Land classification extraction method, system and equipment and storage medium
By introducing convolutional deep learning network, circular deep learning network and automatic encoder into the land classification model, combined with GIS technology and active learning methods, the problem of difficulty in fusion of multiple feature information in remote sensing images is solved, and efficient and accurate land use classification is achieved.
Patent Information
- Application Number
- CN202510203750.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-24
- Publication Date
- 2025-06-13
AI Technical Summary
The prior art is difficult to effectively fuse multiple feature information when processing high-resolution remote sensing images, resulting in insufficient classification accuracy and model generalization capabilities.
The land classification extraction model based on convolutional deep learning network, circular deep learning network and automatic encoder is adopted, combined with GIS technology and active learning methods, samples interfered by environmental factors are automatically identified and eliminated, sample sets are dynamically adjusted, and the generalization ability and classification performance of the model are improved through integrated learning and gradient enhancement mechanisms.
It significantly improves the efficiency and accuracy of land use classification of remote sensing images, improves the generalization ability and classification performance of the model, and maintains the explanatory nature of the model.
Smart Images
Figure CN120147702A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of deep learning technology, and particularly to a method, system, device, and storage medium for land classification and extraction. Background Art
[0002] In the field of land use classification, it is mainly accomplished by using satellite remote sensing technology. With the development of remote sensing technology, the amount of image data obtained is increasingly large and complex, which poses challenges to classification accuracy and the generalization ability of the model. For example, in complex land use scenarios, how to accurately distinguish different types of ground objects has become an urgent problem to be solved.
[0003] Traditional machine learning methods: including support vector machines, K-nearest neighbors, etc. These methods usually rely on manual feature engineering to extract useful information from remote sensing images. Although these methods are easy to understand and implement, they are prone to falling into the curse of dimensionality when facing high-dimensional data, and have limited ability to capture non-linear relationships, resulting in poor classification effects. In recent years, deep learning models such as convolutional neural networks (CNNs) have achieved remarkable results in remote sensing image classification due to their strong feature extraction ability and non-linear mapping ability. However, deep learning models usually require a large amount of labeled data for training, and the model structure is complex, the training time is long, and it is easy to cause waste of resources. In addition, deep learning models are often regarded as black box models, lacking transparency and interpretability.
[0004] One of the main problems faced by current land classification methods when dealing with high-resolution remote sensing images is how to effectively fuse multiple feature information to improve classification accuracy. Traditional single-feature-based methods are difficult to capture the complex attributes of ground objects, especially the lack of information in texture details and time series changes, resulting in insufficient generalization ability of the classification model in complex scenarios. Summary of the Invention
[0005] This application provides a method, system, device, and storage medium for land classification and extraction to solve the above problems.
[0006] On the one hand, this application provides a method for land classification and extraction, and the method includes the following steps:
[0007] Step S1: Obtain remote sensing image data;
[0008] Step S2: Preprocess the remote sensing image data to obtain feature data;
[0009] Step S3: inputting the feature data into a pre-trained land classification extraction model for identification and prediction; wherein the land classification extraction model is based on the introduced convolutional deep learning network, the recurrent deep learning network and the autoencoder to achieve deep feature extraction and dimensionality reduction, and combines GIS technology and active learning methods in the model training process to automatically identify and eliminate samples affected by environmental factors, and dynamically adjust the sample set;
[0010] Step S4: Output the land classification results and perform a visual display.
[0011] In one implementation of the present application, in step S1, the remote sensing image data is acquired based on satellites, drones or aerial photography, and the data covers different time periods of the target area, including multi-spectral, hyperspectral and synthetic terrain features.
[0012] In one implementation of the present application, step S2 specifically includes:
[0013] Step S21: Identify interference factors based on GIS and remove the interference factors, wherein the interference factors include: cloud cover, shadow and noise, and use image restoration technology to remove these interference factors, including using time series data to fill the cloud cover area, or using the statistical characteristics of neighboring pixels to eliminate noise.
[0014] Step S22: extracting feature data from the image after removing interference factors; wherein the feature data includes: spectral features, texture features and terrain features;
[0015] Step S23: Eliminate invalid and poor quality feature data samples based on data cleaning. A threshold can be set to filter out image segments containing a lot of noise or missing values.
[0016] In one implementation of the present application, in step S3, the model training process specifically includes:
[0017] The extracted feature data is integrated into the deep learning network for deep feature extraction; wherein the convolutional deep learning network is used to capture the spatial local correlation of spectral and texture features; the recurrent deep learning network is used to analyze time series data and capture the dynamic characteristics of land use changes; the autoencoder: reduces the dimension of features and extracts key information to improve the generalization ability of the model;
[0018] Input the fused feature vector into the model and output the land classification label; the land classification labels include: forest land, cultivated land, and building land; initialize the weights of all features to the same value, ω 1 =ω 2 =ω 3 =...ωN = 1 / N, randomly select a part of the samples to construct the first decision tree.
[0019] Construct a decision tree and perform ensemble learning and gradient boosting operations;
[0020] Output the land classification results and perform model evaluation and optimization adjustment.
[0021] In an implementation manner of the present application, the process of constructing a decision tree specifically includes the following steps:
[0022] When splitting each node of the decision tree, dynamically adjust the features according to the information gain and / or Gini impurity, and at the initial stage of model training, assign the same initial weight to all features; the formula for the information gain is as follows: Where T is the training set, a is the feature, is the subset where the value of feature a is, and H is the entropy function;
[0023] During the construction of each decision tree, update the weights of the features according to the feature selection results, and reorder the feature list according to the updated weights;
[0024] When splitting each node, calculate the local density and complexity of the data points in the node, and dynamically adjust the splitting threshold. Based on the local density and complexity, increase or decrease the number of features participating in the splitting.
[0025] In an implementation manner of the present application, the ensemble learning and gradient boosting operations specifically include the following process:
[0026] Based on the constructed basic decision tree model, in each round of training, calculate the error between the prediction result of the previous tree and the true label;
[0027] Calculate the gradient of each sample according to the error, where the gradient is the contribution of each sample to the loss function, and the optimization formula for the gradient boosting loss function is as follows: l is the loss function, yi is the true label, and F(x_i) is the predicted model value;
[0028] Optimize the loss function of the overall model through the gradient descent method, and combine the prediction results of each tree with weights to form the final classification result.
[0029] In an implementation manner of the present application, the method further includes:
[0030] Adjust the weight of each sample according to the gradient magnitude, so that the next round of training pays more attention to the samples with larger errors;
[0031] Construct a new decision tree based on the adjusted sample weights;
[0032] Dynamically adjust the learning rate according to the error rate of the pre-order tree to control the degree of model update in each round of training.
[0033] This application also provides a land classification and extraction system, which includes:
[0034] A data acquisition module for acquiring remote sensing image data;
[0035] A feature data generation module for preprocessing the remote sensing image data to obtain feature data;
[0036] A prediction module for inputting the feature data into a pre-trained land classification and extraction model for recognition and prediction; wherein, the land classification and extraction model is based on the introduced convolutional deep learning network, recurrent deep learning network and autoencoder to achieve deep feature extraction and dimensionality reduction. In the training process of the model, GIS technology and active learning methods are combined to automatically identify and eliminate samples disturbed by environmental factors and dynamically adjust the sample set;
[0037] A display module for outputting the land classification result and performing visual display.
[0038] This application also provides a land classification and extraction device, which includes:
[0039] At least one processor; and,
[0040] A memory communicatively connected to the at least one processor; wherein,
[0041] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to complete the foregoing land classification and extraction method.
[0042] This application also provides a non-volatile computer storage medium for land classification and extraction, storing computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the foregoing land classification and extraction method.
[0043] A method, system, device and storage medium for land classification and extraction provided by the present application significantly improve the efficiency and accuracy of land use classification in remote sensing images through an improved algorithm of random forest based on deep learning and adaptive optimization, through dynamic feature importance adjustment, ensemble learning, gradient boosting and adaptive segmentation strategy. Through GIS technology and active learning method, samples disturbed by environmental factors are automatically identified and removed, and the sample set is dynamically adjusted. Through adaptive random forest optimization, the feature importance can be dynamically adjusted, and the gradient boosting mechanism is introduced to improve the generalization ability and classification performance of the model. Through the adaptive segmentation strategy, the segmentation strategy can be dynamically adjusted based on data density and complexity to avoid overfitting while maintaining the model interpretability. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation to the present application. In the drawings:
[0045] Figure 1 is a flowchart of a method for land classification and extraction provided by an embodiment of the present application;
[0046] Figure 2 is a composition diagram of a system for land classification and extraction provided by an embodiment of the present application;
[0047] Figure 3 is a schematic diagram of a device for land classification and extraction provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0048] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.
[0049] An embodiment of the present application provides a method, system, device and storage medium for land classification and extraction. The technical solutions proposed in the embodiments of the present application will be described in detail below with reference to the drawings.
[0050] Figure 1 is a flowchart of a method for land classification and extraction provided by an embodiment of the present application. As Figure 1 shown, the method mainly includes the following steps:
[0051] Step S1: Obtain remote sensing image data;
[0052] Step S2: Preprocess the remote sensing image data to obtain feature data;
[0053] Step S3: Input the feature data into a pre-trained land classification and extraction model for recognition and prediction; wherein, the land classification and extraction model is based on the introduced convolutional deep learning network, recurrent deep learning network and autoencoder to achieve deep feature extraction and dimensionality reduction. In the training process of the model, GIS technology and active learning methods are combined to automatically identify and eliminate samples disturbed by environmental factors, and the sample set is dynamically adjusted;
[0054] Step S4: Output the land classification result and perform visual display.
[0055] The implementation process of the present invention will be described below based on specific embodiments.
[0056] Embodiment 1
[0057] First, high-resolution remote sensing image data from 1991 to 2021 covering the Shandong Peninsula region was selected. Using the GEE platform, according to the land use classification standards (cultivated land, forest land, grassland, water area, construction land, unused land), a sample set of 7 time phases was constructed to ensure the uniform distribution and representativeness of the samples in space. At the same time, the influence of environmental factors such as terrain and climate was considered to avoid the interference of cloud and shadow areas.
[0058] Then, a multi-dimensional feature set was constructed, specifically including: Spectral feature extraction: Different bands (such as B1 to B7) of the Landsat series (Landsat-5TM, Landsat-7TM, Landsat-8OLI) were selected to extract bands with significant object recognition ability. By calculating spectral indices such as the Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), and Normalized Difference Water Index (NDWI), the spectral characteristics of objects were further enhanced. Deep texture features: Deep learning technology, especially convolutional neural networks (CNNs), was innovatively introduced to automatically extract image textures, replacing the traditional Gray Level Co-occurrence Matrix (GLCM) method to capture more complex texture patterns. The CNN model automatically learns the texture features in the image through multiple convolutional and pooling operations. Terrain features: Using USGS / SRTMGL1_003 data, slope and height were extracted through the ee.Algorithms.Terrain() function as terrain auxiliary features.
[0059] Secondly, the deep optimization and training of the random forest model, dynamic feature importance adjustment: In the training process of the random forest model, a dynamic feature importance adjustment mechanism was implemented. First, an initial weight (optionally equal weights) was assigned to each feature.
[0060] During the construction of each decision tree, calculate the information gain (or reduction in Gini impurity) for each candidate splitting feature and update the global weight of that feature accordingly. The weight update can be accumulative, i.e., each time the feature is selected as the splitting feature, its weight is increased.
[0061] In subsequent decision tree construction, perform feature selection based on the global weights of the features. Preferentially select features with high weights for node splitting, thereby improving the classification efficiency and accuracy of the model.
[0062] Ensemble learning and gradient boosting: Incorporate the gradient boosting mechanism on the basis of traditional random forests. First, use the initial random forest model to make predictions on the training set and calculate the residuals between the predicted values and the true values.
[0063] Next, construct a new decision tree to fit these residuals instead of directly fitting the original data. The new decision tree will focus on learning the parts where the model makes prediction errors.
[0064] In each iteration step, dynamically adjust the learning rate according to the prediction results of the previous trees to reduce the contribution of subsequent trees (if the previous trees have well-fitted the data) or increase the contribution of subsequent trees (if the previous trees are not accurately predicted).
[0065] Finally, integrate the prediction results of all decision trees (including the initial random forest and gradient boosting trees) through weighted summation to obtain the final prediction result.
[0066] Adaptive splitting strategy: During the node splitting process of the decision tree, adopt an adaptive splitting strategy. First, calculate the density and distribution complexity (such as entropy, Gini impurity, etc.) of the samples within the node.
[0067] Dynamically adjust the splitting threshold according to the density and complexity of the samples. In regions with high sample density and low complexity, appropriately reduce the number of splits or relax the splitting conditions to avoid overfitting; in regions with low sample density or high complexity, increase the number of splits or tighten the splitting conditions to better capture the subtle differences in the data. Dynamic adjustment is to dynamically adjust the sample set according to the model prediction results, preferentially resample the samples that are difficult to classify, and enhance the adaptability of the model to complex situations. Active learning is to combine expert knowledge to label the samples that are difficult to classify and add these samples to the training set for retraining.
[0068] In this way, the adaptive splitting strategy can improve the generalization ability and classification accuracy of the model while maintaining the model complexity.
[0069] Model Training and Parameter Tuning: Use the randomColumn function to randomly divide the samples into a training set (70%) and a test set (30%) to ensure the generalization ability of the model.
[0070] Apply ee.Classifier.randomForest() to build a random forest model and integrate the above improvement mechanisms. Set parameters such as the number of trees, maximum depth, and minimum number of samples, and adjust them through a cross-validation strategy to achieve the best classification performance.
[0071] During the training process, monitor the performance metrics of the model (such as accuracy, recall, F1-score, etc.), and adjust the model parameters and feature weights as needed.
[0072] Through in-depth optimization and training, it not only has the ability of deep learning-assisted feature extraction, but also further improves the classification accuracy and practicality of the model through improvement mechanisms such as dynamic feature importance adjustment, ensemble learning and gradient boosting, and adaptive segmentation strategy.
[0073] Secondly, after model training, accuracy evaluation and iterative optimization are required. Through confusion matrix analysis, verify the accuracy of the classification results, and calculate the overall accuracy (OA) and Kappa coefficient. According to the preliminary classification results, dynamically adjust the sample set, especially add sample points for misclassified categories, and optimize the model through multiple iterations until the classification accuracy reaches the expected value.
[0074] Apply this application to the analysis of land use changes in the urbanization process. Use the improved random forest model to analyze the land use changes in a certain city in the Shandong Peninsula in the past decade, especially the conversion of cultivated land to construction land in the urban fringe area. The classification results show that with the expansion of the city, the cultivated land area has decreased significantly, while the construction land area has increased sharply. By comparing the classification results year by year, this model accurately reveals the rate and direction of urban expansion, providing important data support for urban planning and ecological environmental protection.
[0075] The present invention effectively improves the accuracy and stability of remote sensing image classification through deep learning and adaptive optimization strategies, providing a more accurate and reliable tool for land resource management, urban planning, environmental monitoring, etc. Through the demonstration of examples and application examples, the effectiveness and practicality of this method in practical applications are verified.
[0076] The above is a land classification and extraction method provided by an embodiment of this application. Based on the same inventive concept, an embodiment of this application also provides a land classification and extraction system. Figure 2 For the composition diagram of a land classification and extraction system provided by an embodiment of this application, as Figure 2As shown in the figure, the system mainly includes: a data acquisition module 201 for acquiring remote sensing image data; a feature data generation module 202 for preprocessing the remote sensing image data to obtain feature data; a prediction module 203 for inputting the feature data into a pre-trained land classification and extraction model for recognition and prediction; wherein, the land classification and extraction model is based on an introduced convolutional deep learning network, a recurrent deep learning network, and an autoencoder to achieve deep feature extraction and dimensionality reduction. During the training process of the model, GIS technology and active learning methods are combined to automatically identify and eliminate samples interfered by environmental factors and dynamically adjust the sample set; a display module 204 for outputting the land classification result and performing visual display.
[0077] The above is a land classification and extraction system provided by an embodiment of the present application. Based on the same inventive concept, an embodiment of the present application also provides a land classification and extraction device. Figure 3 The following is a schematic diagram of a land classification and extraction device provided by an embodiment of the present application. As Figure 3 shown, the device mainly includes: at least one processor 301; and a memory 302 communicatively connected to the at least one processor; wherein, the memory 302 stores instructions executable by the at least one processor 301, and the instructions are executed by the at least one processor 301 so that the at least one processor 301 can complete the foregoing land classification and extraction method.
[0078] In addition, an embodiment of the present application also provides a non-volatile computer storage medium for land classification and extraction, storing computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the foregoing land classification and extraction method.
[0079] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.
[0080] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the functions specified in one or more of the processes Figure 1 or steps and / or Figure 1 boxes specified in one or more of the boxes.
[0081] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes Figure 1 or steps and / or Figure 1 boxes specified in one or more of the boxes.
[0082] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.
[0083] Each embodiment in this application is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description of the method embodiments.
[0084] It should also be noted that the term "comprising", "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus including a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or apparatus including the element.
[0085] The above are only the embodiments of the present application and are not used to limit the present application. For those skilled in the art, various modifications and changes can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.
Claims
1. A land classification extraction method, characterized in that: The method comprises the following steps: Step S1: Acquire remote sensing image data; Step S2: preprocessing the remote sensing image data to obtain feature data; Step S3: inputting the feature data into a pre-trained land classification extraction model for identification and prediction; wherein the land classification extraction model is based on the introduced convolutional deep learning network, the recurrent deep learning network and the autoencoder to achieve deep feature extraction and dimensionality reduction, and combines GIS technology and active learning methods in the model training process to automatically identify and eliminate samples affected by environmental factors, and dynamically adjust the sample set; Step S4: Output the land classification results and perform a visual display.
2. A land classification extraction method according to claim 1, characterized in that: In step S1, the remote sensing image data is acquired based on satellite, drone or aerial photography, and the data covers different time periods of the target area, including multi-spectral, hyper-spectral and synthetic terrain features.
3. A land classification extraction method according to claim 1, characterized in that: The step S2 specifically includes: Step S21: Identify interference factors based on GIS and remove the interference factors; wherein the interference factors include: cloud cover, shadow and noise; Step S22: extracting feature data from the image after removing interference factors; wherein the feature data includes: spectral features, texture features and terrain features; Step S23: Eliminate invalid and poor-quality feature data samples based on data cleaning.
4. A land classification extraction method according to claim 1, characterized in that: In step S3, the model training process specifically includes: The extracted feature data is integrated into the deep learning network for deep feature extraction; wherein the convolutional deep learning network is used to capture the spatial local correlation of spectral and texture features; the recurrent deep learning network is used to analyze time series data and capture the dynamic characteristics of land use changes; the autoencoder: reduces the dimension of features and extracts key information to improve the generalization ability of the model; Input the fused feature vector into the model and output the land classification label; the land classification labels include: forest land, cultivated land, and building land; Build a decision tree and perform ensemble learning and gradient boosting operations; Output land classification results, and conduct model evaluation and optimization adjustments.
5. A land classification extraction method according to claim 4, characterized in that: The process of building a decision tree includes the following steps: When each node of the decision tree is split, the features are dynamically adjusted according to information gain and / or Gini impurity, and at the beginning of model training, all features are assigned the same initial weight; the formula for the information gain is as follows: Among them, T is the training set, a is the feature, is the subset of feature a, and H is the entropy function; In each decision tree construction process, the feature weights are updated according to the feature selection results, and the feature list is reordered according to the updated weights; When each node is split, the local density and complexity of the data points in the node are calculated, and the segmentation threshold is dynamically adjusted to increase or decrease the number of features involved in the segmentation based on the local density and complexity.
6. A land classification extraction method according to claim 4, characterized in that: Ensemble learning and gradient boosting operations, specifically The process includes: Based on the constructed basic decision tree model, in each round of training, the error between the predicted result of the previous sequence number and the true label is calculated; The gradient of each sample is calculated based on the error, where the gradient is the contribution of each sample to the loss function. The gradient boosting loss function optimization formula is as follows: l is the loss function, y i is the true label, F(x i ) is the predicted model value; The loss function of the overall model is optimized by the gradient descent method, and the prediction results of each tree are weighted and combined to form the final classification result.
7. A land classification extraction method according to claim 6, characterized in that: The method further comprises: Adjust the weight of each sample according to the gradient size, so that the next round of training will pay more attention to samples with larger errors; Build a new decision tree based on the adjusted sample weights; The learning rate is dynamically adjusted according to the error rate of the previous tree to control the degree of model update in each round of training.
8. A land classification extraction system, characterized in that: The system comprises: A data acquisition module, used for acquiring remote sensing image data; A feature data generating module, used for preprocessing the remote sensing image data to obtain feature data; A prediction module is used to input the feature data into a pre-trained land classification extraction model for identification and prediction; wherein the land classification extraction model is based on the introduced convolutional deep learning network, recurrent deep learning network and autoencoder to achieve deep feature extraction and dimensionality reduction. In the training process of the model, GIS technology and active learning methods are combined to automatically identify and eliminate samples interfered by environmental factors, and dynamically adjust the sample set; The display module is used to output the land classification results and display them visually.
9. A land classification extraction device, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can complete a land classification extraction method as described in any one of claims 1-8.
10. A non-volatile computer storage medium for land classification extraction, storing computer executable instructions, characterized in that: The computer executable instructions are executed by a processor to implement a land classification extraction method as described in any one of claims 1-8.