Sea ice type remote sensing classification method and device
By extracting the waveform features in sea ice data and building a convolutional neural network model, the problem of cumbersome parameter tuning and long model training time in the existing technology is solved, and accurate classification and efficient monitoring of sea ice types are achieved.
Patent Information
- Application Number
- CN202411644974.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-04-11
AI Technical Summary
The prior art has problems such as cumbersome parameter tuning, long model training time, and different contributions to different waveform characteristics in the classification of sea ice, making it difficult to achieve accurate classification of sea ice.
By collecting Cryosat-2L1b data and AARI ice condition map data, the characteristics of the radar echo waveform, such as pulse width, leading edge width, trailing edge width, etc., are extracted, and spatially matched based on these features, and divided into training sets and test sets. Convolutional neural network model is built, trained and tested, and finally achieved accurate classification of sea ice types.
It realizes accurate classification of sea ice types, simplifies network structure and parameter tuning, improves the diversity and efficiency of classification, and supports sea ice monitoring and research.
Smart Images
Figure CN119807828A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of remote sensing application technology, and in particular to a sea ice type remote sensing classification method and device. Background Art
[0002] At present, sea ice type has an important impact on the global climate and is one of the important factors affecting sea ice thickness inversion and ship navigation. In the early days, sea ice type data could only be obtained through field surveys. Later, classification methods based on optical and SAR remote sensing images were developed, but this method was limited by the cost of remote sensing images, small spatial coverage and strict usage conditions, making it difficult to achieve large-scale sea ice classification. With the development of satellite radar altimeters, more scholars began to study the use of altimeter data to classify sea ice, because the altimeter has a larger spatial coverage and a higher temporal resolution. The use of altimeters to classify sea ice is different from the common application of altimeter data, because altimeters are often used to invert sea ice thickness. Since 1990, many scholars have successively discovered that different types of sea ice have certain differences in their radar echo waveforms, and this difference can be used to classify sea ice. However, to achieve more accurate classification of sea ice, more advanced radar altimeters are often required.
[0003] On the other hand, traditional machine learning or deep learning methods have the following shortcomings when classifying sea ice: 1. Traditional machine learning methods have complicated parameter tuning and waveform feature combination processes, and their feature learning capabilities are insufficient; 2. The complex network structure makes the model training time long; 3. Different waveform features contribute differently to the identification of sea ice types, and traditional classification methods pay equal attention to all input features. Summary of the invention
[0004] In view of the above-mentioned defects, an embodiment of the present invention discloses a method and device for remote sensing classification of sea ice types, which has a simple network structure, fewer parameters, and improves the diversity of sea ice classification.
[0005] A first aspect of an embodiment of the present invention discloses a remote sensing classification method for sea ice types, comprising:
[0006] Collecting Cryosat-2L1b data of a set period and AARI ice condition map data of the same period as the Cryosat-2L1b data, wherein the Cryosat-2L1b data includes latitude and longitude data and radar echo waveform data, and the AARI ice condition map data includes ice condition maps of the South China Sea and ice condition maps of the North China Sea;
[0007] Processing the Cryosat-2L1b data, calculating waveform features in the Cryosat-2L1b data, and selecting target Cryosat-2L1b data based on the waveform features, wherein the waveform features include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis, and backscattering coefficient;
[0008] Based on the waveform characteristics, the target Cryosat-2L1b data is spatially matched with the AARI ice condition map data to obtain waveform characteristic data with sea ice type information, and the waveform characteristic data is divided into a training set and a test set;
[0009] Constructing a convolutional neural network model, inputting a training set into the convolutional neural network model for training to obtain a sea ice classification model, and inputting a test set into the sea ice classification model to output a classification result;
[0010] The sea ice classification model is corrected based on the classification results.
[0011] As an optional implementation manner, in the first aspect of the embodiment of the present invention, processing the Cryosat-2L1b data and calculating the waveform features in the Cryosat-2L1b data includes:
[0012] Extracting the latitude and longitude data and radar echo waveform data from the Cryosat-2L1b data;
[0013] The waveform characteristics of the radar echo waveform data are calculated.
[0014] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the waveform feature calculation of the radar echo waveform data includes:
[0015] According to the formula Calculate the pulse width, where PP is the pulse width, P max is the maximum energy value, the P i is the echo energy at the i-th range gate.
[0016] As an optional implementation manner, in the first aspect of the embodiment of the present invention, selecting target Cryosat-2L1b data based on the waveform feature includes:
[0017] Select Cryosat-2L1b data with a leading edge width less than or equal to 14 from the Cryosat-2L1b data;
[0018] The ryosat-2L1b data with a pulse width less than 0.3 and a stack standard deviation greater than 4 are removed from the selected ryosat-2L1b data, and the remaining ryosat-2L1b data are used as the target ryosat-2L1b data.
[0019] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the waveform feature data is divided into a training set and a test set, including:
[0020] Divide the waveform feature data into two groups according to the ratio of 70 percent and 30 percent;
[0021] The 70% group is used as the training set and the 30% group is used as the test set.
[0022] As an optional implementation manner, in the first aspect of the embodiment of the present invention, inputting the test set into the sea ice classification model to output the classification result includes:
[0023] Get the vector point vector with B waveform features in the test set;
[0024] Obtain a weight vector of each waveform feature data in the test set, and multiply the weight vector at the corresponding position by the vector point vector to obtain a waveform feature vector after importance correction;
[0025] Based on the waveform feature vector, the preset waveform features are learned through the convolution layer in the sea ice classification model;
[0026] The preset waveform features are input into a sea ice classification model to obtain a classification result.
[0027] As an optional implementation manner, in the first aspect of the embodiment of the present invention, it also includes:
[0028] Gridding the vector points in the classification results into grid data of a preset size;
[0029] The mode of all sea ice types in the grid data is obtained as the sea ice type of the grid data.
[0030] A second aspect of an embodiment of the present invention discloses a sea ice type remote sensing classification device, comprising:
[0031] Data collection module: used to collect Cryosat-2L1b data of a set period and AARI ice condition map data of the same period as the Cryosat-2L1b data, wherein the Cryosat-2L1b data includes latitude and longitude data and radar echo waveform data, and the AARI ice condition map data includes ice condition maps of the South China Sea and ice condition maps of the North China Sea;
[0032] Data processing module: used to process the Cryosat-2L1b data, calculate the waveform features in the Cryosat-2L1b data, and select target Cryosat-2L1b data based on the waveform features, wherein the waveform features include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis and backscattering coefficient;
[0033] Data matching module: used to spatially match the target Cryosat-2L1b data with the AARI ice condition map data based on waveform features to obtain waveform feature data with sea ice type information, and divide the waveform feature data into a training set and a test set;
[0034] Model building module: used to build a convolutional neural network model, input the training set into the convolutional neural network model for training to obtain a sea ice classification model, and input the test set into the sea ice classification model to output the classification result;
[0035] Model calibration module: used for correcting the sea ice classification model based on the classification results.
[0036] As an optional implementation manner, in the second aspect of the embodiment of the present invention, processing the Cryosat-2L1b data and calculating the waveform features in the Cryosat-2L1b data include:
[0037] Extracting the latitude and longitude data and radar echo waveform data from the Cryosat-2L1b data;
[0038] The waveform characteristics of the radar echo waveform data are calculated.
[0039] As an optional implementation, in the second aspect of the embodiment of the present invention, the waveform feature calculation of the radar echo waveform data includes:
[0040] According to the formula Calculate the pulse width, where PP is the pulse width, P max is the maximum energy value, the P i is the echo energy at the i-th range gate.
[0041] As an optional implementation, in the second aspect of the embodiment of the present invention, selecting target Cryosat-2L1b data based on the waveform feature includes:
[0042] Select Cryosat-2L1b data with a leading edge width less than or equal to 14 from the Cryosat-2L1b data;
[0043] The ryosat-2L1b data with a pulse width less than 0.3 and a stack standard deviation greater than 4 are removed from the selected ryosat-2L1b data, and the remaining ryosat-2L1b data are used as the target ryosat-2L1b data.
[0044] As an optional implementation, in the second aspect of the embodiment of the present invention, the waveform feature data is divided into a training set and a test set, including:
[0045] Divide the waveform feature data into two groups according to the ratio of 70 percent and 30 percent;
[0046] The 70% group is used as the training set and the 30% group is used as the test set.
[0047] As an optional implementation, in the second aspect of the embodiment of the present invention, inputting the test set into the sea ice classification model to output the classification result includes:
[0048] Get the vector point vector with B waveform features in the test set;
[0049] Obtain a weight vector of each waveform feature data in the test set, and multiply the weight vector at the corresponding position by the vector point vector to obtain a waveform feature vector after importance correction;
[0050] Based on the waveform feature vector, the preset waveform features are learned through the convolution layer in the sea ice classification model;
[0051] The preset waveform features are input into a sea ice classification model to obtain a classification result.
[0052] As an optional implementation manner, in the second aspect of the embodiment of the present invention, it also includes:
[0053] Gridding the vector points in the classification results into grid data of a preset size;
[0054] The mode of all sea ice types in the grid data is obtained as the sea ice type of the grid data.
[0055] A third aspect of an embodiment of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the sea ice type remote sensing classification method disclosed in the first aspect of the embodiment of the present invention.
[0056] A fourth aspect of an embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the sea ice type remote sensing classification method disclosed in the first aspect of an embodiment of the present invention.
[0057] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:
[0058] In the embodiment of the present invention, by collecting Cryosat-2L1b data and AARI ice condition map data of the same period and performing data processing and spatial matching, waveform features reflecting the physical and geometric characteristics of sea ice, such as pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis and backscattering coefficient, etc., can be extracted. The waveform features are combined with the sea ice type information provided by the AARI ice condition map, providing rich training data for the convolutional neural network (CNN) model, so that the model can learn the complex relationship between sea ice type and waveform features; constructing and training the CNN model, and finally realizing accurate classification of sea ice types; traditional sea ice monitoring methods often require a lot of manpower, material resources and time investment, and the monitoring range is limited, while the present invention combines satellite remote sensing data and machine learning technology, and can efficiently classify sea ice types, which not only improves the accuracy of classification, but also provides strong support for sea ice monitoring and research. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0060] Figure 1 It is a flow chart of a remote sensing classification method for sea ice types disclosed in an embodiment of the present invention;
[0061] Figure 2 This is a network structure diagram of the convolutional neural network model provided by an embodiment of the present invention.
[0062] Figure 3 is a schematic structural diagram of a sea ice type remote sensing classification device provided by an embodiment of the present invention;
[0063] Figure 4 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0065] It should be noted that the terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present invention and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0066] The embodiment of the present invention discloses a remote sensing classification method, device, electronic device and storage medium for sea ice type. In the embodiment, by collecting Cryosat-2L1b data and AARI ice condition map data of the same period and performing data processing and spatial matching, waveform features reflecting the physical and geometric characteristics of sea ice, such as pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis and backscattering coefficient, etc., can be extracted. The waveform features are combined with the sea ice type information provided by the AARI ice condition map, providing rich training data for a convolutional neural network (CNN) model, so that the model can learn the complex relationship between sea ice type and waveform features; constructing and training a CNN model, and finally realizing accurate classification of sea ice types; traditional sea ice monitoring methods often require a lot of manpower, material resources and time investment, and the monitoring range is limited, while the present invention combines satellite remote sensing data and machine learning technology, can efficiently classify sea ice types, not only improves the accuracy of classification, but also provides strong support for sea ice monitoring and research.
[0067] Embodiment 1
[0068] See also Figure 1 , Figure 1 It is a flow chart of a remote sensing classification method for sea ice types disclosed in an embodiment of the present invention. Among them, the execution subject of the method described in the embodiment of the present invention is an execution subject composed of software and / or hardware, and the execution subject can receive relevant information by wired or / and wireless means, and can send certain instructions. Of course, it can also have certain processing functions and storage functions. The execution subject can control multiple devices, such as a remote physical server or cloud server and related software, or it can be a local host or server and related software that performs related operations on a device placed somewhere. In some scenarios, multiple storage devices can also be controlled, and the storage devices can be placed in the same place or different places as the devices. For example Figure 1 As shown, the sea ice type remote sensing classification method includes the following steps:
[0069] 101. Collect Cryosat-2L1b data of a set period and AARI ice condition map data of the same period as the Cryosat-2L1b data, wherein the Cryosat-2L1b data includes latitude and longitude data and radar echo waveform data, and the AARI ice condition map data includes an ice condition map of the South China Sea and an ice condition map of the North China Sea.
[0070] Collect Cryosat-2 satellite L1b-level data for a specific period (such as a season or year). These data include latitude and longitude information (for positioning) and radar echo waveform data (for analyzing the physical characteristics of sea ice). At the same time, collect ice condition map data provided by AARI (which may be a specific agency, such as the Russian Arctic and Antarctic Research Institute) for the same period as Cryosat-2 L1b data. These data are divided into ice condition maps for the southern and northern seas, providing intuitive information on the distribution and type of sea ice.
[0071] CryoSat-2 is one of the missions of the European Space Agency (ESA) Earth Exploration Program. The purpose of this mission is to determine the changing trends of the cryosphere. The most important of these is to verify the annual reduction of Arctic sea ice caused by global warming and to clarify the contribution of the Antarctic and Greenland ice sheets to global sea level rise. The observation data range of CryoSat-2 reaches up to the area of 88°N north latitude, with a full cycle of 369 days and a small cycle of 30 days. The vertical measurement accuracy of the synthetic aperture interferometric radar altimeter (SIRAL) carried on the satellite reaches 1 to 3 cm, and the delayed Doppler radar altimeter (DDA) technology is used to reduce the satellite's ground footprint to about 0.3 km along the track and about 1.5 km across the track. In addition, SIRAL performs multi-view processing on surface points to reduce the noise caused by radar spots, thereby improving the measurement accuracy of sea level height.
[0072] AARI ice condition map data is AARI (Arctic and Antarctic Institute) SIGRID-3 data (shapefile format, including sea ice type information), which provides ice condition maps of the Arctic and Antarctic waters once a week. Its sea ice type products are obtained using optical, near-infrared, SAR satellite data and ship navigation data. The AARISIGRID-3 data product (shapefile format) is used, which provides five types of sea ice data: Nile ice, primary ice, one-year ice, multi-year ice and fixed ice. Since Cryosat-2 has a large spatial coverage and one-year ice and multi-year ice are the main ice types in the polar regions, we only use the data of one-year ice and multi-year ice.
[0073] This step collects and processes Cryosat-2L1b data and AARI ice condition map data, and can subsequently extract waveform features that reflect the physical and geometric characteristics of sea ice, such as pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis, and backscatter coefficient. Traditional sea ice detection methods require a lot of manpower, material resources, and time investment, and the monitoring range is prioritized. The present invention uses satellite remote sensing data and machine learning technology to achieve rapid and accurate classification of sea ice types, greatly improving the efficiency of sea ice monitoring. It not only helps to timely understand the distribution and changes of sea ice, but also provides important data support for marine meteorological forecasts, maritime shipping safety, polar resource development and other fields.
[0074] 102. Process the Cryosat-2L1b data, calculate waveform features in the Cryosat-2L1b data, and select target Cryosat-2L1b data based on the waveform features, wherein the waveform features include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis, and backscattering coefficient.
[0075] This step processes the collected Cryosat-2L1b data and extracts the characteristics of the radar echo waveform. These characteristics include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis and backscatter coefficient, which can reflect the physical and geometric characteristics of sea ice. Based on the extracted waveform characteristics, the Cryosat-2L1b data related to the target research are selected as the object of subsequent analysis.
[0076] Specifically in this step, the Cryosat-2L1b data is processed and waveform features in the Cryosat-2L1b data are calculated, including: extracting longitude and latitude data and radar echo waveform data in the Cryosat-2L1b data; and calculating waveform features of the radar echo waveform data.
[0077] First, the longitude and latitude information of each data point is extracted. The longitude and latitude data is the precise location of the data point in the Global Positioning System (GPS), which is crucial for subsequent spatial matching and geographic location analysis. At the same time, the radar echo waveform data is extracted from the data set. The radar echo waveform data is the reflected signal received by the radar altimeter carried by the Cryosat-2 satellite during the flight. They contain rich information about the surface (including sea ice). After extracting the radar echo waveform data, the next step is to calculate the waveform characteristics. Pulse width: represents the duration of the waveform signal, which can reflect the thickness and roughness of the sea ice. Leading edge width and trailing edge width: represent the width of the starting and ending parts of the waveform signal respectively. Stack standard deviation: measures the degree of change of the waveform signal at different time points. Maximum energy value: represents the maximum energy point in the waveform signal. Stack kurtosis: describes the kurtosis of the waveform signal, that is, the sharpness of the waveform signal. Backscatter coefficient: represents the intensity of the radar signal reflected back by the surface (including sea ice), which is a commonly used parameter in radar remote sensing. By processing the Cryosat-2L1b data, extracting the longitude and latitude data and radar echo waveform data, and calculating the waveform features, we provide a foundation for subsequent spatial matching, data classification, and model training.
[0078] The waveform feature calculation of the radar echo waveform data includes:
[0079] According to the formula Calculate the pulse width, where PP is the pulse width, P max is the maximum energy value, the P i is the echo energy at the ith range gate. "Range gate" refers to a series of equally spaced sampling points or intervals into which radar signals are divided during propagation. Each range gate corresponds to a specific distance range. Also known as pulse width, it is the ratio of the maximum energy in the echo waveform to the cumulative energy sum, P max =max(P i ),i=1,2,3,L,256.
[0080] Furthermore, LEW is the number of range gates between 1% and 99% of the maximum echo energy (starting from the first range gate greater than 1% of the maximum energy to the first range gate greater than 99% of the maximum energy; TEW is the number of range gates between 99% and 1% of the maximum echo energy (starting from the last range gate greater than 99% of the maximum energy to the last range gate greater than 1% of the maximum energy). The calculation formulas for LEW and TEW are as follows:
[0081] A1=99%×P max ,A2=1%×P max ,
[0082] LeW=Bin(A1)-Bin(A2),
[0083] Wherein, Bin(A1) represents the position of the first range gate greater than 99% of the maximum energy value, Bin(A2) represents the position of the first range gate greater than 1% of the maximum energy value, and LeW is the range gate width between A1 and A2.
[0084] A′1=1%×P max ,A′2=99%×P max ,
[0085] TEW=Bin′(A′1)-Bin′(A′2)
[0086] Wherein, Bin(A′1) represents the position of the last range gate greater than 1% of the maximum energy value, Bin(A′2)5 represents the position of the last range gate greater than 99% of the maximum energy value, and TEW is the range gate width between A′1 and A′2.
[0087] SSD is the standard deviation of the echo waveform energy value on the common plane formed by the Doppler waveform at different incident angles. The calculation formula of SSD is as follows:
[0088]
[0089] In the formula, It is the average value of the energy of the waveform at 256 range gates.
[0090] SK is a characteristic number that indicates the peak value in the waveform energy distribution. The calculation formula of SK is as follows:
[0091]
[0092] Where E(g) represents the expectation, and σ is the standard deviation of the echo waveform energy value.
[0093] In addition, target Cryosat-2L1b data are selected based on the waveform characteristics, including: selecting Cryosat-2L1b data with a leading edge width less than or equal to 14 from the Cryosat-2L1b data; removing ryosat-2L1b data with a pulse width less than 0.3 and a stack standard deviation greater than 4 from the selected ryosat-2L1b data, and using the remaining ryosat-2L1b data as target ryosat-2L1b data.
[0094] From all the Cryosat-2L1b data, select those with a leading edge width less than or equal to 14. The leading edge width is an important parameter in the waveform characteristics, reflecting the width of the starting part of the waveform signal. Data with a leading edge width less than or equal to 14 are more likely to meet the needs of our subsequent analysis. Among the data obtained through preliminary screening, it is necessary to further eliminate those data with a pulse width less than 0.3 and a stack standard deviation greater than 4. The pulse width is the duration of the waveform signal, while the stack standard deviation measures the degree of change of the waveform signal at different time points. It is assumed that those data with too small pulse widths and too large stack standard deviations may be caused by noise or other interference factors, so they need to be eliminated. Through reasonable screening conditions and accurate waveform feature calculations, representative target data can be obtained, providing strong support for the subsequent remote sensing classification of sea ice types.
[0095] 103. Based on the waveform features, spatially match the target Cryosat-2L1b data with the AARI ice condition map data to obtain waveform feature data with sea ice type information, and divide the waveform feature data into a training set and a test set.
[0096] The selected target Cryosat-2L1b data is spatially matched with the AARI ice map data to ensure that each Cryosat-2L1b data point has corresponding ice map information, so that each data point has information about the type of sea ice. Spatial matching is to match the longitude and latitude of the two data.
[0097] Furthermore, the waveform feature data is divided into two groups according to the ratio of 70% and 30%; the 70% group is used as the training set and the 30% group is used as the test set. The training set is used to train the machine learning model, and the remaining 30% of the data will be used as the test set to evaluate the performance of the model. When dividing the data, make sure to use random sampling to avoid any potential bias, which is achieved by setting the random seed.
[0098] 104. Construct a convolutional neural network model, input the training set into the convolutional neural network model for training to obtain a sea ice classification model, and input the test set into the sea ice classification model to output a classification result.
[0099] This step builds a suitable convolutional neural network (CNN) model based on the Tensorflow deep learning library and Python language. CNN performs well in processing image and waveform data and is suitable for extracting features from radar echo waveforms and classifying sea ice types. The network structure diagram of the convolutional neural network model is shown in the figure below. Figure 2As shown in the figure, the network model is iteratively trained and hyperparameters are tuned (such as the number of convolutional layers, attention module parameters, number of network iterations, learning rate, etc.) using the training set. The test samples are classified using the trained model, and the overall accuracy and Kappa coefficient are used to evaluate the model performance. Among them, the overall accuracy indicates the proportion of correctly classified samples to the total number of samples, and the Kappa coefficient is an indicator for consistency testing and can also be used to measure the classification effect.
[0100] Specifically, the test set is input into the sea ice classification model to output the classification result, including: obtaining a vector point vector with B waveform features in the test set; obtaining a weight vector of each waveform feature data in the test set, multiplying the weight vector at the corresponding position with the vector point vector to obtain a waveform feature vector after importance correction; based on the waveform feature vector and learning the preset waveform features through the convolution layer in the sea ice classification model; inputting the preset waveform features into the sea ice classification model to obtain the classification result.
[0101] Get vector point vectors with B waveform features from the test set. These vector point vectors are the representation of the test set data in the feature space. Each vector contains B eigenvalues, corresponding to B different waveform features. Get the weight vector of each waveform feature data in the test set. The weight vector is used to represent the importance of each feature for the classification task. Multiply the weight vector of the corresponding position with the vector point vector to calculate the waveform feature vector after importance correction. The purpose is to weight the original feature vector according to the importance of the feature to improve the accuracy of classification. The convolution layer is an important component for feature extraction in the neural network. It learns the local features in the input data through convolution operations. In the process of learning preset waveform features by the convolution layer, the model will learn the preset waveform features, and after learning by the convolution layer, the extracted features are input to the subsequent layers of the sea ice classification model (such as the fully connected layer, softmax layer, etc.) to obtain the final classification results.
[0102] The vector points in the classification results are gridded into grid data of a preset size, and the mode of all sea ice types in the grid data is obtained as the sea ice type of the grid data. The preset size is a spatial resolution of 25kmx25km.
[0103] 105. Correct the sea ice classification model based on the classification result.
[0104] Embodiment 2
[0105] See also Figure 3 , Figure 3 Schematic diagram of the structure of a remote sensing classification device for sea ice types disclosed in an embodiment of the present invention. Figure 3As shown, the sea ice type remote sensing classification device may include: a data collection module 301, a data processing module 302, a data matching module 303, a model building module 304 and a model calibration module 305, wherein the data collection module 301 is used to collect Cryosat-2L1b data of a set period and AARI ice condition map data of the same period as the Cryosat-2L1b data, the Cryosat-2L1b data includes longitude and latitude data and radar echo waveform data, and the AARI ice condition map data includes ice condition maps of the South China Sea and ice condition maps of the North China Sea; the data processing module 302 is used to process the Cryosat-2L1b data, calculate the waveform features in the Cryosat-2L1b data, and based on the waveform features Select target Cryosat-2L1b data, the waveform characteristics include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis and backscatter coefficient; data matching module 303: used to spatially match the target Cryosat-2L1b data with AARI ice condition map data based on the waveform characteristics to obtain waveform characteristic data with sea ice type information, and divide the waveform characteristic data into a training set and a test set; model building module 304: used to build a convolutional neural network model, input the training set into the convolutional neural network model for training to obtain a sea ice classification model, and input the test set into the sea ice classification model to output the classification result; model calibration module 305: used to calibrate the sea ice classification model based on the classification result.
[0106] The data processing module 302 processes the Cryosat-2L1b data and calculates the waveform features in the Cryosat-2L1b data, including: extracting the longitude and latitude data and radar echo waveform data in the Cryosat-2L1b data; and calculating the waveform features of the radar echo waveform data. The waveform features of the radar echo waveform data are calculated by: according to the formula Calculate the pulse width, where PP is the pulse width, P max is the maximum energy value, the P i is the echo energy at the i-th range gate.
[0107] The data processing module 302 selects target Cryosat-2L1b data based on the waveform characteristics, including selecting Cryosat-2L1b data with a leading edge width less than or equal to 14 from the Cryosat-2L1b data; removing ryosat-2L1b data with a pulse width less than 0.3 and a stack standard deviation greater than 4 from the selected ryosat-2L1b data, and using the remaining ryosat-2L1b data as target ryosat-2L1b data.
[0108] In the data matching module 303, the waveform feature data is divided into a training set and a test set, including: dividing the waveform feature data into two groups according to a ratio of 70% and 30%; using the 70% group as the training set and the 30% group as the test set.
[0109] In the model building module 304, the test set is input into the sea ice classification model to output the classification result, including: obtaining a vector point vector with B waveform features in the test set; obtaining a weight vector of each waveform feature data in the test set, multiplying the weight vector at the corresponding position with the vector point vector to obtain a waveform feature vector after importance correction; based on the waveform feature vector and learning the preset waveform features through the convolution layer in the sea ice classification model; inputting the preset waveform features into the sea ice classification model to obtain the classification result.
[0110] The embodiment may further include a gridding module for gridding the vector points in the classification results into grid data of a preset size; and obtaining the mode of all sea ice types in the grid data as the sea ice type of the grid data.
[0111] Embodiment 3
[0112] See also Figure 4 , Figure 4 Schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device may be a computer, a server, etc. Of course, in certain circumstances, it may also be a smart device such as a mobile phone, a tablet computer, a monitoring terminal, and an image acquisition device with processing functions. Figure 4 As shown, the electronic device may include:
[0113] A memory 401 storing executable program codes;
[0114] a processor 402 coupled to the memory 401;
[0115] The processor 402 calls the executable program code stored in the memory 401 to execute part or all of the steps in the sea ice type remote sensing classification method in the first embodiment.
[0116] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute part or all of the steps in the remote sensing classification method for sea ice types in the first embodiment.
[0117] The embodiment of the present invention further discloses a computer program product, wherein when the computer program product is run on a computer, the computer is enabled to execute part or all of the steps in the remote sensing classification method for sea ice types in the first embodiment.
[0118] An embodiment of the present invention further discloses an application publishing platform, wherein the application publishing platform is used to publish a computer program product, wherein when the computer program product runs on a computer, the computer executes part or all of the steps in the sea ice type remote sensing classification method in embodiment one.
[0119] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the processes does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0120] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0121] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0122] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several requests for a computer device (which can be a personal computer, a server or a network device, etc., specifically a processor in a computer device) to perform some or all of the steps of the method described in each embodiment of the present invention.
[0123] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.
[0124] A person of ordinary skill in the art can understand that some or all of the steps in the various methods of the embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.
[0125] The sea ice type remote sensing classification method, device, electronic device and storage medium disclosed in the embodiments of the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. A remote sensing classification method for sea ice types, characterized in that: include: Collecting Cryosat-2L1b data of a set period and AARI ice condition map data of the same period as the Cryosat-2L1b data, wherein the Cryosat-2L1b data includes latitude and longitude data and radar echo waveform data, and the AARI ice condition map data includes ice condition maps of the South China Sea and ice condition maps of the North China Sea; Processing the Cryosat-2L1b data, calculating waveform features in the Cryosat-2L1b data, and selecting target Cryosat-2L1b data based on the waveform features, wherein the waveform features include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis, and backscattering coefficient; Based on the waveform characteristics, the target Cryosat-2L1b data is spatially matched with the AARI ice condition map data to obtain waveform characteristic data with sea ice type information, and the waveform characteristic data is divided into a training set and a test set; Constructing a convolutional neural network model, inputting a training set into the convolutional neural network model for training to obtain a sea ice classification model, and inputting a test set into the sea ice classification model to output a classification result; The sea ice classification model is corrected based on the classification results.
2. The remote sensing classification method for sea ice types according to claim 1, characterized in that: Processing the Cryosat-2L1b data and calculating waveform features in the Cryosat-2L1b data includes: Extracting the latitude and longitude data and radar echo waveform data from the Cryosat-2L1b data; The waveform characteristics of the radar echo waveform data are calculated.
3. The remote sensing classification method for sea ice types according to claim 2, characterized in that: The step of calculating waveform characteristics of the radar echo waveform data comprises: According to the formula Calculate the pulse width, where PP is the pulse width, P max is the maximum energy value, the P i is the echo energy at the i-th range gate.
4. The sea ice type remote sensing classification method according to claim 2, characterized in that: Selecting target Cryosat-2L1b data based on the waveform features includes: Select Cryosat-2L1b data with a leading edge width less than or equal to 14 from the Cryosat-2L1b data; The ryosat-2L1b data with a pulse width less than 0.3 and a stack standard deviation greater than 4 are removed from the selected ryosat-2L1b data, and the remaining ryosat-2L1b data are used as the target ryosat-2L1b data.
5. The remote sensing classification method for sea ice types according to claim 1, characterized in that: The waveform feature data is divided into a training set and a test set, including: Divide the waveform feature data into two groups according to the ratio of 70 percent and 30 percent; Use 70% of the group as the training set and 30% of the group as the test set.
6. The sea ice type remote sensing classification method according to claim 1, characterized in that: The test set is input into the sea ice classification model to output the classification results, including: Get the vector point vector with B waveform features in the test set; Obtain a weight vector of each waveform feature data in the test set, and multiply the weight vector at the corresponding position by the vector point vector to obtain a waveform feature vector after importance correction; Based on the waveform feature vector, the preset waveform features are learned through the convolution layer in the sea ice classification model; The preset waveform features are input into a sea ice classification model to obtain a classification result.
7. The remote sensing classification method for sea ice types according to claim 6, characterized in that: Also includes: Gridding the vector points in the classification results into grid data of a preset size; The mode of all sea ice types in the grid data is obtained as the sea ice type of the grid data.
8. A sea ice type remote sensing classification device, characterized in that: include: Data collection module: used to collect Cryosat-2L1b data of a set period and AARI ice condition map data of the same period as the Cryosat-2L1b data, wherein the Cryosat-2L1b data includes latitude and longitude data and radar echo waveform data, and the AARI ice condition map data includes ice condition maps of the South China Sea and ice condition maps of the North China Sea; Data processing module: used to process the Cryosat-2L1b data, calculate the waveform features in the Cryosat-2L1b data, and select target Cryosat-2L1b data based on the waveform features, wherein the waveform features include pulse width, leading edge width, trailing edge width, stack standard deviation, maximum energy value, stack kurtosis and backscattering coefficient; Data matching module: used to spatially match the target Cryosat-2L1b data with the AARI ice condition map data based on waveform features to obtain waveform feature data with sea ice type information, and divide the waveform feature data into a training set and a test set; Model building module: used to build a convolutional neural network model, input the training set into the convolutional neural network model for training to obtain a sea ice classification model, and input the test set into the sea ice classification model to output the classification result; Model calibration module: used for correcting the sea ice classification model based on the classification results.
9. An electronic device, characterized in that: include: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the sea ice type remote sensing classification method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the sea ice type remote sensing classification method according to any one of claims 1 to 7.