Crop fine classification method and device based on ten-meter analysis ready-to-use data product

Abstract

The invention provides a crop fine classification method and device based on a ten-meter-level analysis-and-play data product, and the method comprises the steps: extracting a plurality of types of crop data from a crop recognition region according to a priori knowledge base and a time sequence synthesis data set, and obtaining a first sample set; training a crop detection model by taking the first sample set as a training sample, and identifying all crop types in the crop identification area based on the crop detection model to obtain a crop identification result; segmenting the field parcels in the crop identification area based on a high-resolution remote sensing image segmentation technology to obtain a field parcel segmentation result; and determining a crop fine classification result of the crop identification area based on the crop identification result and the field segmentation result. According to the method, a crop fine classification framework of ten-meter-level analysis ready-to-use data is constructed and combined with a field identification technology, so that fine description of a planting structure is realized, and the crop fine classification accuracy based on remote sensing data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method and device for fine classification of crops based on ready-to-use data products at the ten-meter level analysis. Background Art

[0002] Detailed crop classification and mapping can provide information on the spatial distribution of crops. It is an important basis for agricultural resource surveys, yield assessments, disaster monitoring, digital agriculture, etc. It is of great significance for agricultural sustainable development decision-making and layout, such as improving agricultural production efficiency, achieving food security and ecological balance.

[0003] Remote sensing data has the characteristics of long-term, large-scale, and multi-temporal observation, and can be used to obtain and monitor information such as crop growth conditions and distribution range.

[0004] In the related technology, although the crop classification method based on time series can realize crop classification by utilizing crop phenological information, its spatial resolution is limited and it is only suitable for large-scale crop classification, and cannot reflect the crop distribution characteristics at the regional scale. The crop classification method based on the regional scale has a low temporal resolution and is difficult to form high-frequency observations. The available observation frequency in cloudy and rainy areas is even less, which may result in missing the key nodes of crop monitoring in the classification, thereby affecting the accuracy of crop classification. Moreover, the crop fine classification maps obtained by the above methods lack a detailed description of the crop growth environment, and cannot reflect the crop planting structure, resulting in low accuracy of crop fine classification. Summary of the invention

[0005] The present invention provides a crop fine classification method and device based on ten-meter-level analysis and ready-to-use data products, so as to solve the defect that the crop fine classification method based on remote sensing data in the prior art cannot reflect the planting structure of crops, resulting in low accuracy of crop fine classification, and improves the accuracy of crop fine classification based on remote sensing data.

[0006] The present invention provides a crop fine classification method based on a decameter-level analysis ready-to-use data product, comprising: Extracting multiple types of crop data from the crop identification area according to the prior knowledge base and the time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained by using the ARD product of the long time series of the crop identification area at the ten-meter level analysis; the prior knowledge base is constructed based on all the crop types, growth characteristics and phenological information in the crop identification area; A crop detection model is trained using the first sample set as training samples, and all crop types in the crop identification area are identified based on the crop detection model to obtain a crop identification result; and fields in the crop identification area are segmented based on high-resolution remote sensing image segmentation technology to obtain a field segmentation result; A crop fine classification result of the crop identification area is determined based on the crop identification result and the field segmentation result.

[0007] According to a crop fine classification method based on a decameter-level analysis ready-to-use data product provided by the present invention, the time series synthetic data set is obtained by the following steps: Collect the 10-meter ARD products for a long time series of M consecutive years in the crop identification area; According to the quality mark of the ARD product, pixels with poor preprocessing accuracy in the ARD product are eliminated to obtain a new 10-meter-level ARD product; The new ten-meter-level ARD product is subjected to at least one of cropping, splicing, vegetation index calculation and synthesis according to the crop identification area to obtain an N-day time series synthetic data set, wherein the N-day time series synthetic data set includes multispectral band images and synthetic images of vegetation indices such as NDVI; M and N are both positive integers.

[0008] According to a crop fine classification method based on a decameter-level analysis ready-to-use data product provided by the present invention, the crop identification area includes a plurality of sub-areas; the first sample set is extracted from one of the plurality of sub-areas; After obtaining the first sample set and before training a crop detection model using the first sample set as training samples, the method further includes: Performing time expansion on the first sample set to obtain a sample set based on time expansion; The sample set based on time expansion is spatially expanded to obtain a second sample set.

[0009] According to a crop fine classification method based on a decameter-level analysis ready-to-use data product provided by the present invention, the time expansion of the first sample set to obtain a sample set based on time expansion includes: Obtain an N-day time series synthetic data set corresponding to the decameter-level ARD product of the long time series of the crop identification area for consecutive years Mi to Mj; M, i and j are all positive integers, and i and j are less than or equal to 5; For each sub-area, a crop hierarchical classification decision tree is constructed for each sub-area based on the N-day time series synthetic data set and the N-day time series synthetic data set corresponding to the ten-meter-level ARD product of the continuous M-year long time series, and multiple types of crop data are extracted from the crop hierarchical classification decision tree to obtain a sample set based on time expansion.

[0010] According to a crop fine classification method based on a decameter-level analysis ready-to-use data product provided by the present invention, the spatial expansion of the sample set based on time expansion to obtain a second sample set includes: Spatially expanding each sub-area according to a preset spatial buffer to obtain a plurality of expanded sub-areas; For each expanded sub-area, crop distinguishing features are extracted from the time-expanded sample set based on feature importance screening technology, and different types of crop data in the time-expanded sample set are supervised and labeled based on the crop distinguishing features to obtain a second sample set.

[0011] According to a crop fine classification method based on a decameter-level analysis ready-to-use data product provided by the present invention, after obtaining the second sample set, the method further includes: Noise samples and sample edge pixels are removed from the second sample set to obtain a third sample set.

[0012] The present invention also provides a crop fine classification device based on a decameter-level analysis ready-to-use data product, comprising: A sample extraction module is used to extract multiple types of crop data from the crop identification area according to a priori knowledge base and a time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through a ten-meter-level analysis, i.e., data ARD product, of a long time series of the crop identification area; and the priori knowledge base is constructed based on all crop types, growth characteristics, and phenological information in the crop identification area; an identification and segmentation module, configured to train a crop detection model using the first sample set as training samples, and identify all crop types in the crop identification area based on the crop detection model to obtain a crop identification result; and segment the fields in the crop identification area based on high-resolution remote sensing image segmentation technology to obtain a field segmentation result; The crop fine classification module is used to determine the crop fine classification result of the crop identification area based on the crop identification result and the field segmentation result.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements any of the above-described methods for fine crop classification based on decameter-level analysis and ready-to-use data products.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for fine crop classification based on decameter-level analysis and ready-to-use data products as described in any one of the above.

[0015] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned methods for fine crop classification based on decameter-level analysis and ready-to-use data products.

[0016] The present invention provides a crop fine classification method and device based on a data product that can be used at the 10-meter level. The method and device extract multiple types of crop data from a crop identification area through a priori knowledge base and a time series synthetic data set to obtain a first sample set and train a crop detection model based on the first sample set, thereby realizing the identification of all crop types in the crop identification area, and then segmenting the fields in the crop identification area to obtain field segmentation results. Finally, the crop identification results and the field segmentation results are superimposed to determine the crop fine classification results. By constructing a crop fine classification framework for the data product that can be used at the 10-meter level and combining it with field identification technology, the planting structure can be finely described, thereby improving the accuracy of crop fine classification based on remote sensing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is one of the flow charts of the crop fine classification method based on the ready-to-use data product of ten-meter-level analysis provided by the present invention.

[0019] Figure 2 This is the second flow chart of the crop fine classification method based on the decameter-level analysis ready-to-use data product provided by the present invention.

[0020] Figure 3 It is a structural schematic diagram of a crop fine classification device based on a decameter-level analysis ready-to-use data product provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Combine the following Figure 1-Figure 3 Describe the invented method and device for fine crop classification based on ready-to-use data products with decameter-level analysis. Figure 1 This is one of the flow charts of the crop fine classification method based on the decameter-level analysis ready-to-use data product provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step 110, extracting multiple types of crop data from the crop identification area according to the prior knowledge base and the time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through the ten-meter-level analysis of the long time series of the crop identification area, i.e., the available data ARD product; the prior knowledge base is constructed based on all crop types, growth characteristics, and phenological information in the crop identification area.

[0024] In this step, crop data is obtained from the crop identification area of ​​the remote sensing image; specifically, appropriate satellite remote sensing images can be selected according to the needs of crop monitoring, such as data of different resolutions and bands provided by Landsat, Sentinel, MODIS and other satellites to meet the needs of different application scenarios.

[0025] In this embodiment, the types of crops to be detected in the crop identification area include multiple types, such as corn, soybean, rice, and wheat; these crops present different spectral characteristics and phenological characteristics in remote sensing images.

[0026] In this embodiment, the crop types, growth characteristics and corresponding phenological information of the crop identification area are collected from the ten-meter-level Analysis Ready Data (ARD) product of the crop identification area over a long period of time, and crop characteristics for distinguishing different crop types are extracted. These characteristic information are then converted into mathematical expressions to form a priori knowledge base.

[0027] In this embodiment, the above growth characteristics include the growth cycle of each crop, growth environment requirements (such as temperature, humidity, light, soil type, etc.), growth habits (such as upright, creeping, climbing, etc.) and physiological characteristics (such as photosynthesis efficiency, transpiration intensity, etc.).

[0028] In this embodiment, the phenological information includes the key phenological periods of each crop, such as sowing period, emergence period, tillering period, heading period, flowering period, fruiting period, maturity period, etc., and records the duration of each phenological period, environmental conditions and major growth changes of the crop in that stage.

[0029] In this embodiment, the above-mentioned crop characteristics can be a weighted combination of one or more of spectral characteristics (indicating the spectral reflectance characteristics of different crops at different growth stages, such as spectral reflectance in visible light, near infrared, mid-infrared and other bands), morphological characteristics (indicating morphological characteristics of crops such as plant height, leaf area, canopy structure, root distribution, etc.), physiological characteristics (indicating physiological parameters such as photosynthesis rate, transpiration rate, water use efficiency, etc. of crops) and phenological characteristics (the occurrence time, duration and corresponding growth changes of key phenological periods of crops).

[0030] In this embodiment, the following spectral features may include a normalized difference vegetation index (NDVI), a greenness index (GNDVI), and a red edge position index (REP), etc., wherein the normalized difference vegetation index (NDVI) may be represented by the following mathematical expression: NDVI = (NIR - RED) / (NIR + RED); Among them, NIR is the reflectance of the near infrared band, and RED is the reflectance of the visible red light band; the mathematical expressions of other vegetation indices are the same as the calculation methods in the prior art, and will not be repeated in this embodiment.

[0031] In this embodiment, the above mathematical expressions and related information are integrated into a priori knowledge base to facilitate subsequent analysis and model building.

[0032] In this embodiment, by preprocessing and data synthesis of a long time series of ten-meter-level ARD products in the crop identification area for many consecutive years (for example, 3 years), a synthetic time series dataset is obtained. For example, the time series dataset is a 10-day time series dataset.

[0033] Step 120: train a crop detection model using the first sample set as training samples, and identify all crop types in the crop identification area based on the crop detection model to obtain crop identification results; segment the fields in the crop identification area based on high-resolution remote sensing image segmentation technology to obtain field segmentation results.

[0034] In this step, the first sample set collected above can be directly used as training samples to identify crop types; In this embodiment, after obtaining the above-mentioned first sample set, preprocessing operations including atmospheric correction, geometric correction and radiation correction are performed on each image data in the sample set to eliminate atmospheric influence and improve the geometric accuracy of the image; image mosaicking, cropping and other processing can also be performed on each image data to better analyze a specific area.

[0035] In this embodiment, after the first sample set is preprocessed, the crop characteristics are obtained by performing feature analysis on each data in the first sample set.

[0036] For example, the spectral characteristics of crops can be extracted by utilizing the differences in reflectance characteristics of different crops in multiple bands such as visible light, near-infrared, and mid-infrared. The growth status of vegetation can be reflected by calculating vegetation indices (such as the normalized difference vegetation index NDVI or the enhanced vegetation index EVI).

[0037] In this embodiment, crop features are used to train a traditional machine learning algorithm or a neural network model to obtain a crop detection model.

[0038] In this embodiment, the traditional machine learning model includes support vector machine (SVM), random forest (RF) or K nearest neighbor (KNN), etc.; the neural network model includes convolutional neural network (CNN) model, etc.

[0039] In this embodiment, a suitable test sample is selected from the obtained crop recognition area and input into the crop detection model to obtain a crop recognition result.

[0040] In this embodiment, the field segmentation result is achieved through the following steps: (1) Obtain a high-resolution remote sensing image of the crop identification area A meters (e.g., 0.5 meters to 2 meters), and crop an image of a representative sub-area from the high-resolution remote sensing image; (2) performing object-oriented segmentation on the image of the representative sub-area; (3) Merge and filter the segmented objects to obtain the field distribution results of representative sub-areas as the initial samples for field identification; (4) Inputting the initial samples into a deep learning model or a traditional machine learning model, and training the model to obtain the field distribution results of the entire crop identification area, i.e., the field segmentation results mentioned above.

[0041] Step 130: Determine the crop fine classification result of the crop identification area based on the crop identification result and the field segmentation result.

[0042] In this step, the crop type recognition results can be superimposed with the field results to obtain the final crop fine classification result map of the crop recognition area.

[0043] Specifically, ArcGIS tools are used to convert the field segmentation results into vectors; the crop identification results are then superimposed with the vectors corresponding to the fields, and then the crop objects outside the fields are removed and the crop objects within the fields are supplemented to obtain a detailed crop classification result map of the entire crop identification area.

[0044] The crop fine classification method based on the ready-to-use data product of ten-meter level analysis provided by the embodiment of the present invention extracts multiple types of crop data from the crop identification area through a priori knowledge base and a time series synthetic data set, obtains a first sample set and trains a crop detection model based on it, so as to realize the identification of all crop types in the crop identification area, and then segments the fields in the crop identification area to obtain the field segmentation results, and finally superimposes the crop identification results and the field segmentation results to determine the crop fine classification results. By constructing a crop fine classification framework of the ready-to-use data product of ten-meter level analysis and combining it with the field identification technology, the planting structure can be finely described, and the accuracy of crop fine classification based on remote sensing data is improved.

[0045] In some embodiments, a time series synthetic data set is obtained by the following steps: collecting a long time series of 10-meter ARD products for a crop identification area for M consecutive years; removing pixels with poor preprocessing accuracy in the ARD products according to the quality identification of the ARD products to obtain new 10-meter ARD products; performing at least one of cropping, splicing, vegetation index calculation and synthesis on the new 10-meter ARD products according to the crop identification area to obtain an N-day time series synthetic data set, the N-day time series synthetic data set including multispectral band images and synthetic images of vegetation indices such as NDVI; M and N are both positive integers.

[0046] In this embodiment, ARD products can be obtained from satellite data providers or relevant research institutions; the required ARD products can be downloaded according to the geographic coordinates and time range (e.g., three consecutive years) of the crop identification area, ensuring that the downloaded data contains multi-spectral bands (such as red, green, blue, near infrared, etc.).

[0047] In this embodiment, according to the quality identification file of the ARD product, dedicated data processing software or platform (such as ENVI, ERDAS Imagine, Google Earth Engine, etc.) is used to identify and remove pixels affected by clouds, cloud shadows, snow, ice, etc.; the ARD product can also be cropped according to the boundary of the target area.

[0048] In this embodiment, if the downloaded data contains multiple files or scenes, these files need to be spliced ​​into a complete data set. According to the quality identification of the ARD product, pixels with poor preprocessing accuracy such as clouds and cloud shadows are removed; In this embodiment, the crop spectral characteristics may be calculated using the above-mentioned NDVI calculation method, which will not be described in detail in this embodiment.

[0049] In this embodiment, the above-mentioned synthesis includes time synthesis and space synthesis; wherein, the time synthesis includes: arranging three consecutive years of ARD products in chronological order, with each 10 days as a synthesis cycle, and performing statistical processing such as averaging, maximum value, minimum value, etc. on the data in each synthesis cycle to generate a 10-day synthetic multispectral band image.

[0050] Similarly, vegetation indices such as NDVI can also be synthesized over time to generate a 10-day composite vegetation index image.

[0051] In this embodiment, spatial synthesis includes: when the target area spans multiple ARD product scenarios, spatial synthesis is required to ensure data integrity.

[0052] The crop fine classification method based on the decameter-level analysis and ready-to-use data product provided in the embodiment of the present invention collects decameter-level ARD products of a continuous M-year long time series in a crop identification area, removes pixels with poor preprocessing accuracy in the ARD products, and then clips, splices, calculates vegetation indexes and synthesizes the new decameter-level ARD products according to the crop identification area to obtain an N-day time series synthetic data set, thereby obtaining a first sample set, improving the sample time resolution, and better capturing key growth nodes of crops, thereby improving the accuracy of crop fine classification.

[0053] In some embodiments, the crop identification area includes multiple sub-areas; the first sample set is extracted from one of the multiple sub-areas; after obtaining the first sample set and before training a crop detection model using the first sample set as a training sample, the method also includes: time-expanding the first sample set to obtain a sample set based on time expansion; and spatially expanding the sample set based on time expansion to obtain a second sample set.

[0054] In this embodiment, the crop identification area is divided into several sub-areas by analyzing factors such as precipitation, accumulated temperature, and topography in the crop identification area; from the several sub-areas, several representative sub-areas are determined and selected to construct the first sample set.

[0055] In this embodiment, for the selected representative sub-area, crop extraction rules are set according to the prior knowledge base, and based on the time series synthetic data set of the ten-meter ARD products of the target year, a hierarchical classification decision tree is established to extract various crop types to form a first sample set.

[0056] In this embodiment, the sampling time-space expansion technology is used to expand the first sample set to obtain high-precision crop type samples.

[0057] The time expansion and space expansion of the first sample set are described below.

[0058] (1) The first sample set is time-expanded to obtain a sample set based on time expansion, including: obtaining an N-day time series synthetic data set corresponding to the 10-meter ARD product of the long time series of Mi to Mj consecutive years in the crop identification area; M, i and j are all positive integers, and i and j are less than or equal to 5; for each sub-area, a crop hierarchical classification decision tree is constructed for each sub-area based on the N-day time series synthetic data set and the N-day time series synthetic data set corresponding to the 10-meter ARD product of the long time series of M consecutive years, and multiple types of crop data are extracted from the crop hierarchical classification decision tree to obtain a sample set based on time expansion.

[0059] In this embodiment, the values ​​of M, N, i and j can be set by the user, for example, M=3 or 5, i≠j=1, 2 or 3, etc., and N=10.

[0060] In this embodiment, the 10-day synthetic data set of the ARD product time series of the current sub-region in the two years before the target year is extracted in sequence; the following steps are performed year by year and sub-region by sub-region: The crop extraction rules are set according to the prior knowledge base, and based on the time series synthetic data set of domestic 10-meter ARD products in the target year, a hierarchical classification decision tree is established to extract various crop types to form the first sample set. In this way, the first sample set is extended in time to form the crop classification results of each sub-area for three consecutive years, and a sample set based on time extension is obtained.

[0061] (2) spatially expanding the sample set based on time expansion to obtain a second sample set, including: spatially expanding each sub-area according to a preset spatial buffer zone to obtain a plurality of expanded sub-areas; for each expanded sub-area, extracting crop distinguishing features from the sample set based on time expansion based on feature importance screening technology, and supervised labeling of different types of crop data in the sample set based on time expansion based on the crop distinguishing features to obtain a second sample set.

[0062] In this embodiment, representative sub-regions are selected from the crop identification area, and these sub-regions generally contain typical characteristics and spatial distribution of the target crop; and a spatial buffer zone is set around the representative sub-regions.

[0063] For example, the crop identification area is Province A, which is located in Plain D (the planting structure of the plain is similar to that of Province A). The crop identification area can be expanded from Province A to Plain D.

[0064] In this embodiment, an area with little change can be selected as a spatial buffer according to crop planting structure and topographic features. The size of the spatial buffer can be determined according to actual needs and the spatial distribution of data. Within the set buffer, new samples are generated by sampling or interpolation methods, thereby expanding the spatial range of the sample set based on time expansion.

[0065] In this embodiment, feature importance screening includes extracting feature variables related to the target crop class from a sample set based on time expansion. These feature variables may include color, texture, morphology, etc.; then using statistical methods or machine learning algorithms to evaluate the importance of each feature variable for distinguishing the target crop class from other categories; finally, based on the importance evaluation results, the feature variables that have the greatest influence on the identification of the target crop class are screened out for subsequent sample generation and model training.

[0066] In this embodiment, the similarity between each sample in the original sample set and other samples is calculated (implemented by the distance or similarity score between samples), and then the samples with similar characteristics are associated with the same target category (implemented by clustering algorithm or similarity threshold and other methods); for the newly generated samples, corresponding labels are automatically assigned to them according to their similarity with the labeled samples, so as to realize automatic supervised labeling of the samples and obtain the corresponding second sample set, so as to train the crop detection model and identify the crop type according to the second sample set, thus saving a lot of manual labeling costs.

[0067] The crop fine classification method based on the decameter-level analysis ready-to-use data product provided in the embodiment of the present invention can efficiently obtain a large number of high-precision crop type samples and improve the quality of crop samples by time expanding the first sample set to obtain a sample set based on time expansion, and spatially expanding the sample set based on time expansion to obtain a second sample set.

[0068] In some embodiments, after obtaining the second sample set, the method further includes: removing noise samples and sample edge pixels from the second sample set to obtain a third sample set.

[0069] It should be noted that noise samples refer to samples in the sample set that are inconsistent with the actual crop characteristics or deviate from the normal range due to data collection errors, sensor failures, environmental interference, etc. Noise samples will reduce the accuracy and reliability of the model.

[0070] Specifically, the sample data can be preprocessed by denoising and smoothing to reduce the impact of noise on the sample. The sample data can also be filtered using filtering algorithms (such as mean filtering, median filtering, etc.) to remove high-frequency noise. Statistical methods (such as mean, standard deviation, maximum value, minimum value, etc.) can also be used to calculate the statistical characteristics of the sample to identify outliers or extreme values ​​that are significantly different from the overall sample characteristics, that is, noise samples.

[0071] It should be noted that edge pixels refer to pixels located at the edge of the sample image. These pixels may produce errors due to edge effects during image acquisition or decreased sensor sensitivity; eliminating these edge pixels can improve the accuracy and reliability of the sample.

[0072] Specifically, the sample image is segmented using an image segmentation algorithm (such as multi-scale segmentation, edge-based segmentation, etc.) to identify the target area and background area in the sample image, as well as pixels located at the edge. The sample image can also be edge detected using an edge detection algorithm (such as Canny edge detection, Sobel edge detection, etc.) to identify edge pixels in the sample image, which may contain errors or noise. The pixels located at the edge of the sample image can also be removed based on the results of image segmentation and edge detection to ensure that the sample image after removal contains only pixels in the target area, so as to improve the accuracy and reliability of the sample. After the noise samples and edge pixels are removed, the remaining sample data are integrated and normalized to form a third sample set.

[0073] The crop fine classification method based on the decameter-level analysis and ready-to-use data product provided in the embodiment of the present invention further improves the quality of the sample set and thereby improves the accuracy of crop type recognition by eliminating noise samples and sample edge pixels from the second sample set to obtain the third sample set.

[0074] Figure 2 This is the second flow chart of the crop fine classification method based on the decameter-level analysis ready-to-use data product provided by the present invention. Figure 2 In the embodiment shown, a crop fine classification method based on the ready-to-use data product of ten-meter level analysis can also be implemented by the following steps: S1. Partition the target recognition area, determine and select representative sub-areas.

[0075] In this step, the crop types in the representative sub-area are rich and contain the crop planting characteristics of the entire area. For example, the number of crop types in the representative sub-area is greater than a preset threshold.

[0076] In this step, the sub-areas can be divided manually according to factors such as precipitation, accumulated temperature, topography, etc., or automatically using a partitioning algorithm; the representative sub-areas selected must be able to represent the crop planting conditions in different partitions.

[0077] For example, if the target area is City B in Province A, the representative sub-area is County C in City B.

[0078] S2. Collect domestic ten-meter-level ARD products of the target identification area for three consecutive years and preprocess them to form a 10-day synthetic time series dataset.

[0079] In this step, the domestically produced 10-meter-level ARD products of three consecutive years of long time series include the target year and the two years before the target year. This is because the crop planting pattern remains unchanged in the short term, and the added data of the first two years can be used for time expansion and supplementary samples.

[0080] Specifically, this embodiment uses domestic 10-meter ARD products based on 16m GF-1 and GF-6 and other multi-source remote sensing satellite data, which have been processed after geometric correction, radiation normalization, atmospheric correction, quality identification and standardized framing. When applied to crop classification, it is necessary to cut and splice according to the scope of the target area; in addition, domestic 10-meter ARD products are multi-spectral data with four bands of blue, green, red and near infrared, which can calculate vegetation index, NDVI, EVI, FVC, LAI, etc. Since ARD products integrate multi-source data and have a high observation frequency, a stable time series of 10 days of synthesis within a year can be formed.

[0081] For example, the target year is 2022, and the domestic ten-meter-level ARD products collected are for three years, namely 2021, 2022 and 2023; the calculated vegetation index is NDVI; the 10-day synthetic time series data set includes five bands: blue, green, red, near-infrared, and NDVI.

[0082] S3. Collect crop types, growth characteristics and corresponding phenological information in the target area to form a priori knowledge base.

[0083] In this step, phenological information can be collected through web pages, literature, field surveys, expert consultations, etc., to determine the key nodes and key characteristics of crop growth changes, and express the characteristic information mathematically.

[0084] Specifically, the main crop types in the above-mentioned County C are spring wheat and winter wheat in spring, and corn and soybeans in summer. Vegetables and fruit trees are also widely distributed in the cultivated fields; spring wheat is sown in March, turns green in April, and is harvested in June. The key node for identification is April, when the NDVI is higher than that of other crops; winter wheat is sown in October of the previous year, turns green in March of the following year, and is harvested in June. The key node for identification is from November of the previous year to February of the following year, and the NDVI in winter months is greater than that of other crops; corn and soybeans are both sown in mid-to-late June and harvested in mid-to-late September. The key node for identification is August, when the reflectivity of soybeans in the near-infrared band is higher than that of corn.

[0085] S4. For the selected representative sub-areas, crop extraction rules are set according to the prior knowledge base, and based on the time series synthetic dataset of domestic 10-meter ARD products in the target year, a hierarchical classification decision tree is established to extract crop types and form an initial sample set (corresponding to the first sample set).

[0086] In this step, for the selected representative sub-areas, the extraction rules for each crop are set according to the mathematical expressions in the prior knowledge base, and the threshold of each rule is determined.

[0087] In this embodiment, a 10-day synthetic data set of the time series of domestic ten-meter-level ARD products in the target year is first selected, a hierarchical classification decision tree is established, and the crop types are extracted one by one using rules and thresholds; then the above operation is performed on all sub-areas to obtain an initial sample set of crop classifications for all sub-areas; wherein the most appropriate extraction threshold can be determined based on the information collected in the prior knowledge base, and if there is real sample data, the threshold is also improved using the real sample data.

[0088] Specifically, taking spring wheat in County C as an example, the extraction rules and thresholds are as follows: .

[0089] S5. Based on the time series ARD data for the two years before the target year, the initial sample set of representative sub-regions is temporally expanded.

[0090] In this step, the temporal expansion method is to apply the extraction rules to the time series of the two years before the target year to extract the crop types to enrich the initial sample set.

[0091] In one embodiment, the target year 2023 is extended to 2021 and 2022 to obtain crop distribution conditions for three consecutive years, which is an initial sample set with a time extension of .

[0092] S6. Further expand the spatial range of the representative sub-area, perform spatial expansion on the initial sample set, and obtain a second sample set.

[0093] In this step, when expanding the spatial range of the representative sub-area, it can be expanded by selecting a cache area or by defining a range; after the spatial range is expanded, the key features of the samples are sorted by importance, and high-confidence samples are automatically marked based on similarity to further enrich the initial sample set.

[0094] Specifically, the spatial range of Huantai County is expanded to 1.5 times of the original range, and samples are automatically marked in the expanded area to obtain an initial sample set after spatial expansion.

[0095] S7. Optimize the sample set after time and space expansion to form a final initial sample set (corresponding to the third sample set).

[0096] In this step, the initial sample set after spatiotemporal expansion is optimized, and noise samples and edge pixels of samples in the sample set are removed to ensure the accuracy of each sample in the sample set, thereby forming the final initial sample set; wherein the optimization operation includes removing noise pixels and edge pixels of samples in the sample set, so as to delete samples that may have low accuracy.

[0097] Specifically, the initial sample set after time-space expansion is optimized, noise pixels are removed using a filtering method, and the edge portion of each sample in the sample set is deleted to ensure the accuracy of the sample.

[0098] S8. Use the final initial sample set as training samples and use machine learning or deep learning models to train the crop types of the entire target area.

[0099] In this step, a suitable machine learning or deep learning model is first constructed. Then, the initial sample set after spatiotemporal expansion is used as training samples, and the machine learning or deep learning model is used to train the crop types in the entire target area to form a detailed crop classification map for the target area in the target year. Among them, the machine learning model can use random forest, and the deep learning model can use CNN, UNET, etc., as well as various optimized and improved models.

[0100] Specifically, the deep learning models of random forest and multi-scale CNN combined with time series analysis were used to classify the crops in Zibo City using the sample sets after spatiotemporal expansion.

[0101] S9. Collect high-resolution remote sensing images of the target area with an area of ​​1 to 2 meters, crop the images of representative sub-areas for segmentation, obtain the field distribution of representative sub-areas, and use them as initial samples for training using a deep learning model to obtain the field distribution results of the entire target area.

[0102] In this step, the purpose of high-resolution remote sensing images is to obtain high-precision field distribution. Images of the same season of the target year should be selected as much as possible to ensure the accuracy of the fields.

[0103] Specifically, high-resolution remote sensing images use GF-2's 1m / 4m fused images; field identification uses the HTC multi-scale feature fusion model.

[0104] S10, superimposing the crop type recognition result with the field result to obtain a final crop fine classification result map of the target area.

[0105] In this step, when superimposing the crop identification results and the field identification results, the crop objects outside the field need to be removed and the crop objects inside the field need to be supplemented to obtain the final crop fine classification result map of the target area.

[0106] Specifically, the crop identification results of County C were superimposed on the field results to obtain the detailed crop classification results map of Huantai County in 2023.

[0107] The following is a description of the crop fine classification device based on the ready-to-use data product of ten-meter level analysis provided by the present invention. The crop fine classification device based on the ready-to-use data product of ten-meter level analysis described below and the crop fine classification method based on the ready-to-use data product of ten-meter level analysis described above can be referenced to each other.

[0108] Figure 3 Schematic diagram of the structure of the crop fine classification device based on the decameter-level analysis and ready-to-use data product provided by the present invention. Figure 3 As shown, the crop fine classification device based on the decameter-level analysis and ready-to-use data product includes: a sample extraction module 310, a recognition and segmentation module 320 and a crop fine classification module 330.

[0109] The sample extraction module 310 is used to extract multiple types of crop data from the crop identification area according to the prior knowledge base and the time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through the 10-meter-level analysis data product of the long time series of the crop identification area; the prior knowledge base is constructed based on all crop types, growth characteristics and phenological information in the crop identification area; The recognition and segmentation module 320 is used to train a crop detection model using the first sample set as training samples, and to identify all crop types in the crop recognition area based on the crop detection model to obtain a crop recognition result; and to segment the fields in the crop recognition area based on high-resolution remote sensing image segmentation technology to obtain a field segmentation result; The crop fine classification module 330 is used to determine the crop fine classification result of the crop identification area based on the crop identification result and the field segmentation result.

[0110] The crop fine classification device based on the ready-to-use data product of ten-meter level analysis provided by the embodiment of the present invention extracts multiple types of crop data from the crop identification area through a priori knowledge base and a time series synthetic data set, obtains a first sample set and trains a crop detection model based on it, thereby realizing the identification of all crop types in the crop identification area, and then segments the fields in the crop identification area to obtain field segmentation results, and finally superimposes the crop identification results and the field segmentation results to determine the crop fine classification results. By constructing a crop fine classification framework for the ready-to-use data product of ten-meter level analysis and combining it with field identification technology, the planting structure can be finely described, thereby improving the accuracy of crop fine classification based on remote sensing data.

[0111] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4As shown, the electronic device may include: a processor (processor) 410 , a communication interface (Communications Interface) 420 , a memory (memory) 430 and a communication bus 440 , wherein the processor 410 , the communication interface 420 , and the memory 430 communicate with each other through the communication bus 440 . The processor 410 can call the logic instructions in the memory 430 to execute a crop fine classification method based on a decameter-level analysis ready-to-use data product, the method comprising: extracting multiple types of crop data from a crop identification area according to a priori knowledge base and a time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through a long time series of decameter-level analysis ready-to-use data ARD products in the crop identification area; the priori knowledge base is constructed based on all crop types, growth characteristics and phenological information in the crop identification area; a crop detection model is trained using the first sample set as a training sample, and all crop types in the crop identification area are identified based on the crop detection model to obtain a crop identification result; fields in the crop identification area are segmented based on high-resolution remote sensing image segmentation technology to obtain field segmentation results; and a crop fine classification result in the crop identification area is determined based on the crop identification results and the field segmentation results.

[0112] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the crop fine classification method based on the decameter-level analysis and ready-to-use data products provided by the above-mentioned methods, the method including: extracting multiple types of crop data from the crop identification area according to a priori knowledge base and a time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through a long time series of decameter-level analysis and ready-to-use data ARD products in the crop identification area; the priori knowledge base is constructed based on all crop types, growth characteristics and phenological information in the crop identification area; a crop detection model is trained with the first sample set as a training sample, and all crop types in the crop identification area are identified based on the crop detection model to obtain a crop identification result; the field in the crop identification area is segmented based on high-resolution remote sensing image segmentation technology to obtain a field segmentation result; and the crop fine classification result of the crop identification area is determined based on the crop identification result and the field segmentation result.

[0114] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute a crop fine classification method based on a decameter-level analysis ready-to-use data product provided by the above-mentioned methods, the method comprising: extracting multiple types of crop data from a crop identification area according to a priori knowledge base and a time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through a long time series of decameter-level analysis ready-to-use data ARD products in the crop identification area; the priori knowledge base is constructed based on all crop types, growth characteristics and phenological information in the crop identification area; a crop detection model is trained using the first sample set as a training sample, and all crop types in the crop identification area are identified based on the crop detection model to obtain a crop identification result; fields in the crop identification area are segmented based on high-resolution remote sensing image segmentation technology to obtain field segmentation results; and a crop fine classification result for the crop identification area is determined based on the crop identification results and the field segmentation results.

[0115] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0116] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A crop fine classification method based on decameter-level analysis ready-to-use data products, characterized in that: include: Extracting multiple types of crop data from the crop identification area according to the prior knowledge base and the time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained by using the ARD product of the long time series of the crop identification area at the ten-meter level analysis; the prior knowledge base is constructed based on all the crop types, growth characteristics and phenological information in the crop identification area; A crop detection model is trained using the first sample set as training samples, and all crop types in the crop identification area are identified based on the crop detection model to obtain a crop identification result; and fields in the crop identification area are segmented based on high-resolution remote sensing image segmentation technology to obtain a field segmentation result; A crop fine classification result of the crop identification area is determined based on the crop identification result and the field segmentation result.

2. The crop fine classification method based on decameter-level analysis and ready-to-use data products according to claim 1 is characterized in that: The time series synthetic dataset is obtained through the following steps: Collect the 10-meter ARD products for a long time series of M consecutive years in the crop identification area; According to the quality mark of the ARD product, pixels with poor preprocessing accuracy in the ARD product are eliminated to obtain a new 10-meter-level ARD product; The new ten-meter-level ARD product is subjected to at least one of cropping, splicing, vegetation index calculation and synthesis according to the crop identification area to obtain an N-day time series synthetic data set, wherein the N-day time series synthetic data set includes multispectral band images and synthetic images of vegetation indices such as NDVI; M and N are both positive integers.

3. The crop fine classification method based on decameter-level analysis and ready-to-use data products according to claim 1 is characterized in that: The crop identification area includes a plurality of sub-areas; the first sample set is extracted from one of the plurality of sub-areas; After obtaining the first sample set and before training a crop detection model using the first sample set as training samples, the method further includes: Performing time expansion on the first sample set to obtain a sample set based on time expansion; The sample set based on time expansion is spatially expanded to obtain a second sample set.

4. The crop fine classification method based on decameter-level analysis and ready-to-use data products according to claim 3 is characterized in that: The step of time-expanding the first sample set to obtain a sample set based on time expansion includes: Obtain an N-day time series synthetic data set corresponding to the ten-meter-level ARD product of the long time series for the crop identification area for consecutive years Mi to Mj; M, i and j are all positive integers, and i and j are less than or equal to 5; For each sub-area, a crop hierarchical classification decision tree is constructed for each sub-area based on the N-day time series synthetic data set and the N-day time series synthetic data set corresponding to the ten-meter-level ARD product of the continuous M-year long time series, and multiple types of crop data are extracted from the crop hierarchical classification decision tree to obtain a sample set based on time expansion.

5. The crop fine classification method based on decameter-level analysis and ready-to-use data products according to claim 3 is characterized in that: The step of spatially expanding the sample set based on time expansion to obtain a second sample set comprises: Spatially expanding each sub-area according to a preset spatial buffer to obtain a plurality of expanded sub-areas; For each expanded sub-area, crop distinguishing features are extracted from the time-expanded sample set based on feature importance screening technology, and different types of crop data in the time-expanded sample set are supervised and labeled based on the crop distinguishing features to obtain a second sample set.

6. The crop fine classification method based on decameter-level analysis and ready-to-use data products according to claim 3 or 5, characterized in that: After obtaining the second sample set, the method further includes: Noise samples and sample edge pixels are removed from the second sample set to obtain a third sample set.

7. A crop fine classification device based on decameter-level analysis and ready-to-use data products, characterized in that: include: A sample extraction module is used to extract multiple types of crop data from the crop identification area according to a priori knowledge base and a time series synthetic data set to obtain a first sample set; wherein the time series synthetic data set is obtained through a ten-meter-level analysis, i.e., data ARD product, of a long time series of the crop identification area; and the priori knowledge base is constructed based on all crop types, growth characteristics, and phenological information in the crop identification area; an identification and segmentation module, configured to train a crop detection model using the first sample set as training samples, and identify all crop types in the crop identification area based on the crop detection model to obtain a crop identification result; and segment the fields in the crop identification area based on high-resolution remote sensing image segmentation technology to obtain a field segmentation result; The crop fine classification module is used to determine the crop fine classification result of the crop identification area based on the crop identification result and the field segmentation result.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for fine-grained crop classification based on decameter-level analysis and ready-to-use data products as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for fine-grained crop classification based on decameter-level analysis and ready-to-use data products as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for fine-grained crop classification based on decameter-level analysis and ready-to-use data products as described in any one of claims 1 to 6 is implemented.

Images