A field plot level crop planting area early monitoring method

Abstract

The application provides a field plot level crop planting area early monitoring method, comprising the following steps: acquiring satellite image data and crop sample data set of a target area, and preprocessing the satellite image data to obtain reconstructed time series satellite data; determining classification features for crop classification based on a feature screening algorithm; based on an image segmentation algorithm, segmenting the satellite image data according to plot data to obtain object units; performing spectral aggregation on each object unit to obtain aggregated images; training and evaluating an initial machine learning classifier according to the aggregated images, the classification features and the crop sample data set to obtain a trained crop classification model; inputting remote sensing images of a target year into the crop classification model to obtain field plot level crop early monitoring results of the target year. The application can significantly reduce the uncertainty caused by pixel-based classification, and improve the accuracy and timeliness of crop classification.

Description

Technical Field

[0001] This invention relates to the field of crop classification technology, and in particular to a method for early monitoring of crop planting area at the field level. Background Technology

[0002] Timely and accurate mapping of crop planting distribution and acquisition of area information are crucial for estimating crop yields, assessing food security, disaster prevention, and agricultural policy formulation. Satellite remote sensing, due to its high timeliness and low cost, is considered an effective method for monitoring crops and has been widely used for extracting crop planting distribution and area information.

[0003] like Figure 1 As shown, current crop mapping methods mainly utilize expert knowledge (such as phenology-based algorithms), machine learning (such as random forest algorithms), and deep learning (such as convolutional neural network algorithms). Based on training samples obtained through different methods, such as ground surveys, visual interpretation, open crop distribution data, and statistical sampling, and combined with preprocessed optical remote sensing and synthetic aperture radar data, they construct and filter effective classification feature libraries to distinguish different crops. Then, they obtain crop distribution results through different crop mapping methods.

[0004] There are two main strategies for crop mapping using satellite remote sensing: pixel-based and object-based. Currently, most studies on rapidly extracting crop planting information over large areas employ pixel-based strategies. This is because object-based crop information extraction often requires segmenting the original image using higher-resolution remote sensing data, and its algorithms are sometimes complex, time-consuming, and computationally intensive, making them difficult to implement. With the gradual improvement of satellite remote sensing resolution, such as the emergence of medium-to-high-resolution satellite remote sensing images like Sentinel-2, Planet, and Worldview, individual pixels are typically much smaller than ground objects. Pixel-based crop information extraction methods only utilize the spectral characteristics of a single pixel, ignoring the spatial relationships between different pixels within the same ground feature. This results in significant noise in the crop classification results, such as different classification results within the same plot, and the unavoidable salt-and-pepper effect due to image quality and classification errors. However, agricultural planting is usually carried out on a plot-by-plot basis. Considering that the pixels of medium- and high-resolution remote sensing images are much smaller than the plots where crops are planted, it is necessary to study object classification strategies. Before classifying crops, remote sensing images should be segmented into meaningful object units using spatial clustering methods, thereby improving the uncertainty brought about by pixel-based classification.

[0005] Furthermore, the publication of most accurate crop classification results currently has a certain lag, often becoming available some time after harvest, which significantly reduces the timeliness of crop classification results. This is mainly due to the difficulty in obtaining the training samples needed for crop mapping over large areas, and the loss of effective observations caused by cloud and rain interference during the growing season. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a method for early monitoring of crop planting area at the field level.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A method for early monitoring of crop planting area at the field level includes:

[0009] Satellite imagery data and crop sample datasets for the target area are acquired, and the satellite imagery data is preprocessed to obtain reconstructed time-series satellite data; the crop sample dataset includes ground samples and plot data.

[0010] The optimal combination of classification features for crop classification is determined based on a feature selection algorithm.

[0011] Based on the image segmentation algorithm, the satellite image data is segmented according to the land parcel data to obtain object units;

[0012] Spectral aggregation is performed on each of the object units to obtain an aggregated image;

[0013] The initial machine learning classifier is trained and evaluated based on the aggregated image, the classification features, and the crop sample dataset to obtain a trained crop classification model.

[0014] The remote sensing images of the target year are input into the crop classification model to obtain the early monitoring results of crops at the field level for the target year.

[0015] Preferably, after inputting the remote sensing image of the target year into the crop classification model to obtain the early monitoring results of crops at the field scale for the target year, the method further includes:

[0016] The accuracy and reliability of the early crop monitoring results at the field level are evaluated by calculating the confusion matrix using real ground samples. The area and accuracy of the early crop monitoring results at the field level are then corrected using the calculated accuracy to obtain the corrected planting area information for different crops.

[0017] Preferably, the satellite image data is Sentinel-2 surface reflectance data.

[0018] Preferably, the satellite image data is preprocessed, including: cloud removal, spatial filtering for noise reduction, time series synthesis, linear interpolation, and vegetation index calculation.

[0019] Preferably, the classification features include any one or more of the following: spectrum, vegetation index, texture, topography, and phenology.

[0020] Preferably, the image segmentation algorithm is a simple non-iterative clustering algorithm.

[0021] Preferably, image aggregation is performed on each of the object units to obtain an aggregated image, including:

[0022] Pixels of the object units that have similar spatial, spectral, and temporal characteristics after segmentation are merged into the same object;

[0023] Within the same object, feature parameters from different objects are integrated using an average or median composite method to form the aggregated image.

[0024] Preferably, the machine learning classifier is any one of support vector machine, random forest, or classification and regression tree.

[0025] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0026] This invention provides a method for early monitoring of crop planting area at the field level, comprising: acquiring satellite image data and crop sample dataset of a target area, and preprocessing the satellite image data to obtain reconstructed time-series satellite data; the crop sample dataset includes ground samples and plot data; determining classification features for crop classification based on a feature selection algorithm; segmenting the satellite image data according to the plot data based on an image segmentation algorithm to obtain object units; performing spectral aggregation on each object unit to obtain an aggregated image; training and evaluating an initial machine learning classifier based on the aggregated image, the classification features, and the crop sample dataset to obtain a trained crop classification model; and inputting remote sensing images of the target year into the crop classification model to obtain early monitoring results of crops at the field level for the target year. This invention, combined with an image segmentation algorithm, can monitor the distribution and area of ​​crop planting at the field level during the growing season, significantly reducing the uncertainty caused by pixel-based classification and improving the accuracy and timeliness of crop classification. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of a solution in the prior art;

[0029] Figure 2 This is a flowchart of a method provided in an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] This invention addresses the shortcomings of existing crop remote sensing information extraction methods by proposing a field-scale early crop monitoring approach that combines object-oriented classification strategies and early monitoring classification strategies with medium-to-high resolution Sentinel-2 data and a remote sensing cloud computing platform.

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] Figure 2 The method flowchart provided in the embodiments of the present invention is as follows: Figure 2 As shown, this invention provides a method for early monitoring of crop planting area at the field level, comprising:

[0034] Step 100: Acquire satellite imagery data and crop sample dataset of the target area, and preprocess the satellite imagery data to obtain reconstructed time-series satellite data; the crop sample dataset includes ground samples and plot data;

[0035] Step 200: Select and determine the optimal combination of features for crop classification based on the feature selection algorithm;

[0036] Step 300: Based on the image segmentation algorithm, determine the optimal image segmentation threshold according to the land parcel data, and segment the satellite image data to obtain object units;

[0037] Step 400: Perform spectral aggregation on each of the object units to obtain the aggregated image;

[0038] Step 500: Train and evaluate the initial machine learning classifier based on the aggregated image, the classification features, and the crop sample dataset to obtain a trained crop classification model;

[0039] Step 600: Input the remote sensing image of the target year into the crop classification model to obtain the early monitoring results of crops at the field scale for the target year.

[0040] Specifically, the medium-to-high resolution Sentinel-2 data used in this invention is used for image segmentation, followed by field-level early crop monitoring. Theoretically, field-level early crop information extraction can also be achieved using higher resolution satellite data such as Planet, Worldview, and Google Earth high-resolution images. However, ultra-high resolution images may increase the computational load of the crop classification process, making it difficult to achieve the goal of rapid early crop monitoring on a large scale.

[0041] Preferably, after step 600, the method further includes:

[0042] Step 700: Calculate the confusion matrix using real ground samples, evaluate the accuracy and reliability of the early crop monitoring results at the field level, and correct the accuracy and area of ​​the early crop monitoring results at the field level using the calculated accuracy to obtain the corrected planting area information for different crops.

[0043] Specifically, the coupled object-oriented early crop monitoring principle of this embodiment is as follows: Compared with pixel-based classification strategies and the uncertainties they bring, object-based classification strategies first divide remote sensing images into meaningful object units, and then evaluate their features through spatial, spectral, and temporal scales. This invention combines a remote sensing cloud computing platform to segment images over a large scale, and uses the classifier transfer method to construct classifiers in historical periods and apply them to the years of early monitoring, thereby achieving field-level early crop monitoring.

[0044] Optionally, the specific technical approach in this embodiment includes the following steps:

[0045] 1) Image preprocessing: Obtain all available Sentinel-2 surface reflectance data for the entire year in the corresponding area, and perform a series of preprocessing operations on the data, including cloud removal, spatial filtering and noise reduction, time series synthesis, linear interpolation, vegetation index calculation, etc., to obtain the reconstructed time series Sentinel-2 data.

[0046] 2) Training Sample Collection: Obtain abundant training sample data within the relevant area. This can be data from ground surveys, data interpreted visually, representative samples selected from existing crop distribution maps using various methods, automatically generated samples, or downscaling data using statistical data, or a combination of these methods. In short, ensure a sufficient number of diverse, representative, and evenly distributed training sample sets within the area. Additionally, acquire a certain amount of plot data, primarily for determining the segmentation size in image segmentation.

[0047] 3) Construction of classification features: By constructing numerous classification features including spectrum, vegetation index, texture, topography, phenology, etc., and then using feature selection algorithms, the most effective combination for distinguishing different crops can be selected, which can greatly improve the accuracy and computational efficiency of crop classification models.

[0048] 4) Image segmentation: Combining the image segmentation algorithms embedded in the GEE remote sensing cloud computing platform, such as the Simple Non-Iterative Clustering (SNIC) algorithm, the constructed time series Sentinel-2 data is segmented into spatially meaningful object units by using the spatial relationships between pixels.

[0049] 5) Plot data acquisition: Plot data can be used as a standard to determine the size of image segments. Plot data can be obtained through ground surveys or by artificial delineation by combining high-resolution images or color composite images of key stages of crop growth.

[0050] 6) SNIC segmentation size determination: Combining ground plot data, the variability of different indices within each plot is calculated and characterized by the standard deviation of these indices. The standard deviation of the indices within the image segmentation objects at different segmentation scales is calculated to ultimately determine the optimal segmentation size. Exemplarily, the image segmentation scale involved in this invention can be changed according to factors such as the size, shape, and planting habits of crop-growing plots in different regions. However, despite using different segmentation thresholds, the goal is to achieve plot-level crop monitoring and reduce the uncertainty brought about by pixel-scale classification.

[0051] 7) Image aggregation: After image segmentation by the SNIC algorithm, pixels with similar spatial, spectral, and temporal characteristics after segmentation are merged into the same object through the "cluster" band. Within the same object, the feature parameters of the entire different objects are synthesized by means of average or median values ​​to form an aggregated image.

[0052] 8) Crop classification model construction: On images aggregated into objects, a machine learning classifier is trained by combining training samples and selected classification features. Different machine learning classifiers can be selected, such as support vector machines, random forests, classification and regression trees, etc., to train the crop classification model. Here, Sentinel-2 images from different periods are continuously added and the accuracy of the model training is evaluated. When the model accuracy reaches saturation and no longer increases significantly, the earliest time when the crop can be identified is determined.

[0053] 9) Early monitoring is achieved through classifier transfer. By using the idea of ​​classifier transfer, the crop classification model trained with historical data is applied to the processed remote sensing image of the target year to obtain the early monitoring results of crops at the field scale in the target year.

[0054] 10) Accuracy calculation and area correction: The confusion matrix is ​​calculated using a portion of real ground samples to assess the accuracy and reliability of early crop monitoring results at the field level. The calculated accuracy is then used to further correct the area to obtain information on the planting area of ​​different crops.

[0055] The beneficial effects of this invention are as follows:

[0056] 1) This invention couples an object-oriented classification strategy, which can realize crop distribution mapping at the field level, significantly reducing the uncertainty brought about by the pixel-based classification strategy and improving the effect of crop classification.

[0057] 2) Through the image segmentation algorithm embedded in the GEE cloud computing platform, early monitoring of crops at the field level can be quickly achieved on a large scale.

[0058] 3) The early monitoring of crops in the target year was achieved through the classifier transfer concept, which is more timely than the crop monitoring information provided by most studies.

[0059] 4) The image segmentation scale and feature construction processes in this invention need to be modified and optimized in different regions according to the size, shape, crops, planting habits, etc.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0061] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of early monitoring of crop planting area at field level, characterized by, The method comprises the following steps: acquiring satellite image data and crop sample data set of a target area, and pre-processing the satellite image data to obtain reconstructed time series satellite data; the crop sample data set comprises ground samples and plot data; determining a classification optimal feature combination for crop classification based on a feature screening algorithm; determining an optimal image segmentation threshold based on the plot data, segmenting the satellite image data based on an image segmentation algorithm, and obtaining object units; performing spectral aggregation on each object unit to obtain an aggregated image; training and evaluating an initial machine learning classifier based on the aggregated image, the classification optimal feature combination and the crop sample data set, and obtaining a trained crop classification model; inputting remote sensing images of a target year into the crop classification model to obtain early crop monitoring results at a plot level in the target year; the classification features include any one or more of spectrum, vegetation index, texture, terrain and phenology; the image segmentation algorithm is a simple non-iterative clustering algorithm; performing spectral aggregation on each object unit to obtain an aggregated image, comprising: merging pixels of the object units having similar features in space, spectrum and time into the same object; integrating feature parameters in different objects in the same object by an average or median synthesis method to form the aggregated image.

2. The field-level crop planting area early monitoring method according to claim 1, characterized by, After inputting remote sensing images of a target year into the crop classification model to obtain early crop monitoring results at a plot level in the target year, the method further comprises: calculating a confusion matrix through ground true samples to evaluate the accuracy and reliability of the early crop monitoring results at a plot level, and correcting the early crop monitoring results at a plot level in accuracy and area based on the calculated accuracy to obtain corrected planting area information of different crops.

3. The field-level crop planting area early monitoring method according to claim 1, characterized by, The satellite image data is Sentinel-2 ground reflectance data.

4. The field-level crop planting area early monitoring method according to claim 1, characterized by, The pre-processing of the satellite image data comprises cloud removal, spatial filtering and denoising, time series synthesis, linear interpolation and vegetation index calculation.

5. The field-level crop planting area early monitoring method according to claim 1, characterized by, The machine learning classifier is any one of support vector machine, random forest or classification and regression tree.

Images