Method, device and medium for crop classification at plot level based on time-series SAR images

Abstract

The present invention discloses a method, device and medium for classifying crops at the plot level based on time-series SAR images, relating to the field of remote sensing technology. The method includes: obtaining plot data; constructing a multi-level knowledge reasoning framework, using the attribute information of the plot and external knowledge to speculate on the crop types, and obtaining the reasoning confidence corresponding to the crop types planted in each plot; based on the vector data of the boundaries of the plots, extracting the time-series features of each plot from the time-series SAR images, associating the time-series features of each plot with the crop types, and establishing a standard crop time-series feature library; comparing the plot to be identified with the standard crop time-series feature library, calculating the observation confidence, fusing the observation confidence and the reasoning confidence to obtain a fused confidence, and using the fused confidence to identify the crop type of the plot to be identified. The method of the present invention solves the problem of low recognition accuracy of traditional methods in small-sample and complex scenarios.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology, and more specifically, to a method, device, and medium for classifying crops at the plot level based on time-series SAR images. Background Art

[0002] At present, precision agriculture and smart agriculture have become important directions for the transformation of modern agriculture, gradually promoting the development of agriculture towards intelligence and intensification. Among them, crop species identification is an important application in precision agriculture, which plays an important role in realizing scientific agricultural management and ensuring fair agricultural subsidies.

[0003] Traditional crop identification requires regular field inspections by professional personnel. This method has obvious limitations: First, this method requires a large amount of human and material resources, with high costs; second, the inspection results are lagging and cannot be effectively monitored in real time; third, limited by manpower, the inspection scope is often relatively limited, making it difficult to achieve large-scale crop classification. In recent years, with the development of remote sensing technology, the resolution of remote sensing images has been continuously improved, and at the same time, the acquisition cost of images has been continuously reduced, which has enabled identification methods based on remote sensing images to gradually replace traditional manual inspections.

[0004] Early crop identification technologies based on remote sensing images mainly used threshold segmentation methods, that is, by artificially setting thresholds for features to distinguish different crop species. Such methods have the following deficiencies: (1) highly dependent on expert experience, with strong subjectivity; (2) poor transferability, with limited applicability between different regions and different crop types; (3) insufficient adaptability to complex planting environments, making it difficult to handle complex situations. In recent years, with the development of artificial intelligence, deep learning technology has been widely applied to crop identification tasks, and many related technologies use algorithms to create crop species maps at the pixel level for remote sensing images.

[0005] However, crop identification methods based on deep learning still face the following challenges: First, crop growth is a dynamic process, and different crops have different phenological characteristics. Multiple temporal remote sensing images need to be collected for analysis. However, existing multi-temporal remote sensing images often have a low spatial resolution (usually 10 - 30 meters), and it is easy to have unclear boundary pixel attribution problems when creating crop species maps at the pixel level, affecting the accuracy of subsequent downstream tasks; second, existing methods are prone to unreasonable results of identifying multiple crops in one plot when dealing with complex scenarios; third, deep learning methods highly depend on a large number of high-quality training samples, yet existing samples have problems of scarcity and uneven distribution.

[0006] In addition, existing technologies also have problems such as poor applicability to cloudy and rainy areas and low identification accuracy for small plot plantings. Summary of the Invention

[0007] To solve the above technical problems, the present invention provides a method, device and medium for classifying crops at the plot level based on time-series SAR images, aiming to solve the problem of low recognition accuracy of traditional methods in small-sample and complex scenarios based on multi-source data and knowledge-driven strategies.

[0008] In a first aspect, the present invention provides a method for classifying crops at the plot level based on time-series SAR images, the method comprising:

[0009] Obtain plot data, where the plot data includes vector data of the boundaries of the plot and attribute information of the plot;

[0010] Construct a multi-level knowledge reasoning framework, and use the attribute information of the plot and external knowledge to speculate on the crop types, obtaining the reasoning confidence corresponding to the crop types planted in each plot;

[0011] Obtain time-series SAR images, and based on the vector data of the boundaries of the plot, extract the time-series features of each plot from the time-series SAR images, associate the time-series features of each plot with the crop types, and establish a standard crop time-series feature library;

[0012] Compare the plot to be recognized with the standard crop time-series feature library, calculate the observation confidence, calculate the reasoning confidence of the plot to be recognized based on the multi-level knowledge reasoning framework, fuse the observation confidence and the reasoning confidence to obtain the fusion confidence, and use the fusion confidence to identify the crop type of the plot to be recognized.

[0013] Further, constructing a multi-level knowledge reasoning framework, and using the attribute information of the plot and external knowledge to speculate on the crop types, obtaining the reasoning confidence corresponding to the crop types planted in each plot, includes:

[0014] Integrate multi-source data and establish a crop phenological information library; wherein, the multi-source data includes the geographical information and spatial information of the plot, the geographical information includes land use type, plot altitude and / or slope, and the spatial information includes administrative division and / or natural division; the crop phenological information library is used to record the key phenological moments of different crop growth, and generate a phenological calendar in combination with the climate data of the natural area;

[0015] Based on expert knowledge, construct a set of reasoning conditions;

[0016] Based on the input attribute information of the plot and external knowledge, speculate on the crop types according to the set of reasoning conditions, and output the reasoning confidence corresponding to the crop types planted in the plot.

[0017] Further, the process of speculating on the crop types by the set of reasoning conditions includes:

[0018] Determine the set of crops planted in the plot according to the land use type of the plot; wherein, the set of crops includes the crop types and probabilities of multiple crops;

[0019] Filter the crop types and probabilities of the multiple crops according to the altitude of the plot and the historical planting information of the administrative region to which it belongs, and obtain the target crop set.

[0020] Furthermore, obtain the time-series SAR images, and based on the vector data of the boundaries of the plots, extract the time-series features of each plot from the time-series SAR images, associate the time-series features of each plot with the crop types, and establish a standard crop time-series feature library, including:

[0021] Perform radiometric calibration, terrain correction, and speckle noise filtering on the time-series SAR images to obtain preprocessed data;

[0022] Based on the vector data of the boundaries of the plots, extract the image features of each plot in the preprocessed data, and perform normalization processing on the image features to obtain the time-series features of each plot;

[0023] According to the actual planting situation of the plots, ensure that the plots cover the target crop types and target planting patterns, construct time-series feature curves according to the time-series features of each plot automatically extracted from the time-series SAR images, and store the time-series feature curves in the standard library to obtain the standard crop time-series feature library.

[0024] Furthermore, perform radiometric calibration, terrain correction, and speckle noise filtering on the time-series SAR images to obtain preprocessed data, including:

[0025] The calculation formula for performing radiometric calibration on the time-series SAR images is:

[0026] ;

[0027] In the formula, is the backscattering coefficient, DN is the original digital value, and A and B are the calibration parameters of the sensor;

[0028] Perform speckle noise filtering through the following formula:

[0029] ;

[0030] In the formula, is the pixel value after filtering, is the local mean of the pixel, is the filtering coefficient, is the pixel value at the corresponding position.

[0031] Further, the image features are normalized by the following formula:

[0032] ;

[0033] In the formula, is the normalized image feature, is the original feature of the plot, is the mean of the plot image features, is the standard deviation. The normalized eigenvalue conforms to the standard normal distribution.

[0034] Further, the plot to be recognized is compared with the standard crop time series feature library, and the observation confidence is calculated, including:

[0035] Calculate the similarity between the plot to be recognized and the standard crop time series feature library. The calculation formula is:

[0036] ;

[0037] In the formula, is the similarity between the plot and the features in the standard library, and respectively represent the feature values of the crop time series features of the plot to be recognized and in the standard library at the th time phase, is the length of the feature sequence, and the similarity value range is , the closer to 1, the more similar the two values are.

[0038] Normalize the similarity by the following formula:

[0039] ;

[0040] In the formula, is a value in the similarity, is the normalized value, and respectively represent the minimum and maximum values in the similarity;

[0041] Use the normalized value as the observation confidence.

[0042] Further, fuse the observation confidence and the inference confidence to obtain the fusion confidence, and use the fusion confidence to identify the crop type of the plot to be recognized, including:

[0043] Fuse the observation confidence and the inference confidence by the following formula to obtain the fusion confidence:

[0044] ;

[0045] In the formula, is the fusion confidence is the inference confidence is the observation confidence is the inference weight is the observation weight;

[0046] A preset determination threshold. When the fusion confidence is greater than the determination threshold, the crop type corresponding to the fusion confidence is used as the crop type of the plot to be recognized.

[0047] In a second aspect, the present invention provides a plot-level crop classification device based on time-series SAR images, and the device includes:

[0048] A data acquisition unit configured to acquire plot data, where the plot data includes vector data of the boundary of the plot and attribute information of the plot;

[0049] A knowledge inference unit configured to construct a multi-level knowledge inference framework, use the attribute information of the plot and external knowledge to infer the crop type, and obtain the inference confidence corresponding to the crop type planted in each plot;

[0050] A feature library construction unit configured to acquire time-series SAR images, extract the time-series features of each plot from the time-series SAR images based on the vector data of the boundary of the plot, associate the time-series features of each plot with the crop type, and establish a standard crop time-series feature library;

[0051] A crop classification unit configured to compare the plot to be recognized with the standard crop time-series feature library, calculate the observation confidence, calculate the inference confidence of the plot to be recognized based on the multi-level knowledge inference framework, fuse the observation confidence and the inference confidence to obtain the fusion confidence, and use the fusion confidence to identify the crop type of the plot to be recognized.

[0052] In a third aspect, the present invention provides a readable storage medium storing one or more programs, and the one or more programs can be executed by one or more processors to implement the method as described above.

[0053] The present invention has at least the following beneficial effects:

[0054] 1. Improve the recognition accuracy. The present invention effectively solves the recognition error problem caused by fuzzy boundaries in traditional methods. Experiments show that the overall recognition accuracy of this method exceeds 90%, an increase of more than 8% compared with traditional methods. Especially in small plot areas, the recognition accuracy improvement is particularly obvious, exceeding 15%.

[0055] 2. Reduce sample dependence. By introducing knowledge reasoning, the present invention significantly reduces the dependence on a large amount of labeled data. In practical applications, test results show that when the sample size is reduced by 50%, the recognition accuracy can still remain above 85%. This characteristic makes the present method more applicable in fields with scarce samples.

[0056] 3. Enhance the interpretability of speculation results. The present invention constructs an interpretable decision framework. Through the fusion of knowledge reasoning and observation results, the crop recognition results of the present invention have clear logical bases. This characteristic not only improves the credibility of the results but also provides decision-making support for the agricultural department, contributing to the implementation of precision agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 shows the overall flowchart of the plot-level crop classification method based on time-series SAR images according to an embodiment of the present invention;

[0058] Figure 2 shows the regional and elevation zoning map of the target area according to an embodiment of the present invention;

[0059] Figure 3 shows a partial example diagram of the fine plots in the target area according to an embodiment of the present invention;

[0060] Figure 4 shows a partial example diagram of the rice distribution recognition result in the target area according to an embodiment of the present invention;

[0061] Figure 5 shows the structure diagram of the plot-level crop classification device based on time-series SAR images according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples, but this is not a limitation to the present invention. For the steps described herein, if there is no necessity for a sequential relationship between them, the order in which they are described as examples herein should not be regarded as a limitation. Those skilled in the art should know that they can be adjusted in order as long as the logic between them is not destroyed and the entire process cannot be implemented.

[0063] An embodiment of the present invention provides a plot-level crop classification method based on time-series SAR images. As Figure 1 shown, it is the flowchart of the plot-level crop classification method based on time-series SAR images. The plot-level crop classification method based on time-series SAR images includes the following steps S10 to S40.

[0064] S10: Obtain plot data, where the plot data includes vector data of the boundaries of the plot and attribute information of the plot.

[0065] In this embodiment, existing plot boundary data is utilized, combined with high-resolution images (spatial resolution better than 1 meter), to optimize and verify the plot boundaries, so as to ensure the accuracy of the plot boundaries and provide accurate spatial constraints for subsequent plot crop identification.

[0066] S20: Construct a multi-level knowledge inference framework, and use the attribute information of the plot and external knowledge to speculate on the crop types, and obtain the inference confidence corresponding to the crop types planted in each plot.

[0067] In this embodiment, the purpose of step S20 is the preliminary speculation of crop types based on knowledge inference. By constructing a multi-level knowledge inference framework, the attribute information of the plot and external knowledge are comprehensively utilized to speculate on the crop types. Through multi-dimensional knowledge fusion, the inference confidence corresponding to the crops planted in each plot is obtained, providing prior knowledge for subsequent identification.

[0068] In some embodiments, step S20 is implemented by the following steps S21 to S23.

[0069] S21: Integrate multi-source data and establish a crop phenological information database; wherein, the multi-source data includes the geographical information and spatial information of the plot, the geographical information includes land use type, plot altitude and / or slope, and the spatial information includes administrative division and / or natural division; the crop phenological information database is used to record the key phenological moments of different crop growth, and combine the climate data of the natural area to generate a phenological calendar.

[0070] Specifically, in step S21, first, integrate multi-source data, including geographical information such as land use type, plot altitude, slope, etc., and spatial information such as administrative division and natural division; for example, the land use type includes paddy fields, dry land, orchards, etc., and the altitude is stratified at intervals of 100 meters. Secondly, establish a crop phenological information database, record the key phenological moments of different crop growth, such as sowing period, heading period, maturity period, etc., and combine the specific climate data of different natural areas to generate an accurate and personalized phenological calendar. The phenological calendar is used to judge the crops planted on the plot each month.

[0071] S22: Based on expert knowledge, construct an inference condition set.

[0072] In some embodiments, the process of the inference condition set for crop type speculation includes: determining the set of crops planted in the plot according to the land use type of the plot; wherein the crop set includes the crop types and probabilities of multiple crops; screening the crop types and probabilities of the multiple crops according to the altitude of the plot and the historical planting information of the administrative region to which it belongs to obtain the target crop set.

[0073] Exemplarily, for the inference condition set, if the land type of the plot is an orchard, the possibility of planting aquatic crops such as rice is excluded; if the altitude of the plot is higher than 1000 meters, the possibility of planting rice is reduced, such as less than 10%; if the plot is located within a certain administrative region and the administrative region has historically mainly planted wheat, the plot type is preferably inferred as wheat, such as 80%.

[0074] S23: Based on the input attribute information of the plot and external knowledge, perform crop type speculation according to the inference condition set, and output the inference confidence corresponding to the crop type planted in the plot.

[0075] The purpose of step S23 is to achieve a preliminary speculation of the crop type. Only as an example, the plot attributes and external knowledge are input into the knowledge inference model to obtain the confidence of the crop type for each plot. At the same time, a confidence threshold is set. If the confidence of the crop is higher than a certain threshold, it is used as a candidate crop. If it is lower than a certain threshold, its possibility is excluded. Otherwise, further analysis and judgment need to be combined with temporal features.

[0076] S30: Obtain the temporal SAR images, extract the temporal features of each plot from the temporal SAR images based on the vector data of the boundaries of the plots, and associate the temporal features of each plot with the crop types to establish a standard crop temporal feature library.

[0077] In this embodiment, the purpose of step S30 is to establish a standard crop temporal feature library through feature extraction of temporal SAR images. Using a multi-temporal SAR image sequence, perform regional statistical analysis based on the precise plot boundaries.

[0078] In some embodiments, step S30 can be implemented by the following steps S31 to S33.

[0079] S31: Perform radiometric calibration, terrain correction, and speckle noise filtering on the temporal SAR images to obtain preprocessed data.

[0080] In some embodiments, the calculation formula for radiometric calibration of temporal SAR images is:

[0081] ;

[0082] In the formula, is the backscattering coefficient,DN is the original numerical value, and A and B are the calibration parameters of the sensor;

[0083] Speckle noise filtering is performed through the following formula:

[0084] ;

[0085] In the formula, is the filtered pixel value, is the local mean of the pixel, is the filtering coefficient, is the pixel value at the corresponding position.

[0086] S32: Based on the vector data of the boundaries of the plots, extract the image features of each plot in the preprocessed data, and perform normalization processing on the image features to obtain the time series features of each plot.

[0087] In some embodiments, based on the plot boundaries, extract the statistical value features of each plot, and perform normalization processing on the time series features to eliminate the influence of dimensions. The formula is as follows:

[0088] ;

[0089] In the formula, is the normalized image feature, is the original feature of the plot, is the mean of the plot image features, is the standard deviation. The normalized eigenvalue conforms to the standard normal distribution.

[0090] S33: According to the actual planting situation of the plots, ensure that the plots cover the target crop types and target planting patterns. Based on the time series features of each plot automatically extracted from the time series SAR images, construct a time series feature curve, and store the time series feature curve in the standard library to obtain a standard crop time series feature library.

[0091] In the specific implementation of step S33, the actual planting situation of the plot samples can be obtained through on-site investigation to ensure that the samples cover the main crop types and typical planting patterns. Through a multi-temporal SAR image sequence with high temporal resolution, automatically extract the statistical value feature curve of the plot samples and store it in the standard library, so as to obtain a standard crop time series feature library.

[0092] S40: Compare the plot to be identified with the standard crop time series feature library, calculate the observation confidence, calculate the inference confidence of the plot to be identified based on the multi-level knowledge inference framework, fuse the observation confidence and the inference confidence to obtain a fused confidence, and use the fused confidence to identify the crop type of the plot to be identified.

[0093] In this embodiment, the purpose of step S40 is to organically combine the similarity of temporal features and the inference confidence to achieve crop recognition with multi-source information fusion.

[0094] In some embodiments, step S40 can be specifically implemented by the following steps S41 to S43.

[0095] S41: Temporal feature matching.

[0096] Calculate the similarity between the sequence features of the plot to be recognized and the standard crop temporal feature library, and generate the observation confidence according to their cosine similarity. The formula is as follows:

[0097] ;

[0098] In the formula, is the similarity between the plot and the features in the standard library, and respectively represent the feature values of the crop temporal features of the plot to be recognized and the standard library at the th time phase, is the length of the feature sequence, and the similarity value range is , the closer to 1, the more similar the two values are;

[0099] Set the conversion relationship between similarity and confidence. To increase the distinguishability between plots, the following formula is used to normalize the similarity and map it to the [0,1] interval:

[0100] ;

[0101] In the formula, is a value in the similarity, is the value after normalization, and respectively represent the minimum and maximum values in the similarity.

[0102] Use the value after normalization as the observation confidence.

[0103] S42: Confidence fusion.

[0104] Design a fusion method to fuse the inference confidence based on knowledge reasoning and the observation confidence based on image observation according to the weights. The formula is as follows:

[0105] ;

[0106] In the formula, is the fusion confidence, is the inference confidence, is the observation confidence, is the inference weight, is the observation weight.

[0107] For example, when both the inference weight and the observation weight are set to 0.5, assume that the confidence level of planting wheat in a certain plot based on knowledge inference is 85%, and the observation confidence level is 95%. After fusion, the final confidence level is 90%.

[0108] S43: Preset a determination threshold. When the fused confidence level is greater than the determination threshold, the crop type corresponding to the fused confidence level is used as the crop type of the plot to be recognized.

[0109] Exemplarily, set a determination threshold, such as 80%. If the fused confidence level of a certain crop is higher than the threshold, it is determined as that crop type. Finally, the classification results at the plot level are output, including the plot crop type, confidence level, and spatial distribution.

[0110] Therefore, the method provided by the embodiments of the present invention makes the high-resolution image and the SAR image complement each other's advantages, achieving the balance between spatial accuracy and temporal resolution. In addition, by introducing a knowledge inference mechanism, while alleviating the problem of difficult model training under small sample conditions, it fills the deficiencies of data-driven methods, making the inference results more interpretable and enhancing the credibility and practicality of the results. Through the above technical solutions, the present invention successfully solves the problems existing in traditional methods, such as fuzzy boundaries, insufficient samples, and lack of interpretability of recognition results, providing reliable technical support for precision agriculture.

[0111] Next, specific experiments and analyses will be combined to further illustrate the feasibility and progressiveness of the present invention.

[0112] Based on the plot-level crop classification method based on temporal SAR images proposed in each of the above embodiments, it is applied to a certain target area. The target area spans 105°11' - 110°11' east longitude and 28°10' - 32°13' north latitude, with a width of 470 kilometers from east to west and a length of 450 kilometers from north to south. It is located in the transitional zone between the Qinghai-Tibet Plateau and the Yangtze River Plain. The overall trend is that the east is higher than the west, and the south is higher than the north. The altitude ranges from 73.1 to 2723.7 meters, and most of the areas have an altitude within 800m suitable for crop growth. The target area belongs to the tropical monsoon humid climate, with an annual average temperature of 16 - 18°C, and the precipitation in each area ranges from 940 to 1375mm. The landform is mainly mountainous and hilly, with mountains accounting for 76%. The main soil types are purple soil, yellow soil, paddy soil, and limestone (rock) soil. The main food crops are rice, corn, wheat, sorghum, sweet potato, and potato. The main cash crops are rapeseed, tobacco leaves, tea leaves, and Chinese prickly ash. The main fruit crops are citrus and pomelo.

[0113] Data sources: Multi-spectral high-resolution remote sensing image data, time-series SAR data, and vector data. The multi-spectral high-resolution remote sensing image data is used for the extraction of fine plot boundaries and result verification. The time-series SAR data consists of 12 scenes of IW-mode GRD level-1 product data of Sentinel-1 satellite from January to August 2022, which can ensure multi-period coverage at least once a month. The auxiliary vector data includes phenological zoning, natural zoning, administrative zoning, and previously obtained fine agricultural plots. The plot attributes include statistical information such as plot type, reference land use type, altitude, slope, and aspect.

[0114] By using the division boundaries of the four blocks in the west, central, south, and northeast regions and the crop information in elevation intervals, corresponding to the sowing and growth times of different crop types in each region and elevation interval, the complete crop production zoning data for the target area is obtained.

[0115] Taking the west block as an example, the main crop phenological calendar for the target area with an altitude of 0 - 400 meters is shown in Table 1.

[0116] Table 1 Main crop phenological calendar for the target area with an altitude of 0 - 400 meters in the west region

[0117] Summer crops Earliest sowing date Latest sowing date Crop maturity period Rice Early April Early May September Maize Late March Early April August Potato Mid-December of the previous year Mid-January July Sweet potato Early May Mid-May November Sorghum Early April Mid-April September Winter crops Earliest sowing date Latest sowing date Maturity period Rapeseed Early October Mid-October May of the following year Potato Late August Mid-September January of the following year Wheat Late October Mid-November May of the following year

[0118] Based on the method proposed in this embodiment, crop recognition is performed in multiple blocks of the target area, a crop phenological information database and a knowledge inference model for the study area are established, and then time-series feature matching is carried out in combination with the multi-temporal SAR image sequence of the study area to infer the confidence level of the crop types by knowledge inference. The crop recognition results applied to the target area are as Figure 3 and Figure 4 shown. Figure 4 The confidence level in is the fused confidence level. The method proposed in the present invention has good effects on small-scale fine plots.

[0119] 1347 samples of crop collection points in the plots of the same period are obtained, and accuracy evaluation is carried out based on the comparison and analysis of the verification samples and the speculation results. Among them, the calculation formula for the accuracy rate is:

[0120] ;

[0121] From the crop speculation accuracy table, it can be seen that accurate classification and recognition of crops can be achieved through the plot-level crop classification method based on time-series SAR images. The accuracy is shown in Table 2. The overall classification accuracy of the method in this chapter reaches 90.66%, and the overall classification effect is relatively ideal. The classification accuracies of wheat, rice, and citrus are all above 90%, which can meet the detailed investigation requirements for the crops in the plots.

[0122] Table 2 Crop speculation accuracy table

[0123]

[0124] It should be noted that Regions 1 to 5 mentioned in Table 2 are five administrative regions within the target area.

[0125] The embodiment of the present invention also provides a device for classifying crops at the plot level based on time-series SAR images, as Figure 5 shown. The device includes:

[0126] A data acquisition unit 501, configured to acquire plot data, where the plot data includes vector data of the boundary of the plot and attribute information of the plot;

[0127] A knowledge reasoning unit 502, configured to construct a multi-level knowledge reasoning framework, use the attribute information of the plot and external knowledge to speculate on the crop type, and obtain the reasoning confidence corresponding to the crop type planted in each plot;

[0128] A feature library construction unit 503, configured to acquire time-series SAR images, extract the time-series features of each plot from the time-series SAR images based on the vector data of the boundary of the plot, associate the time-series features of each plot with the crop type, and establish a standard crop time-series feature library;

[0129] A crop classification unit 504, configured to compare the plot to be identified with the standard crop time-series feature library, calculate the observation confidence, calculate the reasoning confidence of the plot to be identified based on the multi-level knowledge reasoning framework, fuse the observation confidence and the reasoning confidence to obtain a fusion confidence, and use the fusion confidence to identify the crop type of the plot to be identified.

[0130] In some embodiments, the knowledge reasoning unit is further configured to:

[0131] Integrate multi-source data and establish a crop phenological information library; wherein, the multi-source data includes the geographical information and spatial information of the plot, the geographical information includes land use type, plot altitude and / or slope, and the spatial information includes administrative division and / or natural division; the crop phenological information library is used to record the key phenological moments of different crop growth, and generate a phenological calendar in combination with the climate data of the natural area;

[0132] Construct a set of reasoning conditions based on expert knowledge;

[0133] Based on the input attribute information of the plot and external knowledge, speculate on the crop type according to the set of reasoning conditions, and output the reasoning confidence corresponding to the crop type planted in the plot.

[0134] In some embodiments, the process of speculating on the crop type by the set of reasoning conditions includes:

[0135] Determine the set of crops planted in the plot according to the land use type of the plot; wherein, the set of crops includes the crop types and probabilities of multiple crops.

[0136] Screen the crop types and probabilities of the multiple crops according to the altitude of the plot and the historical planting information of the administrative region to which it belongs, and obtain the target crop set.

[0137] In some embodiments, the feature library construction unit is further configured to:

[0138] Perform radiometric calibration, terrain correction, and speckle noise filtering on the time-series SAR image to obtain preprocessed data.

[0139] Based on the vector data of the boundary of the plot, extract the image features of each plot in the preprocessed data, and perform normalization processing on the image features to obtain the time-series features of each plot.

[0140] According to the actual planting situation of the plot, ensure that the plot covers the target crop type and target planting mode, construct a time-series feature curve based on the time-series features of each plot automatically extracted from the time-series SAR image, and store the time-series feature curve in the standard library to obtain the standard crop time-series feature library.

[0141] In some embodiments, the feature library construction unit is further configured to:

[0142] The calculation formula for performing radiometric calibration on the time-series SAR image is:

[0143] ;

[0144] In the formula, is the backscattering coefficient, DN is the original digital value, and A and B are the calibration parameters of the sensor.

[0145] Perform speckle noise filtering through the following formula:

[0146] ;

[0147] In the formula, is the filtered pixel value, is the local mean of the pixel, is the filtering coefficient, is the pixel value at the corresponding position.

[0148] In some embodiments, the feature library construction unit is further configured to perform normalization processing on the image features through the following formula:

[0149] ;

[0150] In the formula, is the normalized image feature, is the original feature of the plot, is the mean of the plot image features, is the standard deviation. The normalized eigenvalue conforms to the standard normal distribution.

[0151] In some embodiments, the crop classification unit is further configured to:

[0152] Calculate the similarity between the plot to be recognized and the standard crop time series feature library. The calculation formula is:

[0153] ;

[0154] In the formula, is the similarity between the plot and the features in the standard library, and respectively represent the feature values of the crop time series features of the plot to be recognized and in the standard library at the th time phase, is the length of the feature sequence. The similarity value range is , the closer to 1, the more similar the two values represent;

[0155] Normalize the similarity through the following formula:

[0156] ;

[0157] In the formula, is a value in the similarity, is the value after normalization, and respectively represent the minimum and maximum values in the similarity;

[0158] Use the value after normalization as the observation confidence.

[0159] In some embodiments, the crop classification unit is further configured to:

[0160] Fuse the observation confidence and the inference confidence through the following formula to obtain the fusion confidence:

[0161] ;

[0162] In the formula, is the fusion confidence, is the inference confidence, is the observation confidence, is the inference weight, is the observation weight;

[0163] A preset determination threshold. When the fusion confidence is greater than the determination threshold, the crop type corresponding to the fusion confidence is used as the crop type of the plot to be recognized.

[0164] It should be noted that the structures of the various plot-level crop classification devices based on temporal SAR images described in this embodiment belong to the same technical concept as the previously described plot-level crop classification method based on temporal SAR images, and achieve the same beneficial effects through the same principle, which will not be elaborated here.

[0165] An embodiment of the present invention also provides a readable storage medium. The readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method described in any of the above embodiments.

[0166] The above description is intended to be illustrative and not restrictive. For example, the above examples (or one or more of their solutions) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. Additionally, in the above specific embodiments, various features can be grouped together to simplify the present invention. This should not be construed as an intention that the features of an unclaimed invention are necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of a particular embodiment of the invention. Thus, the following claims are incorporated herein as examples or embodiments into the specific embodiments, where each claim independently serves as a separate embodiment, and considering these embodiments, they can be combined with each other in various combinations or permutations. The scope of the present invention should be determined with reference to the appended claims and the full scope of the equivalent forms empowered by these claims.

Claims

1. A plot-level crop classification method based on time-series SAR images, characterized in that: The method comprises: Acquire land parcel data, wherein the land parcel data includes vector data of the boundary of the land parcel and attribute information of the land parcel; Construct a multi-level knowledge reasoning framework, use the attribute information of the plot and external knowledge to infer the crop type, and obtain the reasoning confidence corresponding to the crop type planted on each plot; Acquire time-series SAR images, extract time-series features of each plot from the time-series SAR images based on vector data of the boundaries of the plots, associate the time-series features of each plot with crop types, and establish a standard crop time-series feature library; Comparing the plot to be identified with the standard crop time series feature library, calculating the observation confidence, calculating the reasoning confidence of the plot to be identified based on the multi-level knowledge reasoning framework, fusing the observation confidence with the reasoning confidence to obtain a fusion confidence, and using the fusion confidence to identify the crop type of the plot to be identified; A multi-level knowledge reasoning framework is constructed to use the attribute information of the plot and external knowledge to infer the crop type, and obtain the reasoning confidence corresponding to the crop type planted on each plot, including: Integrate multi-source data and establish a crop phenology information database; wherein the multi-source data includes geographic information and spatial information of the plot, the geographic information includes land use type, plot altitude and / or slope, and the spatial information includes administrative divisions and / or natural divisions; the crop phenology information database is used to record key phenological moments of growth of different crops, and generate a phenological calendar in combination with climate data of natural areas; Based on expert knowledge, construct a set of reasoning conditions; Based on the inputted property information of the plot and external knowledge, the crop type is inferred according to the reasoning condition set, and the reasoning confidence corresponding to the crop type planted on the plot is outputted; The observation confidence and the reasoning confidence are integrated to obtain a fused confidence, and the crop type of the to-be-identified plot is identified by using the fused confidence, including: The observation confidence and the inference confidence are fused by the following formula to obtain the fusion confidence: ； In the formula, is the fusion confidence, is the inference confidence, is the observation confidence, is the inference weight, is the observation weight; A determination threshold is preset, and when the fusion confidence is greater than the determination threshold, the crop type corresponding to the fusion confidence is used as the crop type of the land parcel to be identified.

2. The plot-level crop classification method based on time-series SAR images according to claim 1 is characterized in that: The process of inferring the crop type according to the inference condition set includes: Determine a crop set planted on the plot according to the land use type of the plot; wherein the crop set includes crop types and probabilities of multiple crops; According to the altitude of the plot and the historical planting information of the administrative division to which it belongs, the crop types and probabilities of the multiple crops are screened to obtain a target crop set.

3. The plot-level crop classification method based on time-series SAR images according to claim 1 is characterized in that: Acquire time-series SAR images, extract the time-series features of each plot from the time-series SAR images based on the vector data of the boundaries of the plots, associate the time-series features of each plot with the crop type, and establish a standard crop time-series feature library, including: Perform radiometric calibration, terrain correction and speckle noise filtering on time-series SAR images to obtain preprocessed data; Based on the vector data of the boundary of the plot, the image features of each plot in the preprocessed data are extracted, and the image features are normalized to obtain the time series features of each plot; According to the actual planting situation of the plot, ensure that the plot covers the target crop type and target planting pattern, construct a time series characteristic curve based on the time series characteristics of each plot automatically extracted from the time series SAR image, and store the time series characteristic curve in the standard library to obtain a standard crop time series characteristic library.

4. The method for crop classification at the plot level based on time-series SAR images according to claim 3 is characterized in that: Perform radiometric calibration, terrain correction and speckle noise filtering on the time series SAR images to obtain preprocessed data, including: The calculation formula for radiometric calibration of time-series SAR images is: ； In the formula, is the backscattering coefficient, DN is the original digital value, A and B are the calibration parameters of the sensor; Speckle noise filtering is performed using the following formula: ； In the formula, is the pixel value after filtering, is the local mean of pixels, is the filter coefficient, is the pixel value at the corresponding position.

5. The method for crop classification at the plot level based on time-series SAR images according to claim 3, characterized in that: The image features are normalized by the following formula: ； In the formula, is the normalized image feature, The original characteristics of the plot. is the mean value of the plot image feature, is the standard deviation.

6. The method for crop classification at the plot level based on time-series SAR images according to claim 1, characterized in that: Compare the plot to be identified with the standard crop time series feature library and calculate the observation confidence, including: Calculate the similarity between the plot to be identified and the standard crop time series feature library. The calculation formula is: ； In the formula, is the similarity between the plot and the features in the standard library, and Represents the crop time series characteristics of the plot to be identified and the standard library in the first The characteristic value of the phase, is the length of the feature sequence, and the similarity range is , the closer the value is to 1, the more similar the two are; The similarity is normalized by the following formula: ； In the formula, is a value in similarity, is the normalized value, and Represent the minimum and maximum values ​​of similarity respectively; The normalized value is used as the observation confidence.

7. A plot-level crop classification device based on time-series SAR images, characterized in that: The device comprises: A data acquisition unit is configured to acquire land parcel data, wherein the land parcel data includes vector data of a boundary of the land parcel and attribute information of the land parcel; The knowledge reasoning unit is configured to construct a multi-level knowledge reasoning framework, use the attribute information of the plot and external knowledge to infer the crop type, and obtain the reasoning confidence corresponding to the crop type planted in each plot; The feature library construction unit is configured to obtain time-series SAR images, extract the time-series features of each plot from the time-series SAR images based on the vector data of the boundaries of the plots, associate the time-series features of each plot with the crop type, and establish a standard crop time-series feature library; The crop classification unit is configured to compare the plot to be identified with the standard crop time series feature library, calculate the observation confidence, calculate the reasoning confidence of the plot to be identified based on the multi-level knowledge reasoning framework, fuse the observation confidence with the reasoning confidence to obtain a fusion confidence, and use the fusion confidence to identify the crop type of the plot to be identified; The knowledge reasoning unit is further configured to: Integrate multi-source data and establish a crop phenology information database; wherein the multi-source data includes geographic information and spatial information of the plot, the geographic information includes land use type, plot altitude and / or slope, and the spatial information includes administrative divisions and / or natural divisions; the crop phenology information database is used to record key phenological moments of growth of different crops, and generate a phenological calendar in combination with climate data of natural areas; Based on expert knowledge, construct a set of reasoning conditions; Based on the inputted property information of the plot and external knowledge, the crop type is inferred according to the reasoning condition set, and the reasoning confidence corresponding to the crop type planted on the plot is outputted; The crop classification unit is further configured to: The observation confidence and the inference confidence are fused by the following formula to obtain the fusion confidence: ； In the formula, is the fusion confidence, is the inference confidence, is the observation confidence, is the inference weight, is the observation weight; A determination threshold is preset, and when the fusion confidence is greater than the determination threshold, the crop type corresponding to the fusion confidence is used as the crop type of the land parcel to be identified. 8 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, executes the method according to claim 1 .

Images