of land management, three types of monitoring modes are divided into regular monitoring mode, key monitoring mode and emergency monitoring preparation mode, and corresponding periodic threshold ranges are set.

8. The method for dynamic monitoring and analysis of land resources according to claim 1, characterized in that: The activation of the emergency monitoring mode in S4 includes: S4.1 Based on the established baseline database of land change rates for each land type, combined with the land control level of the monitoring area, the results of ecological vulnerability assessment, and historical disaster case data, emergency threshold intervals are divided, and the response levels corresponding to different thresholds are clarified. S4.

2. The latest satellite image data of the monitoring area is acquired in real time through the satellite remote sensing data receiving system, and combined with the drone aerial image data to form a real-time monitoring dataset with multi-source data fusion. S4.

3. Call the land type classification algorithm of "deep learning automatic classification + manual accurate verification" in S2 to quickly classify the land type of the real-time monitoring dataset, calculate the current land type change rate through time series difference method, and generate a real-time change rate monitoring report. S4.4 Establish an automatic threshold comparison engine to match the real-time change rate with the hierarchical threshold system one by one to determine whether the emergency triggering conditions have been met.

9. The method for dynamic monitoring and analysis of land resources according to claim 1, characterized in that: S5 accurately extracts land use change information and completes change attribution analysis and trend prediction, including: S5.1 Collects multi-period satellite imagery data during the monitoring period, and dynamically monitors topographic data, meteorological data, and human activity records during the monitoring period; S5.2 Spatially overlay the land use classification maps of the baseline period and the monitoring period, extract the pixels where the classification categories have changed, and form a preliminary land use change map; S5.3 Select the spectral bands of land type characteristics, construct a multi-band spectral difference model, calculate the spectral difference of corresponding pixels between the baseline period and the monitoring period, and determine the spectral difference threshold range of each change type based on the differences in spectral characteristics of different land types through statistical analysis methods. S5.4 Quantify the correlation between each driving factor and the type and magnitude of land use change, identify the dominant driving factors of land change in different regions, and combine GIS spatial analysis technology to overlay the dominant driving factors with land use change patches to generate a thematic map of change attribution. S5.5 uses multi-period land use change data and driving factor data to select a time series prediction model. It trains the model using historical monitoring data, meteorological data, and human activity records. The calibrated model is then used to predict future land use change trends and output land use prediction maps for each scenario.

10. A land resource dynamic monitoring and analysis system, used to execute a land resource dynamic monitoring and analysis method as described in any one of claims 1 to 9, comprising an image acquisition unit, a classification unit, a dynamic monitoring unit, an emergency response unit, and a prediction unit; The image acquisition unit is used to determine the spatial boundary and time range of the monitoring area and to acquire image information of the monitoring area via satellite. The classification unit adopts a hybrid strategy of "deep learning automatic classification + manual precise verification" to classify the land properties of the acquired image information; The dynamic monitoring unit intelligently sets a differentiated monitoring cycle, establishes a dynamic cycle adjustment model based on the land type change rate, and realizes adaptive switching between routine monitoring, key monitoring and emergency monitoring. The emergency unit sets a land type change rate threshold, dynamically acquires satellite image data of the monitoring area in real time, and combines it with drone aerial photography. If the land type change rate exceeds the set threshold, the emergency monitoring mode is activated. The prediction unit accurately extracts land use change information by detecting and analyzing multi-dimensional dynamic changes, and by combining classification comparison with spectral difference verification, thereby completing change attribution analysis and trend prediction.

Images