SAR crop classification method based on multi-scale time-varying information extraction network

Abstract

The application relates to a SAR crop classification method based on a multi-scale time-varying information extraction network, which comprises the following steps: acquiring time-series SAR images to be classified; pre-processing the time-series SAR images to be classified, and registering according to a pre-processing result and preset label data to obtain slice data to be classified; inputting the slice data to be classified into a pre-constructed classification network model to obtain a crop classification result of the time-series SAR images to be classified, wherein the pre-constructed classification network model is obtained by training an initial network model which is fused and constructed by a multi-scale time-varying information module and a multi-scale spatial information module. Thus, the information in the time-series SAR images is classified by constructing the multi-scale time-varying information network model, efficient and accurate automatic and intelligent classification of crops is realized under the condition of low parameter quantity, and all-weather crop planting area monitoring can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar remote sensing image processing, and particularly relates to a SAR crop classification method based on a multi-scale time-varying information extraction network. BACKGROUND

[0002] At present, the crop planting area monitoring method using optical remote sensing has been widely developed and applied, but due to the fact that optical remote sensing is easily affected by weather such as rain and clouds, it is difficult to obtain complete and continuous observation data, and the spectral sensitivity is gradually saturated with the growth of vegetation, which makes the optical remote sensing method have certain limitations in the practical application of crop monitoring. SAR (Synthetic Aperture Radar, time-series synthetic aperture radar) is a kind of active electromagnetic wave emission remote sensing system, which has all-weather observation capability and is very suitable for continuous observation of crops, and SAR is more sensitive to the structural characteristics and dielectric properties of crops, which better compensates for the limitations of optical remote sensing. However, it is difficult to distinguish between sowing and harvesting periods of farmland using single-phase SAR data at a certain time, and it is also difficult to classify crops with different planting times into one category. Moreover, crops have different characteristics at different times, and continuous observation of crops is beneficial to accurately extracting crop information. At the same time, with more and more SAR sensors, the SAR field has accumulated a certain amount of high-temporal-resolution time-series images, which lays a data foundation for using time-series SAR images for crop planting area classification research. Therefore, it is of great research significance and application value to use time-series SAR images for continuous monitoring of crops.

[0003] In related technologies, deep learning methods are usually used to classify crops collected by time-series SAR images.

[0004] However, although the deep learning technology in the above technical means can automatically learn the deep semantic features of crops in training, it also has problems such as the need for other auxiliary data, low utilization rate of network model for crop time-series information, and dependence on more two-dimensional structures, which makes the model parameter amount large, and needs to be solved. SUMMARY

[0005] The present application provides a SAR crop classification method based on a multi-scale time-varying information extraction network, to solve the problems of related technologies, such as the need for other auxiliary data, low utilization rate of network model for crop time-series information, and dependence on more two-dimensional structures, which makes the model parameter amount large, fully excavates the multi-scale time-varying information in the time-series SAR image, reduces the dependence on other auxiliary data, realizes the automatic and intelligent efficient and accurate classification of crops under the condition of low parameter amount, and realizes the all-weather crop planting area monitoring.

[0006] To achieve the above object, the first aspect of the present application provides a SAR crop classification method based on a multi-scale time-varying information extraction network, comprising the following steps:

[0007] Obtaining a to-be-classified time-series SAR image;

[0008] Pretreating the to-be-classified time-series SAR image, and registering according to a pretreatment result and preset label data to obtain to-be-classified slice data;

[0009] Inputting the to-be-classified slice data into a pre-constructed classification network model to obtain a crop classification result of the to-be-classified time-series SAR image, wherein the pre-constructed classification network model is obtained by training an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module.

[0010] According to an embodiment of the present application, before the to-be-classified slice data is input into the pre-constructed classification network model, it further comprises:

[0011] Obtaining a plurality of target time-series SAR images;

[0012] Pretreating the plurality of target time-series SAR images, and registering the pretreated plurality of target time-series SAR images and the preset label data to obtain a slice data set;

[0013] Dividing the slice data set into a training set and a test set, and training an initial network model constructed by fusing the multi-scale time-varying information module and the multi-scale spatial information module using the training set to obtain an initial classification network model;

[0014] Testing the initial classification network model using the test set, and taking the initial classification network model as the pre-constructed classification network model when the test result meets a preset classification standard.

[0015] According to an embodiment of the present application, after the initial classification network model is tested using the test set, it further comprises:

[0016] If the test result does not meet the preset classification standard, adjusting the division ratio of the training set and the test set, and retraining the initial network model according to the newly divided test set until the new test result meets the preset classification standard to obtain the pre-constructed classification network model.

[0017] According to an embodiment of the present application, the pretreatment of the to-be-classified time-series SAR image comprises:

[0018] The orbit correction processing is performed on the to-be-classified time-series SAR image to obtain a first processed image;

[0019] The radiation calibration processing is performed on the first processed image to obtain a second processed image;

[0020] The geometric correction processing is performed on the second processed image to obtain a third processed image;

[0021] The coherent speckle filtering processing is performed on the third processed image to obtain a fourth processed image;

[0022] The registration is performed on the fourth processed image to obtain the preprocessing result.

[0023] According to an embodiment of the present application, the multi-scale time-varying information module is:

[0024]

[0025] Wherein, v is the scattering difference of different time phase radar images, T n is the time phase at t n moment.

[0026] According to the SAR crop classification method based on the multi-scale time-varying information extraction network provided in the embodiment of the present application, the classification network model is trained and verified in advance to ensure the accuracy of the classification network model result, and then the classification slice data obtained by processing the SAR image is input into the classification network model, and after the time-series information extraction and noise suppression processing of the classification network model, the final time-series SAR image crop classification result is obtained. Therefore, by constructing the multi-scale time-varying information network model, the information in the time-series SAR image is classified, the efficient and accurate classification of the crop automation and intelligence is realized under the condition of low parameter quantity, and the all-weather crop planting area monitoring is realized.

[0027] To achieve the above purpose, the second aspect embodiment of the present application provides a SAR crop classification device based on a multi-scale time-varying information extraction network, comprising:

[0028] The acquisition module is configured to acquire a to-be-classified time-series SAR image;

[0029] The preprocessing module is configured to preprocess the to-be-classified time-series SAR image, and perform registration according to a preprocessing result and preset label data to obtain to-be-classified slice data;

[0030] The classification module is configured to input the to-be-classified slice data into a pre-constructed classification network model to obtain a crop classification result of the to-be-classified time-series SAR image, wherein the pre-constructed classification network model is obtained by training an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module.

[0031] According to one embodiment of the present application, before the to-be-classified slice data is input into the pre-constructed classification network model, the classification module is further configured to:

[0032] Obtain a plurality of target time-series SAR images;

[0033] Preprocess the plurality of target time-series SAR images, and register the preprocessed plurality of target time-series SAR images and the preset label data to obtain a slice data set;

[0034] Divide the slice data set into a training set and a test set, and train an initial network model constructed by fusing the multi-scale time-varying information module and the multi-scale spatial information module using the training set to obtain an initial classification network model;

[0035] Test the initial classification network model using the test set, and use the initial classification network model as the pre-constructed classification network model when the test result meets a preset classification standard.

[0036] According to one embodiment of the present application, after testing the initial classification network model using the test set, the classification module is further configured to:

[0037] If the test result does not meet the preset classification standard, adjust the division ratio of the training set and the test set, and retrain the initial network model according to the newly divided test set until the new test result meets the preset classification standard to obtain the pre-constructed classification network model.

[0038] According to one embodiment of the present application, the preprocessing module is specifically configured to:

[0039] Perform orbit correction processing on the to-be-classified time-series SAR image to obtain a first processed image;

[0040] Perform radiation calibration processing on the first processed image to obtain a second processed image;

[0041] Perform geometric correction processing on the second processed image to obtain a third processed image;

[0042] Perform coherent speckle filtering processing on the third processed image to obtain a fourth processed image;

[0043] Register the fourth processed image to obtain the preprocessing result.

[0044] According to one embodiment of the present application, the multi-scale time-varying information module is:

[0045]

[0046] wherein v is the scattering difference of radar images of different time phases, T n is the time phase at t n time.

[0047] The SAR crop classification device based on the multi-scale time-varying information extraction network according to the embodiment of the present application can ensure the accuracy of the classification network model result by training and verifying the classification network model in advance, and then input the classification slice data obtained by processing the SAR image into the classification network model, so that the final time sequence SAR image crop classification result is obtained after the time sequence information extraction and noise suppression processing of the classification network model. Therefore, by constructing the multi-scale time-varying information network model to classify the information in the time sequence SAR image, the automatic and intelligent efficient and accurate classification of crops is realized under the condition of low parameter quantity, so that the all-weather crop planting area monitoring can be realized.

[0048] To achieve the above object, the third aspect of the present application provides an electronic device, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the program to realize the SAR crop classification method based on the multi-scale time-varying information extraction network as described in the above embodiments.

[0049] To achieve the above object, the fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the SAR crop classification method based on the multi-scale time-varying information extraction network as described in the above embodiments.

[0050] To achieve the above object, the fifth aspect of the present application provides a computer program product, comprising a computer program, which is executed by a processor to realize the SAR crop classification method based on the multi-scale time-varying information extraction network as described in the above embodiments.

[0051] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0052] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0053] Figure 1 A flowchart of a SAR crop classification method based on a multi-scale time-varying information extraction network provided by an embodiment of the present application is shown in the figure;

[0054] Figure 2 A flowchart of a SAR crop classification method based on a multi-scale time-varying information extraction network according to an embodiment of the present application is provided.

[0055] Figure 3 A structural diagram of a classification network model according to an embodiment of the present application is provided.

[0056] Figure 4 A structural diagram of a multi-scale time-varying information module according to an embodiment of the present application is provided.

[0057] Figure 5 A flowchart of a multi-scale spatial information module according to an embodiment of the present application is provided.

[0058] Figure 6 A block diagram of a SAR crop classification device based on a multi-scale time-varying information extraction network according to an embodiment of the present application is provided.

[0059] Figure 7 A structural diagram of an electronic device according to an embodiment of the present application is provided.

[0060] Reference Signs:

[0061] 10 - SAR crop classification device based on a multi-scale time-varying information extraction network, 100 - acquisition module, 200 - preprocessing module, 300 - classification module, 701 - memory, 702 - processor, 703 - communication interface. DETAILED DESCRIPTION

[0062] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0063] A SAR crop classification method based on a multi-scale time-varying information extraction network according to an embodiment of the present application is described below with reference to the accompanying drawings. First, a SAR crop classification method based on a multi-scale time-varying information extraction network according to an embodiment of the present application is described with reference to the accompanying drawings.

[0064] To address the problems in the background technology, such as the need for auxiliary data, low utilization rate of crop time-series information by network models, and large number of model parameters due to reliance on two-dimensional structures, this invention provides a SAR crop classification method based on a multi-scale time-varying information extraction network. By pre-training and validating the classification network model, the accuracy of the results is ensured. Then, the classification slice data after SAR image processing is input into the classification network model. After time-series information extraction and noise suppression processing by the classification network model, the final time-series SAR image crop classification result is obtained. Thus, by constructing a multi-scale time-varying information network model to classify information in time-series SAR images, efficient and accurate automated and intelligent classification of crops is achieved under low parameter conditions, enabling all-day, all-weather monitoring of crop planting areas.

[0065] Figure 1 This is a flowchart of a SAR crop classification method based on a multi-scale time-varying information extraction network provided by an embodiment of the present invention.

[0066] like Figure 1 As shown, the SAR crop classification method based on a multi-scale time-varying information extraction network includes the following steps:

[0067] In step S101, the time-series SAR image to be classified is acquired.

[0068] Specifically, time-series SAR imagery consists of surface images acquired by SAR through the transmission and reception of microwave signals at different time points. By imaging the same area at different points in time, a time series can be formed for monitoring changes in crops on the surface. In this embodiment of the invention, SAR imagery can be acquired by downloading satellite data provided by a satellite remote sensing platform, for example, by downloading SAR imagery transmitted from satellites through the European Space Agency platform.

[0069] In step S102, the temporal SAR image to be classified is preprocessed, and the preprocessing results and preset label data are registered to obtain the slice data to be classified.

[0070] Specifically, such as Figure 2 As shown, Figure 2The flowchart of the SAR crop classification method based on the multi-scale time-varying information extraction network according to one specific embodiment of the present application is provided, and the embodiment of the present application sequentially performs orbit correction, radiation calibration, geometric correction, coherent spot filtering and registration and other preprocessing on the time-series SAR image to be classified, so as to suppress the noise of the time-series SAR image. At the same time, the time-series SAR image is registered to ensure that the multi-temporal images are aligned in the same coordinate system, improve the recognition accuracy, and crop the preprocessed time-series SAR image and the registered label data to obtain a slice data set to be classified. The cropping of the preprocessed time-series SAR image and the registered label data can be cropped according to the same size, for example, both are cropped according to the size of 256*256 pixels to obtain the slice data to be classified.

[0071] As a possible implementation manner, in some embodiments, the preprocessing of the time-series SAR image to be classified includes: performing orbit correction processing on the time-series SAR image to be classified to obtain a first processed image; performing radiation calibration processing on the first processed image to obtain a second processed image; performing geometric correction processing on the second processed image to obtain a third processed image; performing coherent spot filtering processing on the third processed image to obtain a fourth processed image; and performing registration on the fourth processed image to obtain a preprocessing result.

[0072] Specifically, the embodiment of the present application first uses orbit data to perform orbit correction processing on the time-series SAR image to be classified to ensure the geometric accuracy of the SAR image, and then performs radiation calibration processing on the orbit correction result to convert the SAR data into a backscattering coefficient. Then, geometric correction processing based on the distance Doppler model is performed on the processing result to further ensure the geometric accuracy of the data. The processed image is subjected to coherent spot filtering processing using a delicate Lee filtering algorithm, while maintaining the detail features such as points, lines and edges in the image. Based on the feature points in the SAR image, the plurality of images in the time-series SAR image to be processed are registered to make the plurality of images in the spatial position in the aligned state, and the final preprocessing result is obtained.

[0073] In step S103, the slice data to be classified is input into a pre-constructed classification network model to obtain a crop classification result of the time-series SAR image to be classified, wherein the pre-constructed classification network model is obtained by training an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module.

[0074] In some embodiments, the multi-scale time-varying information module is:

[0075]

[0076] wherein v is the scattering difference of different temporal radar images, T n is tn The phase of the time.

[0077] It should be noted that, as Figure 3 indicated, Figure 3 is a structural diagram of a classification network model provided according to an embodiment of the present application. The present application fuses a multi-scale time-varying information module and a multi-scale spatial information module in the classification network model for time-varying features on time-series SAR images, wherein the multi-scale time-varying information module uses a time difference method to process time-series SAR images.

[0078] Specifically, as Figure 4 indicated, Figure 4 is a structural diagram of a multi-scale time-varying information module provided according to an embodiment of the present application. The main framework of the multi-scale time-varying information extraction model is formed by difference extraction between two phases at different time intervals and combination of the extraction results at different time intervals. The processing mode with respect to T1 phase is shown in formula (1), and the processing mode with respect to T2 and even T n The difference information extraction method of the phase can be analogized by referring to formula (1). The multi-scale time-varying information extraction module is built by combining these different scale time-varying information and processing by using one-dimensional convolution. After processing by the multi-scale time-varying information module, the classification of crops in the SAR image can be realized, but the classification result is often fragmented due to the influence of SAR image speckle noise, which is difficult to be practically used. Therefore, the embodiment of the present application further fuses a multi-scale spatial information module. The feature map obtained after the multi-scale time-varying information extraction module is input into the multi-scale spatial information extraction module, noise suppression is performed based on spatial semantic information, and convolution calculation is performed on the output results of the two modules to obtain the crop classification result of the time-series SAR image to be classified. The multi-scale spatial information extraction module can be an Atrous Spatial Pyramid Pooling (ASPP) commonly used in the field of deep learning. The specific structure of the module is shown in Figure 5 by performing a dilated convolution (also known as a dilated convolution or a hole convolution) operation on different scales, so as to capture the context information of different scales in the image, that is, to insert "holes" (i.e. zeros) between the standard convolution kernel elements. The module is shown in Figure 5 and contains the following parts: 1x1 convolution, 3x3 dilated convolution kernel with a dilated rate of 6, 3x3 dilated convolution kernel with a dilated rate of 12, 3x3 dilated convolution kernel with a dilated rate of 18, and a global average pooling unit.

[0079] Therefore, based on the pre-constructed classification network model, the time-series SAR image is pre-processed to suppress the coherent speckle noise of the radar image, and then the pre-processed time-series SAR image is registered and cropped with the preset label, the slice data set is input into the classification network model, and the classification of the time-series SAR image is realized.

[0080] Optionally, in some embodiments, before the slice data to be classified is input into the pre-constructed classification network model, further comprising: acquiring a plurality of target time-series SAR images; pre-processing the plurality of target time-series SAR images, and registering the pre-processed plurality of target time-series SAR images and the preset label data to obtain a slice data set; dividing the slice data set into a training set and a test set, and training an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module using the training set to obtain an initial classification network model; testing the initial classification network model using the test set, and when the test result meets the preset classification standard, using the initial classification network model as the pre-constructed classification network model.

[0081] Wherein, the preset classification standard can be determined by those skilled in the art according to actual needs, and the present application does not make any limitation.

[0082] Specifically, the embodiment of the present application can acquire a plurality of target time-series SAR images for testing and training, and sequentially pre-process the acquired plurality of target time-series SAR images through orbit correction, radiation calibration, geometric correction, coherent speckle filtering and registration, etc., and then register the pre-processed plurality of target time-series SAR images and the preset label data, and use the same size to crop to obtain a slice data set, and divide the slice data set into a training set and a test set, wherein the data size of the test set should be no less than 50% of the entire slice data. The initial network model constructed by fusing the multi-scale time-varying information module and the multi-scale spatial information module is trained by the data in the training set, and the initial classification network model is obtained after training, so that the network model can classify the time-series SAR image more accurately; and the trained initial classification network model is tested using the data in the test set, and it is judged whether the test result meets the preset classification standard, when the test result meets the preset classification standard, the initial classification network model is used as the preset classification network model.

[0083] Wherein, the trained initial classification network model can be tested using five-fold cross-validation, the data set is randomly divided into five equal-sized subsets, one subset is used as the validation set and the remaining four are combined as the training set, and the five rounds of testing are performed to obtain the final test result, and the better training result is selected for model testing.

[0084] Further, in some embodiments, after testing the initial classification network model with the test set, further comprising: if the test result does not meet the preset classification standard, adjusting the division ratio of the training set and the test set, and retraining the initial network model according to the newly divided test set until the new test result meets the preset classification standard, obtaining the pre-constructed classification network model.

[0085] Specifically, when the test result does not meet the preset classification standard, retesting can be performed by adjusting the division ratio of the training set and the test set until the test result meets the preset classification standard. In this process, in order to ensure the reliability of the test result, it is still necessary to ensure that the size of the test set data is not less than 50% of the entire slice data.

[0086] According to the SAR crop classification method based on the multi-scale time-varying information extraction network provided by the embodiment of the present application, the classification network model is trained and verified in advance to ensure the accuracy of the classification network model result, and then the classification slice data obtained by processing the SAR image is input into the classification network model. After the time sequence information extraction and noise suppression processing of the classification network model, the final time sequence SAR image crop classification result is obtained. Therefore, by constructing a multi-scale time-varying information network model to classify the information in the time sequence SAR image, the automatic and intelligent efficient and accurate classification of crops is realized under the condition of low parameter quantity, and the all-weather crop planting area monitoring is realized.

[0087] Secondly, the SAR crop classification device based on the multi-scale time-varying information extraction network provided by the embodiment of the present application is described with reference to the accompanying drawings.

[0088] Figure 6 is a block schematic diagram of the SAR crop classification device based on the multi-scale time-varying information extraction network according to an embodiment of the present application.

[0089] As shown in Figure 6 , the SAR crop classification device based on the multi-scale time-varying information extraction network 10 comprises an acquisition module 100, a preprocessing module 200 and a classification module 300.

[0090] The acquisition module 100 is configured to acquire the time sequence SAR image to be classified.

[0091] The preprocessing module 200 is configured to pre-process the time sequence SAR image to be classified, and register according to the pre-processing result and the preset label data to obtain the classification slice data.

[0092] The classification module 300 is configured to input the to-be-classified slice data into a pre-constructed classification network model to obtain a crop classification result of the to-be-classified time-series SAR image, wherein the pre-constructed classification network model is trained from an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module.

[0093] According to an embodiment of the present application, before the to-be-classified slice data is input into the pre-constructed classification network model, the classification module 300 is further configured to: acquire a plurality of target time-series SAR images; pre-process the plurality of target time-series SAR images, and register the pre-processed plurality of target time-series SAR images and preset label data to obtain a slice data set; divide the slice data set into a training set and a test set, and train an initial network model constructed by fusing the multi-scale time-varying information module and the multi-scale spatial information module using the training set to obtain an initial classification network model; test the initial classification network model using the test set, and when the test result meets a preset classification standard, use the initial classification network model as the pre-constructed classification network model.

[0094] According to an embodiment of the present application, after the initial classification network model is tested using the test set, the classification module 300 is further configured to: if the test result does not meet the preset classification standard, adjust the division ratio of the training set and the test set, and retrain the initial network model according to the newly divided test set until the new test result meets the preset classification standard to obtain the pre-constructed classification network model.

[0095] According to an embodiment of the present application, the to-be-classified time-series SAR image is pre-processed by the pre-processing module 200, which is specifically configured to: perform orbit correction processing on the to-be-classified time-series SAR image to obtain a first processed image; perform radiation calibration processing on the first processed image to obtain a second processed image; perform geometric correction processing on the second processed image to obtain a third processed image; perform coherent speckle filtering processing on the third processed image to obtain a fourth processed image; and perform registration on the fourth processed image to obtain a pre-processing result.

[0096] According to an embodiment of the present application, the multi-scale time-varying information module is:

[0097]

[0098] wherein v is a scattering difference of different time-phase radar images, T n is a time phase at t n .

[0099] According to the SAR crop classification device based on the multi-scale time-varying information extraction network provided by the embodiment of the application, the classification network model is trained and verified in advance to ensure the accuracy of the classification network model result, and then the classification slice data obtained by processing the SAR image is input into the classification network model, and after the time sequence information extraction and noise suppression processing of the classification network model, the final time sequence SAR image crop classification result is obtained. Therefore, by constructing the multi-scale time-varying information network model, the information in the time sequence SAR image is classified, the efficient and accurate classification of the crop automation and intelligence is realized under the condition of low parameter quantity, and the all-weather crop planting area monitoring is realized.

[0100] Figure 7 The structure schematic diagram of the electronic device provided by the embodiment of the application is provided. The electronic device can include:

[0101] The memory 701, the processor 702 and the computer program stored in the memory 701 and executable on the processor 702.

[0102] The processor 702 implements the SAR crop classification method based on the multi-scale time-varying information extraction network provided in the above embodiment when executing the program.

[0103] Further, the electronic device further includes:

[0104] The communication interface 703 is used for communication between the memory 701 and the processor 702.

[0105] The memory 701 is used for storing the computer program executable on the processor 702.

[0106] The memory 701 can include a high-speed RAM (Random Access Memory, Random Access Memory) memory, and can also include a non-volatile memory, for example, at least one disk memory.

[0107] If the memory 701, the processor 702 and the communication interface 703 are independently implemented, the communication interface 703, the memory 701 and the processor 702 can be connected to each other through a bus and complete the communication between each other. The bus can be an ISA (Industry Standard Architecture, Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect, Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture, Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 7Only one bus or only one type of bus can exist, however.

[0108] Optionally, in a specific implementation, if the memory 701, the processor 702 and the communication interface 703 are integrated on a chip, the memory 701, the processor 702 and the communication interface 703 can complete mutual communication through an internal interface.

[0109] The processor 702 can be a CPU (Central Processing Unit, central processing unit) or an ASIC (Application Specific Integrated Circuit, application specific integrated circuit) or one or more integrated circuits configured to implement embodiments of the present application.

[0110] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the SAR crop classification method based on the multi-scale time-varying information extraction network as described above.

[0111] The embodiment of the present application also provides a computer program product, which includes a computer program, and the program is executed by a processor to realize the SAR crop classification method based on the multi-scale time-varying information extraction network as described above.

[0112] In addition, the terms "first", "second", "third" and the like are used only to describe different instances, and do not imply or suggest relative importance or imply the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0113] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms is not necessarily for the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0114] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and that variations, modifications, substitutions and changes can be made by those skilled in the art without departing from the scope of the present application.

Claims

1. A SAR crop classification method based on a multi-scale time-varying information extraction network, characterized in that, The method comprises the following steps: acquire a time-series SAR image to be classified; pre-process the time-series SAR image to be classified, and register the pre-processing result and preset label data to obtain slice data to be classified; input the slice data to be classified into a pre-constructed classification network model to obtain a crop classification result of the time-series SAR image to be classified, wherein the pre-constructed classification network model is obtained by training an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module; wherein, before the slice data to be classified is input into the pre-constructed classification network model, the method further comprises: acquiring a plurality of target time-series SAR images; pre-processing the plurality of target time-series SAR images, and registering the pre-processed plurality of target time-series SAR images and the preset label data to obtain a slice data set; dividing the slice data set into a training set and a test set, and training an initial network model constructed by fusing the multi-scale time-varying information module and the multi-scale spatial information module using the training set to obtain an initial classification network model; testing the initial classification network model using the test set, and taking the initial classification network model as the pre-constructed classification network model when the test result meets a preset classification standard; the multi-scale time-varying information module is: ； wherein, is the difference in scattering for radar imagery at different times, is the time of day.

2. The method of claim 1, wherein, after the test set is used to test the initial classification network model, the method further comprises: if the test result does not meet the preset classification standard, adjusting the division ratio of the training set and the test set, and retraining the initial network model according to the newly divided test set until the new test result meets the preset classification standard to obtain the pre-constructed classification network model.

3. The method of claim 1, wherein, the pre-processing of the time-series SAR image to be classified comprises: orbit correction processing is performed on the time-series SAR image to be classified to obtain a first processed image; radiometric calibration processing is performed on the first processed image to obtain a second processed image; geometric correction processing is performed on the second processed image to obtain a third processed image; coherent speckle filtering processing is performed on the third processed image to obtain a fourth processed image; the fourth processed image is registered to obtain the pre-processing result.

4. A SAR crop classification device based on a multi-scale time-varying information extraction network, characterized in that, comprise: an acquisition module configured to acquire a time-series SAR image to be classified; a pre-processing module configured to pre-process the time-series SAR image to be classified, and register a pre-processing result and preset label data to obtain slice data to be classified; a classification module configured to input the slice data to be classified into a pre-constructed classification network model to obtain a crop classification result of the time-series SAR image to be classified, wherein the pre-constructed classification network model is obtained by training an initial network model constructed by fusing a multi-scale time-varying information module and a multi-scale spatial information module; The classification module is further configured to: acquire a plurality of target time-series SAR images; pre-process the plurality of target time-series SAR images, and register the pre-processed plurality of target time-series SAR images and the preset label data to obtain a slice data set; divide the slice data set into a training set and a test set, and train an initial network model constructed by fusing the multi-scale time-varying information module and the multi-scale spatial information module using the training set to obtain an initial classification network model; and test the initial classification network model using the test set, and use the initial classification network model as the pre-constructed classification network model when the test result meets a preset classification standard. The multi-scale time-varying information module is: ； wherein, is the scattering difference for different time phase radar imagery, is the time phase of the moment.

5. An electronic device, comprising: The method comprises: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the SAR crop classification method based on the multi-scale time-varying information extraction network according to any one of claims 1-3.

6. A computer storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the SAR crop classification method based on the multi-scale time-varying information extraction network according to any one of claims 1-3.

7. A computer program product, characterised in that, The computer program is executed by the processor to implement the SAR crop classification method based on the multi-scale time-varying information extraction network according to any one of claims 1-3.

Images