Object-oriented coral reef remote sensing image landform classification method and device

Abstract

The invention provides an object-oriented coral reef remote sensing image landform classification method and device, and belongs to the technical field of remote sensing monitoring, and the method comprises the steps: constructing a landform unit classification system of a coral reef, and carrying out the data preprocessing of an obtained coral reef original remote sensing image; segmenting an original remote sensing image by using a multi-scale segmentation algorithm in object-oriented segmentation, finding an optimal parameter, and finally obtaining an optimal segmentation result; feature extraction and classification are carried out on the preprocessed remote sensing image, remote sensing index features and texture features of a coral reef landform unit classification data set are calculated, and a landform unit classification data set of coral reefs is obtained; training the coral reef landform unit classification data set by using an object-oriented machine learning algorithm, and adjusting and optimizing model parameters to obtain a trained identification and classification model; and predicting by using the trained classification model, and making a corresponding coral reef remote sensing landform unit classification result map. According to the invention, the classification accuracy and efficiency can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing monitoring, and in particular relates to an object-oriented method and device for landform classification of coral reef remote sensing images. Background Art

[0002] Coral reefs are biomass limestone formed by the long-term accumulation and deposition of calcium substances by reef-building corals and other reef-building organisms. They are mainly distributed in tropical shallow waters. Coral reefs are called "oases in the blue desert". The coral reef ecosystem is one of the most species-rich and productive ecosystems on Earth. Coral reefs are rich in species and play an important role in maintaining the diversity of marine life. At the same time, they also provide many biological resources for human production and life, and have huge development potential, economic value and ecological and environmental value.

[0003] However, the sensitivity and vulnerability of coral reefs to global change has led to a rapid degradation of coral reef ecosystems with high biodiversity and high economic value in recent decades under the influence of global change. Coral reefs are vital to marine ecosystems, and the fishery resources provided by coral reefs have made important contributions to people's livelihoods and national economic growth. Moreover, coral reefs are an important carrier for recording global change, and can accurately record the changing processes of tropical marine environments at different time scales and the mechanisms of their interaction with global change. Therefore, protecting coral reefs is a key measure to maintain marine ecological balance and sustainable development. In order to better understand and protect coral reefs, effective classification and monitoring methods have become essential.

[0004] With the advancement of science and technology, remote sensing technology has made great achievements in the field of earth science, and remote sensing technology has been proven to be an effective technical means to classify and monitor coral reefs. In this context, remote sensing technology has become one of the important tools for studying coral reefs. Remote sensing technology has the advantages of large-scale synchronous coverage, repeated detection of the same area, high accuracy, and rich information. In particular, high-resolution remote sensing has become an important means of classifying coral islands and reefs on a large scale, and it has excellent performance in spatial resolution and positioning accuracy. Multi-platform, multi-sensor, and multiple spatial and spectral resolution remote sensing data have been widely used in coral island and reef research, providing a powerful tool for in-depth understanding and protection of this ecosystem.

[0005] Traditional coral reef classification methods are limited by the subjectivity and time-consuming nature of manual identification, and a more efficient and accurate classification method is urgently needed. Currently, machine learning, with its outstanding performance in large-scale data processing and complex pattern recognition, provides unique advantages for the classification of coral reef geomorphic units. However, how to improve the accuracy and running speed of machine learning in the classification results of coral reef geomorphic units is a problem that needs to be solved urgently. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art. The present invention provides an object-oriented method and device for coral reef remote sensing image landform classification, which can improve the accuracy and efficiency of classification.

[0007] In order to achieve the above object, the present invention adopts the following technical scheme: An object-oriented method for landform classification of coral reef remote sensing images comprises the following steps: Step 1: Construct a classification system of coral reef geomorphic units and perform data preprocessing on the original remote sensing images of coral reefs; Step 2: Use the multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image, find the optimal parameters, and finally obtain the optimal segmentation result; Step 3: extract and classify the preprocessed remote sensing images, calculate the remote sensing index features and texture features of the coral reef geomorphic unit classification data set, and obtain the coral reef geomorphic unit classification data set; Step 4: Use an object-oriented machine learning algorithm to train the coral reef geomorphic unit classification data set, and use the k-fold cross-validation method with adaptive weight allocation to optimize the model parameters to obtain the trained recognition and classification model; Step 5: Use the trained classification model to make predictions and produce the corresponding coral reef remote sensing geomorphological unit classification result map.

[0008] Furthermore, the step 1 comprises: Step 11. Based on the existing classification system of coral reef geomorphic units and combined with the spectral characteristics of coral reef geomorphology in remote sensing images, a classification system of coral reef geomorphic units is established, and coral reef geomorphic units are divided into: deep reef fore slope, shallow reef fore slope, reef ridge, coral dense area, coral sparse area, lagoon slope, lagoon bottom, point reef, sandbar and land; each coral reef island will change the classification system according to its unique geomorphic characteristics; Step 12: The original remote sensing image of the coral reef adopts the World-View2 high-resolution remote sensing image. The data preprocessing of the original remote sensing image includes image denoising, geometric correction, image fusion and image enhancement.

[0009] Furthermore, the step 2 comprises: Step 21, using a multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image; the multi-scale segmentation algorithm performs segmentation according to the weights, compactness and smoothness of different bands in the image to obtain image objects of different sizes and shapes; Step 22, find the optimal segmentation parameters and finally obtain the best segmentation result; The optimal segmentation result satisfies the following two conditions: (1) maximizing the homogeneity within the object and (2) maximizing the heterogeneity between objects.

[0010] Furthermore, the step 3 comprises: Step 31, extracting and classifying features from the pre-processed remote sensing images, calculating remote sensing index features and texture features of the coral reef geomorphic unit classification data, and obtaining a geomorphic unit classification data set of the coral reef; Step 32: The classification data set of coral reef geomorphic units is divided into stratified sampling methods in random sampling. By dividing the population into different levels, the sampled data set has the same proportion as the original data set. , set the extraction ratio , according to the ratio Divide the coral reef geomorphic unit dataset into training sets and test set .

[0011] Furthermore, in step 31, the remote sensing index feature uses a series of variant index features related to the normalized vegetation index NDVI and the normalized difference water index NDWI; the texture feature is calculated using the gray level co-occurrence matrix GLCM, and is defined as is a texture feature function, which includes texture features such as contrast , dissimilarity , Uniformity , second-order distance ,entropy , mean ,variance , Correlation , inverse difference .

[0012] Furthermore, the step 4 comprises: By building a collection of multiple machine learning models , including random forest, support vector machine, K nearest neighbor algorithm, XGBoost and other algorithms, training model collection Each model in To predict the classification labels of coral reef geomorphic units; use the grid search method to tune the model's hyperparameters, and use the k-fold cross-validation method with adaptive weight allocation to adaptively adjust according to the size of the data set The size of the model is adjusted, and the weights of the categories are redistributed according to the degree of category imbalance in each fold. The optimal parameter combination is determined to improve the accuracy of the model and retrain the model.

[0013] Furthermore, the k-fold cross validation of the adaptive weight allocation includes: Divide the original data set into Each of these subsets is used as a validation set in turn, and the rest are subsets are combined into a training set; in adaptive adjustment Based on the value, the weight of each category is dynamically adjusted according to the category distribution in each fold ; For each fold, the training set is used to train the model, and then the validation set is used to evaluate the performance indicators of the model; Weight The calculation formula is: ; Among them, It is The weight of the class, is the total number of samples in the dataset, It is The number of samples in the class; Repeat the steps so that each fold serves as a validation set once, until each fold has been used as a validation set once; based on the results of each validation, calculate the average performance indicator as an estimate of the performance of the model on the entire dataset.

[0014] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the above-mentioned object-oriented method for landform classification of coral reef remote sensing images are implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned object-oriented coral reef remote sensing image landform classification method are implemented.

[0016] Beneficial effects: The present invention is based on the World-View2 high-resolution remote sensing image, and simultaneously trains multiple machine learning models - random forest model (RF), support vector machine (SVM), K nearest neighbor model (KNN), and distributed gradient boosting model (XGBoost). In the process of classifying coral reef geomorphic units, different machine learning models show their own characteristics and advantages. For example, in the classification task of point reefs, the RF model and the XGBoost model show high accuracy and can better identify and distinguish the characteristics of point reefs. The SVM model has a high overall classification accuracy, and the KNN model has the fastest classification speed. After the four models are trained, the classification accuracy of the four models for each category of geomorphic units can be automatically extracted, and the overall classification accuracy and the classification accuracy of each category can be calculated. And the model can realize the automatic production of the classification result map of coral reef geomorphic units based on the prediction record. Compared with traditional machine learning methods, the object-oriented classification method shows significant advantages in the identification and classification of geomorphic units. This method can more accurately capture the essential characteristics of geomorphic units by identifying natural objects in the image, such as land cover types, buildings, water bodies, etc., rather than relying solely on pixel-level analysis. It can also better handle the redundancy of spatial data, reduce resource waste, and improve the practicality and reliability of classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other features, purposes and advantages of the present invention will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings: Figure 1 A flowchart of an object-oriented method for landform classification of coral reef remote sensing images of the present invention; Figure 2 It is a schematic diagram of the multi-scale segmentation process of the present invention; Figure 3 This is a schematic diagram of the accuracy distribution process of the model training of the present invention; Figure 4 A schematic diagram of the accuracy and Kappa coefficient of the model training of the present invention; Figure 5 A schematic diagram of the accuracy of a single landform category trained by the model of the present invention; Figure 6a , Figure 6b , Figure 6c , Figure 6d This is the classification result map of coral reef geomorphic units of the present invention; wherein, Figure 6a is a random forest, Figure 6b is a support vector machine, Figure 6c is the K nearest neighbors, Figure 6d Distributed Gradient Boosting. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0019] In the description of the present invention, it should be understood that the terms "upper", "lower", "top", "bottom", "inner", "outer", etc. used in the present invention to indicate the orientation or position relationship is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0020] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0021] The present invention discloses an object-oriented method for landform classification of coral reef remote sensing images. The method adopts supervised classification ideas and classifies coral reef landform units based on machine learning models. The method constructs a quasi-sample of coral reef landform unit classification through visual interpretation, World-View2 satellite classification products or coarse classification algorithms; by establishing a set of multiple classification models , training set Each model in , thereby predicting the classification labels of coral reef geomorphic units.

[0022] This method can ensure the speed of operation under the condition of limited training samples and save system operation costs; this method effectively samples the hyperparameters and model design space through a complete evaluation system, improves the advanced performance of the machine learning model and the quality of the machine learning model set, thereby obtaining the machine learning model with the best performance to classify coral reef geomorphic units, greatly improving the accuracy of the operation. Therefore, this method can ensure high speed and accuracy in the process of coral reef geomorphic unit classification.

[0023] The details of the embodiments of the present invention are described in detail below through specific examples.

[0024] like Figure 1 As shown, the present invention provides an object-oriented method for coral reef remote sensing image landform classification, comprising the following steps: Step S1: construct a classification system for coral reef geomorphic units and perform data preprocessing on the acquired original remote sensing images of coral reefs, including: Step S11, fully investigate the existing coral reef geomorphology classification system, and combine the spectral characteristics of coral reef geomorphology on remote sensing images to establish a coral reef geomorphology unit classification system, and divide the coral reef geomorphology units into: deep reef front slope, shallow reef front slope, reef ridge, coral dense area, coral sparse area, lagoon slope, lagoon bottom, point reef, sandbar and land. Since coral reefs have their own unique coral reef geomorphology unit distribution, the coral reef geomorphology unit classification system established is the geomorphology unit characteristics of many coral reef islands, and with reference to the existing coral reef geomorphology unit classification system, an overall classification system for coral reef islands is proposed, and each coral reef island will modify its classification system according to its unique geomorphology characteristics.

[0025] Step S12: The original remote sensing image of the coral reef adopts the World-View2 high-resolution remote sensing image. The data preprocessing includes image denoising, geometric correction, image fusion and image enhancement. These preprocessings help the machine learning model to better capture the association between different bands, better capture the details and features of the geomorphic units, and improve the model's classification performance for different geomorphic units.

[0026] Step S2: Use the multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image of the coral reef, and find the optimal segmentation scale parameter, and finally obtain the best segmentation result, including: The multi-scale segmentation algorithm can segment the original coral reef remote sensing images according to the weights, compactness and smoothness of different bands, and can obtain image objects of different sizes and shapes. This method can make full use of the spatial, spectral, texture and other information of the original coral reef remote sensing images, and is suitable for the extraction of complex areas or detailed land types.

[0027] like Figure 2 As shown, the object-oriented segmentation is to obtain the optimal solution of the segmentation parameters by setting different segmentation parameters, and to obtain the optimal segmentation result by evaluating the segmentation scheme results of different parameters.

[0028] First, by setting a segmentation scale parameter set , the original image is segmented through a multi-scale segmentation algorithm, the regional features of the segmentation results are calculated, the segmentation quality is evaluated, and the segmentation results are judged. If the segmentation results meet the requirements, similar areas in the results are merged. The ESP tool is used to determine whether the segmentation parameters are optimal, and finally the optimal segmentation scale is determined and the segmentation results are obtained. Calculation process: Calculate the local variance of different objects in a band to calculate the mean of the local variance of multiple bands: ; in, Expressed as the local variance of each band, i=1,2,…,8. Indicates the average calculation of multiple bands.

[0029] If the segmentation scale is much larger than the objects in the scene, most of the segmented objects are highly correlated, and the local variance measurement will be low. If the segmentation scale is close to the size of the target object, the heterogeneity between the target objects will increase, and the local variance will increase. In order to obtain the optimal segmentation scale, use The curve shows that when When the curve reaches a maximum value, the scale is the optimal segmentation scale, and the formula is as follows: ; in, Represents the Receiver Operating Characteristic curve Represents the local variance of the target layer; Representatives will The local variance of the target layer is taken as the local variance of the next layer of the benchmark. When the curve reaches a maximum value, the scale is the optimal segmentation scale.

[0030] Finally, the best result should basically meet the following two conditions: (1) maximizing homogeneity within objects and (2) maximizing heterogeneity between objects.

[0031] The formula for calculating the within-subject average variance is as follows: ; in, Is an object The variance of the values ​​in the inner cells, Is an object The average within-subject variance is calculated as an area-weighted average, with n being the total number of objects. More weight is given to larger objects, thus reducing the destabilizing effects that may be caused by small objects. The higher the value of , the greater the overall heterogeneity of the objects; the lower the value, the more homogeneous the objects are with respect to the measured attribute. In general, there is a direct relationship between the mean size of the objects and the within-object variance: if the objects are large, then higher and vice versa.

[0032] Moran's I It can be expressed as: ; in, Is an object The average value of Is an object The average value of is the average value of the scene, It is the object of measurement and objects The weight of the spatial adjacency of . and objects Adjacent, then The value of is 1; otherwise . Moran Index The value range is from -1 to +1. Indexes close to zero and negative values ​​indicate that the differences between the average values ​​of adjacent objects are usually large. When both conditions are met, the optimal segmentation result of the remote sensing image can be obtained.

[0033] In this embodiment, the segmentation parameters determined by fitting the ROC curve using the ESP tool are: scale parameter: 80; shape parameter: 0.3; compactness: 0.5. In other embodiments, the segmentation parameters need to be adjusted according to actual conditions.

[0034] Step S3: extracting and classifying features from the pre-processed remote sensing images, calculating remote sensing index features and texture features of the coral reef geomorphic unit classification data, and obtaining a coral reef geomorphic unit classification data set, including: Step S31: Since multi-scale segmentation is performed in step S2, the shape features of the remote sensing image are no longer calculated in step S31. The remote sensing index features mainly use a series of variant index features related to the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI). The texture features are calculated using the Gray Level Co-occurrence Matrix (GLCM), which is defined as is a texture feature function, which includes texture features such as contrast , dissimilarity , Uniformity , second-order moment ,entropy , mean ,variance , Correlation , inverse difference The following are the calculation methods of all texture feature functions: ; ; ; ; ; ; ; ; in, For the matrix The number of rows or columns in (equal to the number of gray levels), For the matrix The row index of (equal to the gray level of the reference cell), For the matrix The column index of (equal to the gray level of the adjacent cell), With grayscale and grayscale The probability (relative frequency) of the neighboring units. is the mean of the row or column, is the standard deviation of the row or column, As a condition.

[0035] ; in, , expressed as The sum of is grayscale and vector. Grayscale and vector is a matrix representing the calculation based on Two gray levels The vector of probability distributions of the sum of . The position in the vector (starting from 0) represents The value of this position corresponds to the gray level and grayscale The probability that the sum of is equal to this value.

[0036] The gray level co-occurrence matrix in this embodiment The sliding window size is 3×3; the step size is , that is, the central pixel is directly compared with its adjacent pixels; direction selection: calculate the gray level co-occurrence matrix The direction is 45°.

[0037] In other embodiments, the sliding window size may be 5×5 or 7×7, or may be set according to the size of the data set; the direction selection is: after the eigenvalues ​​of the four direction matrices are obtained, the average value of the four eigenvalues ​​is calculated as the final value.

[0038] Step S32: The classification data set of coral reef geomorphic units is divided into stratified sampling in random sampling, which divides the population into different levels so that the sampled data set has the same proportion as the original data set. , set the extraction ratio , according to the extraction ratio Divide the coral reef geomorphic unit classification dataset into training sets and test set .

[0039] The label data set in this implementation example is annotated by experts based on their experience. The extraction ratio set in this embodiment is is 30%. In other embodiments, the extraction ratio is The optimal setting of should be determined through experiments based on actual conditions in order to find the ratio that best suits the current task.

[0040] Step S4: train a prediction model using the training set, the prediction model including multiple sub-machine learning models; optimize the model parameters by using a k-fold cross-validation method with adaptive weight allocation, and the trained prediction model is used to identify the geomorphic unit classification in the data set, including: By building a collection of various machine learning models (such as random forest (RF), support vector machine (SVM), K nearest neighbor model (KNN), distributed gradient boosting model (XGBoost)) , training set Each model in , thereby predicting the classification labels of coral reef geomorphic units. The grid search method is used to optimize the hyperparameters of the model, and the k-fold cross-validation method with adaptive weight allocation is used to adaptively adjust the hyperparameters according to the size of the data set. The size of the fold is adjusted, and the weights of the categories are redistributed according to the degree of category imbalance in each fold to improve the accuracy of the model, and the model is retrained to ensure that objective and stable training results can be obtained.

[0041] Grid search is a technique that determines the optimal hyperparameter combination for a machine learning model by traversing a predefined parameter grid. First, a finite parameter range set is set. , by traversing all possible parameter combinations for training and validation, and finally selecting the parameter set with the highest cross-validation score.

[0042] The specific process of k-fold cross validation for adaptive weight allocation is as follows: Divide the original data set into Each subset is used as a validation set in turn, and the rest are The subsets are combined into the training set. Based on the value, the weight of each category is dynamically adjusted according to the category distribution in each fold For each fold, the training set is used to train the model, and then the validation set is used to evaluate the model's performance indicators, such as accuracy, precision, recall, etc. A larger value and a smaller dataset can adjust the class weights more finely, so that the class with fewer samples in each fold gets a higher weight. , thereby compensating for the bias caused by insufficient sample size of the category.

[0043] Weight The calculation formula is: ; Among them, It is The weight of the class, is the total number of samples in the dataset, It is The number of samples of the class.

[0044] Repeat the steps so that each fold serves as a validation set once, until each fold has been used as a validation set once. Based on the results of each validation, the average performance indicator is calculated as the performance estimate of the model on the entire dataset. This strategy helps reduce training time while providing more flexible weight distribution for imbalanced data of different categories.

[0045] The model parameter adjustment method in this embodiment is not unique. The value is 10, the category weight The optimal parameter combination is obtained by calculation through the formula and finally verified. In other embodiments, other methods can be used to adjust according to prior knowledge. In this embodiment, a box plot is used to show the accuracy distribution obtained by training different models. Figure 3 A schematic diagram of the distribution of model training accuracy according to an embodiment of the present invention is shown. Figure 3 As shown, this embodiment shows the distribution of model accuracy. The horizontal axis represents the model, and the vertical axis represents the model accuracy. In other embodiments, the way to display the distribution of model accuracy can be set according to actual conditions. In this embodiment, a line graph is used to display the accuracy of the model training process. Figure 4 A schematic diagram of model training accuracy according to an embodiment of the present invention is shown as follows: Figure 4As shown, in this embodiment, the model accuracy and Kappa coefficient are displayed. The horizontal axis represents the model, and the vertical axis represents the model accuracy. In other embodiments, the method of displaying the model accuracy can be set according to the actual situation. In this embodiment, a bar chart is used to display the accuracy of each landform category during the model training process. Figure 5 A schematic diagram of the accuracy of a single category of model training according to an embodiment of the present invention is shown. Figure 5 As shown, in this embodiment, the model accuracy is displayed. The horizontal axis represents the landform category, and the vertical axis represents the model accuracy. In other embodiments, the method of displaying the accuracy of the model for each category can be set according to actual conditions.

[0046] Step S5: Use the trained classification model to make predictions and automatically create the corresponding coral reef remote sensing geomorphic unit classification result map. Figure 6a , Figure 6b , Figure 6c , Figure 6d Shown is a diagram showing the classification results of the coral reef geomorphic units of the present invention. Figure 6a is a random forest, Figure 6b is a support vector machine, Figure 6c is the K nearest neighbors, Figure 6d It is distributed gradient boosting. It can be found that each model has different recognition accuracy for each category of geomorphic units. For example, the random forest model based on decision tree and the distributed gradient boosting model are more suitable for distinguishing multiple geomorphic unit categories and can better capture the differences between different categories. The random forest model (RF) has the highest accuracy and is better than other models in identifying reefs; although the support vector machine (SVM) can recognize the outline of each geomorphic unit as a whole, there are cases of misclassification and omission; the K-nearest neighbor model (KNN) is not sensitive to the lagoon slope and lagoon bottom, resulting in poor classification results; the distributed gradient boosting model (XGBoost) is better overall, but due to the limitations of its implementation principle, the running speed is relatively slow.

[0047] The present invention uses World-View2 remote sensing images as data sources, adopts a method combined with object-oriented machine learning, and classifies coral reef geomorphic units based on the texture feature information of World-View2 remote sensing images and remote sensing indices, providing a specific implementation method for the classification of coral reef geomorphic units, and can automatically generate classification result maps, which is convenient for coral reef analysis and investigation.

[0048] The present invention also provides an object-oriented device for coral reef remote sensing image landform classification, comprising the following modules: System construction module, constructing the geomorphic unit classification system of coral reefs and performing data preprocessing on the original remote sensing images of coral reefs; The segmentation module uses the multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image, find the optimal parameters, and finally obtain the optimal segmentation result; The data set acquisition module extracts and classifies the preprocessed remote sensing images, calculates the remote sensing index characteristics and texture characteristics of the coral reef geomorphic unit classification data set, and obtains the coral reef geomorphic unit classification data set; The recognition and classification module uses an object-oriented machine learning algorithm to train the coral reef geomorphic unit classification data set, and uses the k-fold cross-validation method with adaptive weight allocation to optimize the model parameters to obtain the trained recognition and classification model; The prediction module uses the trained classification model to make predictions and produce corresponding classification result maps of coral reef remote sensing geomorphic units.

[0049] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the above-mentioned object-oriented method for landform classification of coral reef remote sensing images are implemented.

[0050] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned object-oriented coral reef remote sensing image landform classification method are implemented.

[0051] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and interpreted scripting language JavaScript, etc.

[0052] The present invention is described with reference to 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 flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0053] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0055] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0056] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An object-oriented method for landform classification of coral reef remote sensing images, characterized in that: The steps include: Step 1: Construct a classification system of coral reef geomorphic units and perform data preprocessing on the original remote sensing images of coral reefs; Step 2: Use the multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image, find the optimal parameters, and finally obtain the optimal segmentation result; Step 3: extract and classify the preprocessed remote sensing images, calculate the remote sensing index features and texture features of the coral reef geomorphic unit classification data set, and obtain the coral reef geomorphic unit classification data set; Step 4: Use an object-oriented machine learning algorithm to train the coral reef geomorphic unit classification data set, and use the k-fold cross-validation method with adaptive weight allocation to optimize the model parameters to obtain the trained recognition and classification model; Step 5: Use the trained classification model to make predictions and produce the corresponding coral reef remote sensing geomorphological unit classification result map.

2. The object-oriented method for landform classification of coral reef remote sensing images according to claim 1 is characterized in that: The step 1 comprises: Step 11. Based on the existing classification system of coral reef geomorphic units and combined with the spectral characteristics of coral reef geomorphology in remote sensing images, a classification system of coral reef geomorphic units is established, and coral reef geomorphic units are divided into: deep reef fore slope, shallow reef fore slope, reef ridge, coral dense area, coral sparse area, lagoon slope, lagoon bottom, point reef, sandbar and land; each coral reef island will change the classification system according to its unique geomorphic characteristics; Step 12: The original remote sensing image of the coral reef adopts the World-View2 high-resolution remote sensing image. The data preprocessing of the original remote sensing image includes image denoising, geometric correction, image fusion and image enhancement.

3. The object-oriented method for landform classification of coral reef remote sensing images according to claim 1 is characterized in that: The step 2 comprises: Step 21, using a multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image; the multi-scale segmentation algorithm performs segmentation according to the weights, compactness and smoothness of different bands in the image to obtain image objects of different sizes and shapes; Step 22, find the optimal segmentation parameters and finally obtain the best segmentation result; The optimal segmentation result satisfies the following two conditions: (1) maximizing the homogeneity within the object and (2) maximizing the heterogeneity between objects.

4. The object-oriented method for landform classification of coral reef remote sensing images according to claim 1, characterized in that: The step 3 comprises: Step 31, extracting and classifying features from the pre-processed remote sensing images, calculating remote sensing index features and texture features of the coral reef geomorphic unit classification data, and obtaining a geomorphic unit classification data set of the coral reef; Step 32: The classification data set of coral reef geomorphic units is divided into stratified sampling methods in random sampling. By dividing the population into different levels, the sampled data set has the same proportion as the original data set. , set the extraction ratio , according to the ratio Divide the coral reef geomorphic unit dataset into training sets and test set .

5. The object-oriented method for landform classification of coral reef remote sensing images according to claim 4 is characterized in that: In step 31, the remote sensing index feature uses a series of variant index features related to the normalized vegetation index NDVI and the normalized difference water index NDWI; the texture feature is calculated using the gray level co-occurrence matrix GLCM, and is defined as is a texture feature function, which includes texture features such as contrast , dissimilarity , Uniformity , second-order distance ,entropy , mean ,variance , Correlation , inverse difference .

6. The object-oriented method for landform classification of coral reef remote sensing images according to claim 1, characterized in that: The step 4 comprises: By building a collection of multiple machine learning models , including random forest, support vector machine, K nearest neighbor algorithm, XGBoost and other algorithms, training model collection Each model in To predict the classification labels of coral reef geomorphic units; use the grid search method to tune the model's hyperparameters, and use the k-fold cross-validation method with adaptive weight allocation to adaptively adjust according to the size of the data set The size of the model is adjusted, and the weights of the categories are redistributed according to the degree of category imbalance in each fold. The optimal parameter combination is determined to improve the accuracy of the model and retrain the model.

7. The object-oriented method for landform classification of coral reef remote sensing images according to claim 6 is characterized in that: The k-fold cross validation of the adaptive weight allocation includes: Divide the original data set into Each of these subsets is used as a validation set in turn, and the rest are subsets are combined into a training set; in adaptive adjustment Based on the value, the weight of each category is dynamically adjusted according to the category distribution in each fold ; For each fold, the training set is used to train the model, and then the validation set is used to evaluate the performance indicators of the model; Weight The calculation formula is: ； Among them, It is The weight of the class, is the total number of samples in the dataset, It is The number of samples in the class; Repeat the steps so that each fold serves as a validation set once, until each fold has been used as a validation set once; based on the results of each validation, calculate the average performance indicator as an estimate of the performance of the model on the entire dataset.

8. An object-oriented device for landform classification of coral reef remote sensing images, characterized in that: Includes the following modules: System construction module, constructing the geomorphic unit classification system of coral reefs and performing data preprocessing on the original remote sensing images of coral reefs; The segmentation module uses the multi-scale segmentation algorithm in object-oriented segmentation to segment the original remote sensing image, find the optimal parameters, and finally obtain the optimal segmentation result; The data set acquisition module extracts and classifies the preprocessed remote sensing images, calculates the remote sensing index characteristics and texture characteristics of the coral reef geomorphic unit classification data set, and obtains the coral reef geomorphic unit classification data set; The recognition and classification module uses an object-oriented machine learning algorithm to train the coral reef geomorphic unit classification data set, and uses the k-fold cross-validation method with adaptive weight allocation to optimize the model parameters to obtain the trained recognition and classification model; The prediction module uses the trained classification model to make predictions and produce corresponding classification result maps of coral reef remote sensing geomorphic units.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the object-oriented method for coral reef remote sensing image landform classification according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of an object-oriented method for landform classification of coral reef remote sensing images as described in any one of claims 1 to 7 are implemented.

Images