WO2021226976A1 - Soil available nutrient inversion method based on deep neural network - Google Patents

Soil available nutrient inversion method based on deep neural network Download PDF

Info

Publication number
WO2021226976A1
WO2021226976A1 PCT/CN2020/090396 CN2020090396W WO2021226976A1 WO 2021226976 A1 WO2021226976 A1 WO 2021226976A1 CN 2020090396 W CN2020090396 W CN 2020090396W WO 2021226976 A1 WO2021226976 A1 WO 2021226976A1
Authority
WO
WIPO (PCT)
Prior art keywords
soil
database
data
remote sensing
neural network
Prior art date
Application number
PCT/CN2020/090396
Other languages
French (fr)
Chinese (zh)
Inventor
张炜
吴晓伟
Original Assignee
安徽中科智能感知产业技术研究院有限责任公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 安徽中科智能感知产业技术研究院有限责任公司 filed Critical 安徽中科智能感知产业技术研究院有限责任公司
Priority to CN202080000932.XA priority Critical patent/CN111727443B/en
Priority to PCT/CN2020/090396 priority patent/WO2021226976A1/en
Publication of WO2021226976A1 publication Critical patent/WO2021226976A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/02—Agriculture; Fishing; Mining
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Definitions

  • the invention relates to a soil quick-acting nutrient inversion method based on a deep neural network.
  • Precision fertilization is an effective means to reduce environmental pollution caused by excessive fertilization and improve the quality of agricultural products.
  • the basis of precision fertilization is the accurate description of farmland soil fertility.
  • the existing soil testing schemes in agricultural practice have ground detection points that are too sparse and cannot be refined according to the field; the soil sampling and testing range is huge, the cycle is long, and the cost is high; the available nutrients vary greatly, and the frequency of soil sample collection cannot be satisfied.
  • Timely demand for agricultural production Limited by the cost of soil sample collection and testing, the evaluation of cultivated land fertility has low spatial density, high cost, and the separation of time and space of soil information accumulated over the years. Therefore, a large-scale, fast and dynamic method for obtaining information on the dynamic distribution of soil fertility is needed.
  • the purpose of the present invention is to provide a soil quick-acting nutrient inversion method based on a deep neural network to solve the problem that the modeling of quick-acting nutrient in the prior art is greatly affected by agricultural production activities and meteorological factors, and data is involved in inversion and prediction.
  • the high dimensionality makes it difficult to effectively perform prediction and calculation.
  • the existing satellite remote sensing requires high time intervals for images, which leads to the problem of inability to track the rapid changes of available nutrients in time.
  • the described method for retrieving soil available nutrients based on deep neural network includes the following steps:
  • Step 1 Establishment of satellite remote sensing database
  • Step 2 Establish a remote sensing database for drones, the data collection interval of the remote sensing database for drones is shorter than that of the satellite remote sensing database;
  • Step 3 Establish a basic database of farmland environment
  • Step 4 In the training experiment area, obtain the test value of ground available nutrients by sampling and surveying soil nutrients at several survey points on-site, and form a soil nutrient distribution test map according to the location of the survey points;
  • Step 5 Modeling soil quick-acting nutrients. Satellite remote sensing database, UAV remote sensing database and farmland environment basic database provide data as covariate information of soil quick-acting nutrients. Convolutional neural network learns and trains through gradient backpropagation algorithm to obtain soil quick-acting Nutrient inversion model;
  • Step 6 Obtain the covariate information of the area that needs to be predicted, and input the soil available nutrient inversion model after preprocessing to produce a soil nutrient distribution map in the area.
  • the step 5 includes:
  • S5.1 Generate regression matrix. According to the location of the survey point, satellite remote sensing database, drone remote sensing database and farmland environment basic database provide corresponding data as the covariate information of soil available nutrients, covering the superimposed covariate information to generate the regression matrix.
  • the learning process of the convolutional neural network in step S5.2 is: convolve through three trainable filters and an addable bias, generate three feature maps in the C1 layer after convolution, and then In the feature map, each group of several features are summed, weighted, and biased, and then the three S2 layer soil available nutrient feature maps are obtained through the Sigmoid function; the soil available nutrient feature map is filtered to obtain the C3 layer, The hierarchical structure of the C3 layer generates the S4 layer by generating the S2 layer; finally, the covariate information formation vector is input to the convolutional neural network, and the output is obtained to obtain the soil available nutrient inversion model.
  • the satellite remote sensing database in step 1 includes satellite remote sensing data, and also includes field boundaries, crop planting types, crop growth and yield level data obtained from the inversion of satellite remote sensing data; in step 2, drone remote sensing
  • the database includes periodically acquired UAV remote sensing data and inverted data on field boundaries, crop planting types, crop conditions, and yield levels.
  • the farmland environment basic database in the step 3 includes a crop phenology database, a farmland environment database, an agrometeorological database, a soil basic geographic database, and a land use cover database, and data from the farmland environment database and the agrometeorological database
  • the collection interval is shorter than the satellite remote sensing database.
  • the data for constructing the crop phenology database includes information such as main crop varieties, planting time, harvesting time, time of different growth periods of crops, crop biomass, leaf area index, row density, ridge density, and sampling plant yield;
  • the data for constructing the farmland environment database includes soil temperature, soil humidity, pH value, air temperature and humidity, and occurrence data of diseases, pests and weeds at different time nodes of the crop growth process;
  • the data for constructing the agrometeorological database includes solar radiation, minimum temperature, maximum temperature, water vapor pressure, average wind speed, precipitation, and meteorological satellite data;
  • the data for constructing the soil basic geographic database includes soil historical soil testing formula data, soil type, slope and altitude;
  • the construction of a land use cover database includes land cover types.
  • the satellite remote sensing database when the satellite remote sensing database lacks data belonging to a certain inversion time interval, the closest satellite remote sensing data before and after the time interval and the related data obtained from the inversion are regarded as the inversion At the same time, the UAV remote sensing data is the data collected during the inversion time interval.
  • the present invention has the following advantages: through the construction of the soil quick-acting nutrient retrieval method based on deep neural network, deep reinforcement learning and the air-space-earth integrated data model are combined to improve the adaptability of the model, even if the data dimension is relatively high, the data source It can also perform inversion and prediction more accurately, and obtain a reliable soil nutrient distribution map, which overcomes the shortcomings of low spatial resolution and low overall accuracy of the prior art prediction model.
  • the more dimensional prediction model can pass the closest satellite remote sensing data and combine the time.
  • the input data such as UAV remote sensing data collected in the interval obtains more accurate output results, avoiding the problem that the existing satellite remote sensing technology cannot satisfy the modeling of quick-acting nutrients that require high data timeliness due to the long effective data collection time interval. problem.
  • satellite remote sensing data is used to participate in the prediction, which can avoid the shortcomings of UAV remote sensing data due to the limited collection range of UAVs and the inversion and prediction of more space constraints.
  • Fig. 1 is an overall flow chart of the method for retrieving soil available nutrients based on deep neural network according to the present invention
  • FIG. 2 is a schematic diagram of data transmission and specific flow of the large-scale soil nutrient inversion method based on deep neural network of the present invention
  • Fig. 3 is a schematic diagram showing the modeling process in the present invention.
  • the present invention provides a method for retrieving soil available nutrients based on a deep neural network, which includes the following steps:
  • Satellite remote sensing data includes satellite high-resolution remote sensing image data and high-precision digital elevation data.
  • the satellite remote sensing database includes 0.26 meter resolution remote sensing image data collected by Sentinel-2 satellite, GF-1 PMS, GF-2 PMS, LandSat ETM/OLI, HJ-1 CCD and Google satellites, and digital elevation with a precision of about 30M.
  • (DEM) data also includes field boundaries, crop planting types, crop growth and yield levels derived from satellite remote sensing data inversion. Combining multi-spectral and hyper-spectral satellite data with other data can remotely sense crop conditions in a large area.
  • Step 2 The establishment of a UAV remote sensing database.
  • the UAV remote sensing database includes periodically acquired UAV remote sensing data and data on field boundaries, crop planting types, crop growth conditions, and yield levels obtained from the inversion of UAV remote sensing data.
  • the data collection interval of the UAV remote sensing database is shorter than that of the satellite remote sensing database, but the UAV has low collection efficiency in a large area, so the spatial range of the collected data is smaller than the satellite remote sensing data.
  • Step 3 Establishment of a basic database of farmland environment.
  • the farmland environment basic database includes a crop phenology database, a farmland environment database, an agrometeorological database, a soil basic geographic database, and a land use cover database, and the data collection interval of the farmland environment database and the agrometeorological database is shorter than that of the satellite remote sensing database.
  • the data for constructing the crop phenology database includes information such as main crop varieties, planting time, harvesting time, different growth periods of crops, crop biomass, leaf area index, row density, ridge density, and sampling plant yield.
  • the data for constructing the farmland environment database includes soil temperature, soil humidity, pH value, air temperature and humidity, and occurrence data of diseases, pests and weeds at different time nodes of the crop growth process.
  • the data for constructing agrometeorological database includes solar radiation, minimum temperature, maximum temperature, water vapor pressure, average wind speed, precipitation, and meteorological satellite data.
  • the data for constructing the soil basic geographic database includes soil historical soil testing formula data, soil type, slope and altitude.
  • the construction of the land use cover database includes land cover types, such as paddy field, dry land, forest land, shrub forest, sparse woodland, grassland, etc.
  • Step 4 In the training experiment area, the ground available nutrient test values are obtained by field sampling and survey of soil nutrients at several survey points, and a soil nutrient distribution test map is formed according to the location of the survey points.
  • Sampling and testing indicators include soil nutrient (organic matter, total nitrogen, total phosphorus, total potassium) storage indicators, nutrient availability status (the content and ratio of nutrients that can be absorbed and used by plants, such as available phosphorus/total phosphorus, available potassium/total potassium) ), the quantity and activity of soil organisms.
  • the index of soil nutrients is determined through the principles of soil chemistry, and the effect of ground available nutrients on soil fertility is numerically expressed. Then combined with the location of the survey point and map information, it can be transformed into a soil nutrient distribution test map.
  • Step 5 Modeling soil quick-acting nutrients. Satellite remote sensing database, UAV remote sensing database and farmland environment basic database provide data as covariate information of soil quick-acting nutrients. Convolutional neural network learns and trains through gradient backpropagation algorithm to obtain soil quick-acting Nutrient inversion model.
  • S5.1 Generate regression matrix. According to the location of the survey point, satellite remote sensing database, drone remote sensing database and farmland environment basic database provide corresponding data as the covariate information of soil available nutrients, covering the superimposed covariate information to generate the regression matrix.
  • the neural network learning framework and process as shown in Figure 3 are as follows: the input value is convolved through three trainable filters and an addable bias. After convolution, three feature maps are generated in the C1 layer, and then the feature maps After each group of several features are summed, weighted, and biased, the three S2 layer soil available nutrient feature maps are obtained through the Sigmoid function; the soil available nutrient feature map is filtered to obtain the C3 layer and the C3 layer.
  • the hierarchical structure generates the S4 layer by generating the S2 layer; finally, the covariate information formation vector is input to the convolutional neural network, and the output is obtained to obtain the soil available nutrient inversion model.
  • Step 6 Obtain the covariate information of the area that needs to be predicted, and input the soil available nutrient inversion model after preprocessing to produce a soil nutrient distribution map in the area.
  • This step needs to collect the multi-spectral/hyper-spectral satellite image of the inversion area, the historical meteorological data within one year of the area, the monitoring data of the crop growth of the area by the drone, etc.
  • the collected data includes multiple dimensions and types. The number is large, and the prediction result is accurate.
  • this method uses the closest satellite remote sensing data before and after the time interval and the related data obtained from the inversion as the input value of the inversion, and the UAV remote sensing data is collected in the inversion time interval. data.
  • input data such as UAV remote sensing data collected in the time interval
  • more accurate output results can be obtained by calculation through the model, which avoids the problem of inaccurate results of the modeling of quick-acting nutrients due to the long interval of satellite remote sensing data collection.

Abstract

Disclosed is a soil available nutrient inversion method based on a deep neural network. The method comprises the following steps: step 1, establishing a satellite remote sensing database; step 2, establishing an unmanned aerial vehicle remote sensing database; step 3, establishing a farmland environmental foundation database; step 4, performing in-situ sampling and surveying at several survey points in a training experiment area, so as to form a soil nutrient distribution test diagram; step 5, performing modeling for available nutrients of soil, and performing learning training on a convolutional neural network by means of a gradient back-propagation algorithm, so as to obtain a soil available nutrient inversion model; and step 6, acquiring covariate information, and pre-processing the covariable information and then inputting same into the soil available nutrient inversion model, so as to generate a soil nutrient distribution diagram of an area. By means of the present method, inversion prediction can be more accurately performed to obtain a reliable soil nutrient distribution diagram, thereby overcoming the defects in the prior art of low spatial resolution and low overall accuracy of a prediction model.

Description

一种基于深度神经网络的土壤速效养分反演方法An Inversion Method of Soil Available Nutrients Based on Deep Neural Network 技术领域Technical field
本发明涉及一种基于深度神经网络的土壤速效养分反演方法。The invention relates to a soil quick-acting nutrient inversion method based on a deep neural network.
背景技术Background technique
精准施肥是减少滥施肥引起的环境污染、提升农产品品质的有效手段，而精准施肥的基础是对农田土壤肥力的精准刻画。然而现有测土方案在农业实践中存在着地面检测点过于稀疏，不能精细的“因田施策”；土壤采样测试范围巨大、周期长、成本高；速效养分变化大，土样采集频次不能满足农业生产及时的需求。受到土样采集测试成本的限制，耕地地力评价存在检测空间密度低、成本高、历年积累的土壤信息存在时间与空间的割裂。因此需要大规模、快速和动态的土壤肥力动分布信息的获取方法。Precision fertilization is an effective means to reduce environmental pollution caused by excessive fertilization and improve the quality of agricultural products. The basis of precision fertilization is the accurate description of farmland soil fertility. However, the existing soil testing schemes in agricultural practice have ground detection points that are too sparse and cannot be refined according to the field; the soil sampling and testing range is huge, the cycle is long, and the cost is high; the available nutrients vary greatly, and the frequency of soil sample collection cannot be satisfied. Timely demand for agricultural production. Limited by the cost of soil sample collection and testing, the evaluation of cultivated land fertility has low spatial density, high cost, and the separation of time and space of soil information accumulated over the years. Therefore, a large-scale, fast and dynamic method for obtaining information on the dynamic distribution of soil fertility is needed.
随着遥感技术和图像处理技术的快速发展，很多研究开展了基于遥感多光谱和高光谱数据对裸土或植被的监测，从而反演土壤肥力分布。但目前基于遥感技术对土壤肥力进行建模，主要集中在非速效养分的土壤属性的建模，如pH值，有机质和土壤容重等，而对测土配方检测指标中速效养分的建模较少。因为速效养分的建模受农业生产活动和气象因素影响较大，因而需要更高时间分辨率的遥感数据和更快速的土壤采样检测方法。基于卫星遥感对影像的时间间隔要求高，导致无法跟踪速效养分的快速变化；建模过程中，随着大数据的爆炸式增长，数据维度比较高，目前没系统对多种来源的数据进行自动学习和处理。With the rapid development of remote sensing technology and image processing technology, many studies have carried out monitoring of bare soil or vegetation based on remote sensing multispectral and hyperspectral data to invert the distribution of soil fertility. However, the current modeling of soil fertility based on remote sensing technology mainly focuses on the modeling of soil properties of non-available nutrients, such as pH value, organic matter and soil bulk density, etc., while modeling of available nutrients in soil testing formula detection indicators is less. . Because the modeling of available nutrients is greatly affected by agricultural production activities and meteorological factors, remote sensing data with higher time resolution and faster soil sampling and detection methods are required. Satellite remote sensing requires high time intervals for images, which makes it impossible to track the rapid changes of available nutrients. During the modeling process, with the explosive growth of big data, the data dimension is relatively high. Currently, there is no system to automatically automate data from multiple sources. Learning and processing.
因此，如何在复杂环境下进行土壤养分建模，提升模型泛化能力、预测精度和迭代能力已成为急需解决的技术问题。Therefore, how to perform soil nutrient modeling in a complex environment to improve the generalization ability, prediction accuracy and iteration ability of the model has become an urgent technical problem to be solved.
发明内容Summary of the invention
本发明的目的在于提供一种基于深度神经网络的土壤速效养分反演方法，以解决现有技术中因为速效养分的建模受农业生产活动和气象因素影响较大，进行反演预测时涉及数据维度高，难以有效进行预测计算，同时现有卫星遥感对影像的时间间隔要求高，导致无法及时跟踪速效养分的快速变化的问题。The purpose of the present invention is to provide a soil quick-acting nutrient inversion method based on a deep neural network to solve the problem that the modeling of quick-acting nutrient in the prior art is greatly affected by agricultural production activities and meteorological factors, and data is involved in inversion and prediction. The high dimensionality makes it difficult to effectively perform prediction and calculation. At the same time, the existing satellite remote sensing requires high time intervals for images, which leads to the problem of inability to track the rapid changes of available nutrients in time.
所述的一种基于深度神经网络的土壤速效养分反演方法，包括以下步骤：The described method for retrieving soil available nutrients based on deep neural network includes the following steps:
步骤1、卫星遥感数据库的建立；Step 1. Establishment of satellite remote sensing database;
步骤2：无人机遥感数据库的建立，所述无人机遥感数据库的数据采集间隔时间小于所述卫星遥感数据库；Step 2: Establish a remote sensing database for drones, the data collection interval of the remote sensing database for drones is shorter than that of the satellite remote sensing database;
步骤3：农田环境基础数据库的建立；Step 3: Establish a basic database of farmland environment;
步骤4：在训练实验区域通过在若干个调查点现场采样调查土壤养分获得地面速效养分测试值，根据调查点位置形成土壤养分分布测试图；Step 4: In the training experiment area, obtain the test value of ground available nutrients by sampling and surveying soil nutrients at several survey points on-site, and form a soil nutrient distribution test map according to the location of the survey points;
步骤5：土壤速效养分建模，卫星遥感数据库、无人机遥感数据库和农田环境基础数据库提供数据作为土壤速效养分的协变量信息，卷积神经网络通过梯度反向传播算法进行学习训练得到土壤速效养分反演模型；Step 5: Modeling soil quick-acting nutrients. Satellite remote sensing database, UAV remote sensing database and farmland environment basic database provide data as covariate information of soil quick-acting nutrients. Convolutional neural network learns and trains through gradient backpropagation algorithm to obtain soil quick-acting Nutrient inversion model;
步骤6：获取需要进行预测的区域的协变量信息，经预处理后输入土壤速效养分反演模型，产出该区域土壤养分分布图。Step 6: Obtain the covariate information of the area that needs to be predicted, and input the soil available nutrient inversion model after preprocessing to produce a soil nutrient distribution map in the area.
优选的，所述步骤5包括：Preferably, the step 5 includes:
S5.1：生成回归矩阵，根据调查点的位置，卫星遥感数据库、无人机遥感数据库和农田环境基础数据库提供相应数据作为土壤速效养分的协变量信息，覆盖叠加协变量信息生成回归矩阵。S5.1: Generate regression matrix. According to the location of the survey point, satellite remote sensing database, drone remote sensing database and farmland environment basic database provide corresponding data as the covariate information of soil available nutrients, covering the superimposed covariate information to generate the regression matrix.
S5.2：拟合空间预测模型，设计卷积神经网络，其同一平面上的神经元权值相等，将回归矩阵转化为向量输入，将土壤养分分布测试图作为学习目标，通过梯度反向传播算法进行学习训练，得到土壤速效养分反演模型。S5.2: Fit the spatial prediction model, design the convolutional neural network, the weights of the neurons on the same plane are equal, the regression matrix is converted into vector input, the soil nutrient distribution test chart is used as the learning target, and the gradient is backpropagated The algorithm performs learning and training, and obtains the soil available nutrient inversion model.
优选的，所述步骤S5.2中卷积神经网络的学习过程为：通过三个可训练的滤波器和可加偏置进行卷积，卷积后在C1层产生三个特征映射图，然后特征映射图中每组若干特征再进行求和、加权值、加偏置，再通过Sigmoid函数得到三个S2层的土壤速效养分特征映射图；土壤速效养分特征映射图再经过滤波得到C3层，C3层的层级结构通过产生S2层的方法产生S4层；最终，协变量信息形成向量输入到卷积神经网络，得到输出获得土壤速效养分反演模型。Preferably, the learning process of the convolutional neural network in step S5.2 is: convolve through three trainable filters and an addable bias, generate three feature maps in the C1 layer after convolution, and then In the feature map, each group of several features are summed, weighted, and biased, and then the three S2 layer soil available nutrient feature maps are obtained through the Sigmoid function; the soil available nutrient feature map is filtered to obtain the C3 layer, The hierarchical structure of the C3 layer generates the S4 layer by generating the S2 layer; finally, the covariate information formation vector is input to the convolutional neural network, and the output is obtained to obtain the soil available nutrient inversion model.
优选的，所述步骤1中卫星遥感数据库中包括卫星遥感数据，还包括卫星遥感数据反演所得的田块边界、作物种植类型、作物长势及单产水平数据；所述步骤2中无人机遥感数据库包括周期性获取的无人机遥感数据及其反演所得的田块边界、作物种植类型、作物长势及单产水平数据。Preferably, the satellite remote sensing database in step 1 includes satellite remote sensing data, and also includes field boundaries, crop planting types, crop growth and yield level data obtained from the inversion of satellite remote sensing data; in step 2, drone remote sensing The database includes periodically acquired UAV remote sensing data and inverted data on field boundaries, crop planting types, crop conditions, and yield levels.
优选的，所述步骤3中所述农田环境基础数据库包括农作物物候数据库、农田环境数据库、农业气象数据库、土壤基础地理数据库和土地利用覆盖数据库，所述农田环境数据库和所述农业气象数据库的数据采集间隔时间小于所述卫星遥感数据库。Preferably, the farmland environment basic database in the step 3 includes a crop phenology database, a farmland environment database, an agrometeorological database, a soil basic geographic database, and a land use cover database, and data from the farmland environment database and the agrometeorological database The collection interval is shorter than the satellite remote sensing database.
优选的，所述构建农作物物候数据库的数据包括主要作物品种、种植时间、收割时间、作物不同生育期时间、作物生物量、叶面积指数、行密度、垄密度和抽样株单产等信息；Preferably, the data for constructing the crop phenology database includes information such as main crop varieties, planting time, harvesting time, time of different growth periods of crops, crop biomass, leaf area index, row density, ridge density, and sampling plant yield;
所述构建农田环境数据库的数据包括作物生长过程不同时间节点的土壤温度、土壤湿度、pH值、空气温湿度和病虫草害发生情况数据；The data for constructing the farmland environment database includes soil temperature, soil humidity, pH value, air temperature and humidity, and occurrence data of diseases, pests and weeds at different time nodes of the crop growth process;
所述构建农业气象数据库的数据包括太阳辐射、最低温度、最高温度、水汽压、平均风速、降水量以及气象卫星数据；The data for constructing the agrometeorological database includes solar radiation, minimum temperature, maximum temperature, water vapor pressure, average wind speed, precipitation, and meteorological satellite data;
所述构建土壤基础地理数据库的数据包括土壤历史测土配方数据、土壤类型、坡度和海拔高度；The data for constructing the soil basic geographic database includes soil historical soil testing formula data, soil type, slope and altitude;
所述构建土地利用覆盖数据库包括土地覆被类型。The construction of a land use cover database includes land cover types.
优选的，所述步骤6中，当卫星遥感数据库缺乏属于某一反演时间区间内的数据时，将该时间区间前后最接近的卫星遥感数据及其反演所得的相关数据作为该次反演的输入值，同时无人机遥感数据为该反演时间区间内采集的数据。Preferably, in the step 6, when the satellite remote sensing database lacks data belonging to a certain inversion time interval, the closest satellite remote sensing data before and after the time interval and the related data obtained from the inversion are regarded as the inversion At the same time, the UAV remote sensing data is the data collected during the inversion time interval.
本发明具有如下优点：通过基于深度神经网络的土壤速效养分反演方法的构建，将深度强化学习和空天地一体化的数据模型相结合，提高模型的适应能力，即使数据维度比较高，数据来源多样，也能较准确地进行反演预测，得到可靠的土壤养分分布图，克服现有技术预测模型空间分辨率不高，总体准确率低的缺陷。The present invention has the following advantages: through the construction of the soil quick-acting nutrient retrieval method based on deep neural network, deep reinforcement learning and the air-space-earth integrated data model are combined to improve the adaptability of the model, even if the data dimension is relatively high, the data source It can also perform inversion and prediction more accurately, and obtain a reliable soil nutrient distribution map, which overcomes the shortcomings of low spatial resolution and low overall accuracy of the prior art prediction model.
同时由于采用无人机遥感数据库和农田环境基础数据库进行补充验证，因此更多维度的预测模型即使出现缺少反演的时间区间内采集的卫星遥感数据，也能通过最接近卫星遥感数据，结合时间区间内采集的无人机遥感数据等输入数据得到较为准确的输出结果，避免了现有卫星遥感技术因有效数据采集时间间隔较长无法满足对数据时效性要求较高的速效养分的建模的问题。同时大部分情况下通过卫星遥感数据参与预测，能避免无人机遥感数据因无人机采集范围有限，反演预测受空间约束较多的缺陷。At the same time, due to the supplementary verification of the UAV remote sensing database and the basic farmland environment database, even if the satellite remote sensing data collected in the time interval lacking inversion occurs, the more dimensional prediction model can pass the closest satellite remote sensing data and combine the time. The input data such as UAV remote sensing data collected in the interval obtains more accurate output results, avoiding the problem that the existing satellite remote sensing technology cannot satisfy the modeling of quick-acting nutrients that require high data timeliness due to the long effective data collection time interval. problem. At the same time, in most cases, satellite remote sensing data is used to participate in the prediction, which can avoid the shortcomings of UAV remote sensing data due to the limited collection range of UAVs and the inversion and prediction of more space constraints.
附图说明Description of the drawings
图1为本发明基于深度神经网络的土壤速效养分反演方法的整体流程图；Fig. 1 is an overall flow chart of the method for retrieving soil available nutrients based on deep neural network according to the present invention;
图2为本发明基于深度神经网络的大规模土壤养分反演方法数据传输和具体流程的示意图；2 is a schematic diagram of data transmission and specific flow of the large-scale soil nutrient inversion method based on deep neural network of the present invention;
图3为本发明中表示建模过程的示意图。Fig. 3 is a schematic diagram showing the modeling process in the present invention.
具体实施方式Detailed ways
下面对照附图，通过对实施例的描述，对本发明具体实施方式作进一步详细的说明，以帮助本领域的技术人员对本发明的发明构思、技术方案有更完整、准确和深入的理解。With reference to the accompanying drawings, the specific implementation of the present invention will be further described in detail through the description of the embodiments to help those skilled in the art have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.
如图1-3所示，本发明提供了一种基于深度神经网络的土壤速效养分反演方法，包括以下步骤：As shown in Figures 1-3, the present invention provides a method for retrieving soil available nutrients based on a deep neural network, which includes the following steps:
步骤1、卫星遥感数据库的建立。卫星遥感数据包括卫星高分辨率遥感影像数据和高精度数字高程数据。卫星遥感数据库中包括历年Sentinel-2卫星、GF-1 PMS、GF-2 PMS，LandSat ETM/OLI、HJ-1 CCD以及谷歌卫星采集的0.26米分辨率遥感影像数据，以及30M左右精度的数字高程(DEM)数据，还包括卫星遥感数据反演所得的田块边界、作物种植类型、作物长势及单产水平数据。多光谱、高光谱卫星数据与其他数据结合能在大范围区域对作物情况进行遥感采集。Step 1. The establishment of satellite remote sensing database. Satellite remote sensing data includes satellite high-resolution remote sensing image data and high-precision digital elevation data. The satellite remote sensing database includes 0.26 meter resolution remote sensing image data collected by Sentinel-2 satellite, GF-1 PMS, GF-2 PMS, LandSat ETM/OLI, HJ-1 CCD and Google satellites, and digital elevation with a precision of about 30M. (DEM) data also includes field boundaries, crop planting types, crop growth and yield levels derived from satellite remote sensing data inversion. Combining multi-spectral and hyper-spectral satellite data with other data can remotely sense crop conditions in a large area.
步骤2：无人机遥感数据库的建立，无人机遥感数据库包括周期性获取的无人机遥感数据及其反演所得的田块边界、作物种植类型、作物长势及单产水平数据。所述无人机遥感数据库的数据采集间隔时间小于所述卫星遥感数据库，但无人机在大范围地域采集的效率较低，因此采集数据的空间范围小于卫星遥感数据。Step 2: The establishment of a UAV remote sensing database. The UAV remote sensing database includes periodically acquired UAV remote sensing data and data on field boundaries, crop planting types, crop growth conditions, and yield levels obtained from the inversion of UAV remote sensing data. The data collection interval of the UAV remote sensing database is shorter than that of the satellite remote sensing database, but the UAV has low collection efficiency in a large area, so the spatial range of the collected data is smaller than the satellite remote sensing data.
步骤3：农田环境基础数据库的建立。所述农田环境基础数据库包括农作物物候数据库、农田环境数据库、农业气象数据库、土壤基础地理数据库和土地利用覆盖数据库，所述农田环境数据库和所述农业气象数据库的数据采集间隔时间小于所述卫星遥感数据库。Step 3: Establishment of a basic database of farmland environment. The farmland environment basic database includes a crop phenology database, a farmland environment database, an agrometeorological database, a soil basic geographic database, and a land use cover database, and the data collection interval of the farmland environment database and the agrometeorological database is shorter than that of the satellite remote sensing database.
所述构建农作物物候数据库的数据包括主要作物品种、种植时间、收割时间、作物不同生育期时间、作物生物量、叶面积指数、行密度、垄密度和抽样株单产 等信息。The data for constructing the crop phenology database includes information such as main crop varieties, planting time, harvesting time, different growth periods of crops, crop biomass, leaf area index, row density, ridge density, and sampling plant yield.
所述构建农田环境数据库的数据包括作物生长过程不同时间节点的土壤温度、土壤湿度、pH值、空气温湿度和病虫草害发生情况数据。The data for constructing the farmland environment database includes soil temperature, soil humidity, pH value, air temperature and humidity, and occurrence data of diseases, pests and weeds at different time nodes of the crop growth process.
所述构建农业气象数据库的数据包括太阳辐射、最低温度、最高温度、水汽压、平均风速、降水量以及气象卫星数据。The data for constructing agrometeorological database includes solar radiation, minimum temperature, maximum temperature, water vapor pressure, average wind speed, precipitation, and meteorological satellite data.
所述构建土壤基础地理数据库的数据包括土壤历史测土配方数据、土壤类型、坡度和海拔高度。The data for constructing the soil basic geographic database includes soil historical soil testing formula data, soil type, slope and altitude.
所述构建土地利用覆盖数据库包括土地覆被类型，如水田、旱田、有林地、灌木林、疏林地、草地等。The construction of the land use cover database includes land cover types, such as paddy field, dry land, forest land, shrub forest, sparse woodland, grassland, etc.
步骤4：在训练实验区域通过在若干个调查点现场采样调查土壤养分获得地面速效养分测试值，根据调查点位置形成土壤养分分布测试图。采样检测指标包含土壤养分(有机质、全氮、全磷、全钾)储量指标、养分有效状态(能被植物吸收利用的养分的含量及其比例，如有效磷/全磷、有效钾/全钾)，土壤生物数量、活性等。通过土壤化学原理进行测定土壤养分的指标，数值化表现地面速效养分对土壤肥力的影响。之后结合调查点的位置和地图信息就能转化为土壤养分分布测试图。Step 4: In the training experiment area, the ground available nutrient test values are obtained by field sampling and survey of soil nutrients at several survey points, and a soil nutrient distribution test map is formed according to the location of the survey points. Sampling and testing indicators include soil nutrient (organic matter, total nitrogen, total phosphorus, total potassium) storage indicators, nutrient availability status (the content and ratio of nutrients that can be absorbed and used by plants, such as available phosphorus/total phosphorus, available potassium/total potassium) ), the quantity and activity of soil organisms. The index of soil nutrients is determined through the principles of soil chemistry, and the effect of ground available nutrients on soil fertility is numerically expressed. Then combined with the location of the survey point and map information, it can be transformed into a soil nutrient distribution test map.
步骤5：土壤速效养分建模，卫星遥感数据库、无人机遥感数据库和农田环境基础数据库提供数据作为土壤速效养分的协变量信息，卷积神经网络通过梯度反向传播算法进行学习训练得到土壤速效养分反演模型。Step 5: Modeling soil quick-acting nutrients. Satellite remote sensing database, UAV remote sensing database and farmland environment basic database provide data as covariate information of soil quick-acting nutrients. Convolutional neural network learns and trains through gradient backpropagation algorithm to obtain soil quick-acting Nutrient inversion model.
S5.1：生成回归矩阵，根据调查点的位置，卫星遥感数据库、无人机遥感数据库和农田环境基础数据库提供相应数据作为土壤速效养分的协变量信息，覆盖叠加协变量信息生成回归矩阵。S5.1: Generate regression matrix. According to the location of the survey point, satellite remote sensing database, drone remote sensing database and farmland environment basic database provide corresponding data as the covariate information of soil available nutrients, covering the superimposed covariate information to generate the regression matrix.
S5.2：拟合空间预测模型，设计卷积神经网络，其同一平面上的神经元权值相等，将回归矩阵转化为向量输入，将土壤养分分布测试图作为学习目标，通过梯度反向传播算法进行学习训练，得到土壤速效养分反演模型。S5.2: Fit the spatial prediction model, design the convolutional neural network, the weights of the neurons on the same plane are equal, the regression matrix is converted into vector input, the soil nutrient distribution test chart is used as the learning target, and the gradient is backpropagated The algorithm performs learning and training, and obtains the soil available nutrient inversion model.
如图3所示的神经网络学习框架和过程如下：输入值通过三个可训练的滤波器和可加偏置进行卷积，卷积后在C1层产生三个特征映射图，然后特征映射图中每组若干特征再进行求和、加权值、加偏置，再通过Sigmoid函数得到三个S2层的土壤速效养分特征映射图；土壤速效养分特征映射图再经过滤波得到C3 层，C3层的层级结构通过产生S2层的方法产生S4层；最终，协变量信息形成向量输入到卷积神经网络，得到输出获得土壤速效养分反演模型。The neural network learning framework and process as shown in Figure 3 are as follows: the input value is convolved through three trainable filters and an addable bias. After convolution, three feature maps are generated in the C1 layer, and then the feature maps After each group of several features are summed, weighted, and biased, the three S2 layer soil available nutrient feature maps are obtained through the Sigmoid function; the soil available nutrient feature map is filtered to obtain the C3 layer and the C3 layer. The hierarchical structure generates the S4 layer by generating the S2 layer; finally, the covariate information formation vector is input to the convolutional neural network, and the output is obtained to obtain the soil available nutrient inversion model.
步骤6：获取需要进行预测的区域的协变量信息，经预处理后输入土壤速效养分反演模型，产出该区域土壤养分分布图。该步骤需要采集反演地块区域的多光谱/高光谱卫星图、该区域一年类内的历史气象数据、无人机对该区域作物长势的监测数据等，采集数据包含多种维度，种类多数量大，预测结果准确。Step 6: Obtain the covariate information of the area that needs to be predicted, and input the soil available nutrient inversion model after preprocessing to produce a soil nutrient distribution map in the area. This step needs to collect the multi-spectral/hyper-spectral satellite image of the inversion area, the historical meteorological data within one year of the area, the monitoring data of the crop growth of the area by the drone, etc. The collected data includes multiple dimensions and types. The number is large, and the prediction result is accurate.
考虑到卫星遥感对影像的时间间隔要求高，而速效养分的建模后预测的时效性要求较高，二者不相匹配，因此容易出现某一反演时间区间内缺乏及时采集的卫星遥感数据。此时，本方法将该时间区间前后时间上最接近的卫星遥感数据及其反演所得的相关数据作为该次反演的输入值，同时无人机遥感数据为该反演时间区间内采集的数据。这样结合时间区间内采集的无人机遥感数据等输入数据，通过模型进行计算能得到较为准确的输出结果，避免了速效养分的建模因卫星遥感数据采集间隔长而结果不准确的问题。Considering that satellite remote sensing has high requirements for the time interval of images, and the timeliness of prediction after modeling of quick-acting nutrients is relatively high, the two do not match, so it is easy to lack timely collection of satellite remote sensing data in a certain inversion time interval. . At this time, this method uses the closest satellite remote sensing data before and after the time interval and the related data obtained from the inversion as the input value of the inversion, and the UAV remote sensing data is collected in the inversion time interval. data. In this way, combined with input data such as UAV remote sensing data collected in the time interval, more accurate output results can be obtained by calculation through the model, which avoids the problem of inaccurate results of the modeling of quick-acting nutrients due to the long interval of satellite remote sensing data collection.
上面结合附图对本发明进行了示例性描述，显然本发明具体实现并不受上述方式的限制，只要采用了本发明的发明构思和技术方案进行的各种非实质性的改进，或未经改进将本发明构思和技术方案直接应用于其它场合的，均在本发明保护范围之内。The present invention is exemplarily described above with reference to the accompanying drawings. It is obvious that the specific implementation of the present invention is not limited by the above-mentioned manners, as long as various insubstantial improvements made by the inventive concept and technical solutions of the present invention are adopted, or no improvements are made. Application of the concept and technical solution of the present invention to other occasions directly falls within the protection scope of the present invention.

Claims (7)

  1. 一种基于深度神经网络的土壤速效养分反演方法，其特征在于：包括以下步骤：A method for retrieving soil available nutrients based on deep neural network, which is characterized in that it includes the following steps:
    步骤1、卫星遥感数据库的建立；Step 1. Establishment of satellite remote sensing database;
    步骤2：无人机遥感数据库的建立，所述无人机遥感数据库的数据采集间隔时间小于所述卫星遥感数据库；Step 2: Establish a remote sensing database for drones, the data collection interval of the remote sensing database for drones is shorter than that of the satellite remote sensing database;
    步骤3：农田环境基础数据库的建立；Step 3: Establish a basic database of farmland environment;
    步骤4：在训练实验区域通过在若干个调查点现场采样调查土壤养分获得地面速效养分测试值，根据调查点位置形成土壤养分分布测试图；Step 4: In the training experiment area, obtain the test value of ground available nutrients by sampling and surveying soil nutrients at several survey points on-site, and form a soil nutrient distribution test map according to the location of the survey points;
    步骤5：土壤速效养分建模，卫星遥感数据库、无人机遥感数据库和农田环境基础数据库提供数据作为土壤速效养分的协变量信息，卷积神经网络通过梯度反向传播算法进行学习训练得到土壤速效养分反演模型；Step 5: Modeling soil quick-acting nutrients. Satellite remote sensing database, UAV remote sensing database and farmland environment basic database provide data as covariate information of soil quick-acting nutrients. Convolutional neural network learns and trains through gradient backpropagation algorithm to obtain soil quick-acting Nutrient inversion model;
    步骤6：获取需要进行预测的区域的协变量信息，经预处理后输入土壤速效养分反演模型，产出该区域土壤养分分布图。Step 6: Obtain the covariate information of the area that needs to be predicted, and input the soil available nutrient inversion model after preprocessing to produce a soil nutrient distribution map in the area.
  2. 根据权利要求1所述的一种基于深度神经网络的土壤速效养分反演方法，其特征在于：所述步骤5包括：The method for retrieving soil quick-acting nutrients based on deep neural network according to claim 1, wherein said step 5 comprises:
    S5.1：生成回归矩阵，根据调查点的位置，卫星遥感数据库、无人机遥感数据库和农田环境基础数据库提供相应数据作为土壤速效养分的协变量信息，覆盖叠加协变量信息生成回归矩阵；S5.1: Generate regression matrix. According to the location of the survey point, satellite remote sensing database, drone remote sensing database and farmland environment basic database provide corresponding data as the covariate information of soil available nutrients, covering the superimposed covariate information to generate the regression matrix;
    S5.2：拟合空间预测模型，设计卷积神经网络，其同一平面上的神经元权值相等，将回归矩阵转化为向量输入，将土壤养分分布测试图作为学习目标，通过梯度反向传播算法进行学习训练，得到土壤速效养分反演模型。S5.2: Fit the spatial prediction model, design the convolutional neural network, the weights of the neurons on the same plane are equal, the regression matrix is converted into vector input, the soil nutrient distribution test chart is used as the learning target, and the gradient is backpropagated The algorithm performs learning and training, and obtains the soil available nutrient inversion model.
  3. 根据权利要求2所述的一种基于深度神经网络的土壤速效养分反演方法，其特征在于：所述步骤S5.2中卷积神经网络的学习过程为：输入值通过三个可训练的滤波器和可加偏置进行卷积，卷积后在C1层产生三个特征映射图，然后特征映射图中每组若干特征再进行求和、加权值、加偏置，再通过Sigmoid函数得到三个S2层的土壤速效养分特征映射图；土壤速效养分特征映射图再经过滤波得到C3层，C3层的层级结构通过产生S2层的方法产生S4层；最终，协变量信息形成向量输入到卷积神经网络，得到输出获得土壤速效养分反演模型。The method for retrieving soil quick-acting nutrients based on a deep neural network according to claim 2, wherein the learning process of the convolutional neural network in step S5.2 is: the input value passes through three trainable filters After convolution, three feature maps are generated in the C1 layer, and then each group of several features in the feature map are summed, weighted, and biased, and then the three are obtained through the Sigmoid function. An S2 layer of soil available nutrient feature maps; the soil available nutrient feature maps are filtered to obtain the C3 layer. The C3 layer's hierarchical structure generates the S4 layer by generating the S2 layer; finally, the covariate information formation vector is input to the convolution Neural network, get the output to obtain the soil available nutrient inversion model.
  4. 根据权利要求1-3中任一所述的一种基于深度神经网络的土壤速效养分 反演方法，其特征在于：所述步骤1中卫星遥感数据库中包括卫星遥感数据，还包括卫星遥感数据反演所得的田块边界、作物种植类型、作物长势及单产水平数据；所述步骤2中无人机遥感数据库包括周期性获取的无人机遥感数据及其反演所得的田块边界、作物种植类型、作物长势及单产水平数据。The method for retrieving soil quick-acting nutrients based on deep neural network according to any one of claims 1-3, characterized in that: the satellite remote sensing database in step 1 includes satellite remote sensing data, and also includes satellite remote sensing data retrieval. The field boundary, crop planting type, crop growth and yield level data obtained from the simulation; the UAV remote sensing database in the step 2 includes the periodically acquired UAV remote sensing data and the field boundary and crop planting obtained from the inversion. Type, crop condition and yield data.
  5. 根据权利要求4所述的一种基于深度神经网络的土壤速效养分反演方法，其特征在于：所述步骤3中所述农田环境基础数据库包括农作物物候数据库、农田环境数据库、农业气象数据库、土壤基础地理数据库和土地利用覆盖数据库，所述农田环境数据库和所述农业气象数据库的数据采集间隔时间小于所述卫星遥感数据库。The method for retrieving available soil nutrients based on a deep neural network according to claim 4, wherein the basic farmland environment database in step 3 includes a crop phenology database, a farmland environment database, agrometeorological database, and soil A basic geographic database and a land use coverage database, and the data collection interval of the farmland environment database and the agrometeorological database is shorter than that of the satellite remote sensing database.
  6. 根据权利要求5所述的一种基于深度神经网络的土壤速效养分反演方法，其特征在于：The method for retrieving soil available nutrients based on deep neural network according to claim 5, characterized in that:
    所述构建农作物物候数据库的数据包括主要作物品种、种植时间、收割时间、作物不同生育期时间、作物生物量、叶面积指数、行密度、垄密度和抽样株单产等信息；The data for constructing the crop phenology database includes information such as main crop varieties, planting time, harvesting time, different growth periods of crops, crop biomass, leaf area index, row density, ridge density, and sampling plant yield;
    所述构建农田环境数据库的数据包括作物生长过程不同时间节点的土壤温度、土壤湿度、pH值、空气温湿度和病虫草害发生情况数据；The data for constructing the farmland environment database includes soil temperature, soil humidity, pH value, air temperature and humidity, and occurrence data of diseases, pests and weeds at different time nodes of the crop growth process;
    所述构建农业气象数据库的数据包括太阳辐射、最低温度、最高温度、水汽压、平均风速、降水量以及气象卫星数据；The data for constructing the agrometeorological database includes solar radiation, minimum temperature, maximum temperature, water vapor pressure, average wind speed, precipitation, and meteorological satellite data;
    所述构建土壤基础地理数据库的数据包括土壤历史测土配方数据、土壤类型、坡度和海拔高度；The data for constructing the soil basic geographic database includes soil historical soil testing formula data, soil type, slope and altitude;
    所述构建土地利用覆盖数据库包括土地覆被类型。The construction of a land use cover database includes land cover types.
  7. 根据权利要求6所述的一种基于深度神经网络的土壤速效养分反演方法，其特征在于：所述步骤6中，当卫星遥感数据库缺乏属于某一反演时间区间内的数据时，将该时间区间前后最接近的卫星遥感数据及其反演所得的相关数据作为该次反演的输入值，同时无人机遥感数据为该反演时间区间内采集的数据。The method for retrieving soil quick-acting nutrients based on a deep neural network according to claim 6, characterized in that: in step 6, when the satellite remote sensing database lacks data belonging to a certain retrieval time interval, the The closest satellite remote sensing data before and after the time interval and the related data obtained from the inversion are used as the input value of this inversion, and the UAV remote sensing data is the data collected in the inversion time interval.
PCT/CN2020/090396 2020-05-15 2020-05-15 Soil available nutrient inversion method based on deep neural network WO2021226976A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202080000932.XA CN111727443B (en) 2020-05-15 2020-05-15 Soil available nutrient inversion method based on deep neural network
PCT/CN2020/090396 WO2021226976A1 (en) 2020-05-15 2020-05-15 Soil available nutrient inversion method based on deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/090396 WO2021226976A1 (en) 2020-05-15 2020-05-15 Soil available nutrient inversion method based on deep neural network

Publications (1)

Publication Number Publication Date
WO2021226976A1 true WO2021226976A1 (en) 2021-11-18

Family

ID=72574204

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/090396 WO2021226976A1 (en) 2020-05-15 2020-05-15 Soil available nutrient inversion method based on deep neural network

Country Status (2)

Country Link
CN (1) CN111727443B (en)
WO (1) WO2021226976A1 (en)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114120147A (en) * 2021-11-30 2022-03-01 上海交通大学 Ecological environment evaluation system for railway tunnel water inrush flow path along line region
CN114140591A (en) * 2021-11-29 2022-03-04 中国科学院南京土壤研究所 Soil organic matter remote sensing mapping method combining machine learning and land statistics
CN114359725A (en) * 2021-12-30 2022-04-15 黑龙江省农业科学院农业遥感与信息研究所 Crop growth remote sensing monitoring system and method based on crop model and assimilation technology
CN114414491A (en) * 2021-12-30 2022-04-29 青海省草原总站 Dynamic monitoring and analyzing system for grassland ecology
CN114460013A (en) * 2022-01-28 2022-05-10 自然资源部第一海洋研究所 Coastal wetland vegetation ground biomass GAN model self-learning remote sensing inversion method
CN114461971A (en) * 2022-01-13 2022-05-10 桂林理工大学 Surface soil water content inversion method fusing soil physical properties and remote sensing data
CN114529838A (en) * 2022-04-24 2022-05-24 江西农业大学 Soil nitrogen content inversion model construction method and system based on convolutional neural network
CN115100521A (en) * 2022-06-23 2022-09-23 中科卫星应用德清研究院 Salinization hierarchical inversion method and system based on polarization radar remote sensing data
CN115116558A (en) * 2022-07-06 2022-09-27 农业农村部环境保护科研监测所 Dynamic prediction method for heavy metal content
CN115453868A (en) * 2022-08-31 2022-12-09 中国农业大学 Full-growth-period light intensity regulation and control method based on tomato photoresponse difference characteristics
CN115661671A (en) * 2022-12-22 2023-01-31 中国科学院地理科学与资源研究所 Soil moisture microwave product downscaling method based on feature space
CN116596326A (en) * 2023-04-11 2023-08-15 常州双炬智能科技有限公司 Urban environment detection and comprehensive evaluation method based on remote sensing data
WO2023168781A1 (en) * 2022-03-10 2023-09-14 浙江大学 Soil cadmium risk prediction method based on spatial-temporal interaction relationship
CN117132883A (en) * 2023-05-08 2023-11-28 江苏商贸职业学院 GIS-based intelligent agricultural disaster discrimination method and system
CN117153291A (en) * 2023-10-31 2023-12-01 水利部交通运输部国家能源局南京水利科学研究院 Method and system for calculating carbon exchange value of paddy field in irrigation area
CN117151354A (en) * 2023-11-01 2023-12-01 北京建工环境修复股份有限公司 Farmland restoration and improvement management method and system
CN115453868B (en) * 2022-08-31 2024-04-12 中国农业大学 Full-growth-period light intensity regulation and control method based on tomato light response difference characteristics

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112711900A (en) * 2020-12-02 2021-04-27 济南浪潮高新科技投资发展有限公司 Crop digital twin modeling method
CN112733780A (en) * 2021-01-19 2021-04-30 哈尔滨航天恒星数据系统科技有限公司 System and method for constructing soil available phosphorus analysis model based on satellite images
CN114051810A (en) * 2021-11-10 2022-02-18 上海联适导航技术股份有限公司 Unmanned high-precision agricultural machine fertilization control method and system and intelligent agricultural machine
CN116451886B (en) * 2023-06-20 2023-10-10 创辉达设计股份有限公司 High-standard farmland water balance calculation method
CN117035471B (en) * 2023-10-10 2023-12-22 中科领讯(北京)科技有限公司 Fertilizing decision method and system based on multi-vegetation index image

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109001127A (en) * 2018-06-01 2018-12-14 华南农业大学 A kind of heavy metal content in soil space predicting method
CN109142679A (en) * 2018-08-13 2019-01-04 中国科学院东北地理与农业生态研究所 The space predicting method of forest soil nutrient based on artificial neural network Kriging regression
CN110110595A (en) * 2019-03-28 2019-08-09 国智恒北斗好年景农业科技有限公司 A kind of farmland portrait and medicine hypertrophy data analysing method based on satellite remote-sensing image
US10417556B1 (en) * 2017-12-07 2019-09-17 HatchB Labs, Inc. Simulation-based controls optimization using time series data forecast
CN111047099A (en) * 2019-12-16 2020-04-21 杭州鲁尔物联科技有限公司 Regional torrential flood risk prediction method and system

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10408810B2 (en) * 2015-03-02 2019-09-10 Unibest International, Llc Methods and apparatus for determining fertilizer/treatment requirements and/or predicting plant growth response
CN105181607B (en) * 2015-09-07 2018-04-10 中国科学院遥感与数字地球研究所 Readily available nutrient of soil inversion method and device
WO2018049289A1 (en) * 2016-09-09 2018-03-15 Cibo Technologies, Inc. Systems for adjusting agronomic inputs using remote sensing, and related apparatus and methods
CN109661979B (en) * 2017-10-13 2021-04-06 中国农业机械化科学研究院 Informatization planting method based on accurate control of crop position
US11596119B2 (en) * 2018-08-13 2023-03-07 Climate Llc Digital nutrient models using spatially distributed values unique to an agronomic field
CN109142359A (en) * 2018-08-23 2019-01-04 中国水利水电科学研究院 A kind of crop growth monitoring method based on time series remotely-sensed data
CN109269475A (en) * 2018-09-29 2019-01-25 成都信息工程大学 A kind of vacant lot one plant automatic checkout system and method
CN109374860A (en) * 2018-11-13 2019-02-22 西北大学 A kind of soil nutrient prediction and integrated evaluating method based on machine learning algorithm
CN109757175B (en) * 2019-03-18 2020-10-30 中国农业科学院农业资源与农业区划研究所 Unmanned aerial vehicle monitoring-based corn water and fertilizer integrated variable fertilization method
CN110115142A (en) * 2019-04-25 2019-08-13 中科赛诺（北京）科技有限公司 A kind of farmland variable fertilization method based on remotely-sensed data
CN110213376B (en) * 2019-06-05 2021-03-09 黑龙江北大荒农业股份有限公司 Information processing system and method for insect pest prevention
CN110530901B (en) * 2019-09-11 2020-09-01 中国科学院南京土壤研究所 Medium and small scale soil water monitoring system and method integrating cosmic ray neutron method and unmanned aerial vehicle remote sensing
CN111028096A (en) * 2019-12-23 2020-04-17 内蒙古自治区生物技术研究院 System and method for integrating space, air and ground data

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10417556B1 (en) * 2017-12-07 2019-09-17 HatchB Labs, Inc. Simulation-based controls optimization using time series data forecast
CN109001127A (en) * 2018-06-01 2018-12-14 华南农业大学 A kind of heavy metal content in soil space predicting method
CN109142679A (en) * 2018-08-13 2019-01-04 中国科学院东北地理与农业生态研究所 The space predicting method of forest soil nutrient based on artificial neural network Kriging regression
CN110110595A (en) * 2019-03-28 2019-08-09 国智恒北斗好年景农业科技有限公司 A kind of farmland portrait and medicine hypertrophy data analysing method based on satellite remote-sensing image
CN111047099A (en) * 2019-12-16 2020-04-21 杭州鲁尔物联科技有限公司 Regional torrential flood risk prediction method and system

Cited By (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114140591A (en) * 2021-11-29 2022-03-04 中国科学院南京土壤研究所 Soil organic matter remote sensing mapping method combining machine learning and land statistics
CN114120147B (en) * 2021-11-30 2022-12-09 上海交通大学 Ecological environment evaluation system for railway tunnel water inrush flow path along line region
CN114120147A (en) * 2021-11-30 2022-03-01 上海交通大学 Ecological environment evaluation system for railway tunnel water inrush flow path along line region
CN114359725A (en) * 2021-12-30 2022-04-15 黑龙江省农业科学院农业遥感与信息研究所 Crop growth remote sensing monitoring system and method based on crop model and assimilation technology
CN114414491A (en) * 2021-12-30 2022-04-29 青海省草原总站 Dynamic monitoring and analyzing system for grassland ecology
CN114359725B (en) * 2021-12-30 2024-03-26 黑龙江省农业科学院农业遥感与信息研究所 Crop growth condition remote sensing monitoring system and method based on crop model and assimilation technology
CN114414491B (en) * 2021-12-30 2023-10-27 青海省草原总站 Grass ecology dynamic monitoring and analysis system
CN114461971A (en) * 2022-01-13 2022-05-10 桂林理工大学 Surface soil water content inversion method fusing soil physical properties and remote sensing data
CN114461971B (en) * 2022-01-13 2024-04-16 桂林理工大学 Earth surface soil water content inversion method integrating soil physical properties and remote sensing data
CN114460013A (en) * 2022-01-28 2022-05-10 自然资源部第一海洋研究所 Coastal wetland vegetation ground biomass GAN model self-learning remote sensing inversion method
CN114460013B (en) * 2022-01-28 2023-10-17 自然资源部第一海洋研究所 Coastal wetland vegetation overground biomass GAN model self-learning remote sensing inversion method
WO2023168781A1 (en) * 2022-03-10 2023-09-14 浙江大学 Soil cadmium risk prediction method based on spatial-temporal interaction relationship
CN114529838A (en) * 2022-04-24 2022-05-24 江西农业大学 Soil nitrogen content inversion model construction method and system based on convolutional neural network
CN115100521A (en) * 2022-06-23 2022-09-23 中科卫星应用德清研究院 Salinization hierarchical inversion method and system based on polarization radar remote sensing data
CN115116558A (en) * 2022-07-06 2022-09-27 农业农村部环境保护科研监测所 Dynamic prediction method for heavy metal content
CN115116558B (en) * 2022-07-06 2024-03-29 农业农村部环境保护科研监测所 Dynamic prediction method for heavy metal content
CN115453868A (en) * 2022-08-31 2022-12-09 中国农业大学 Full-growth-period light intensity regulation and control method based on tomato photoresponse difference characteristics
CN115453868B (en) * 2022-08-31 2024-04-12 中国农业大学 Full-growth-period light intensity regulation and control method based on tomato light response difference characteristics
CN115661671B (en) * 2022-12-22 2023-03-21 中国科学院地理科学与资源研究所 Soil moisture microwave product downscaling method based on feature space
CN115661671A (en) * 2022-12-22 2023-01-31 中国科学院地理科学与资源研究所 Soil moisture microwave product downscaling method based on feature space
CN116596326A (en) * 2023-04-11 2023-08-15 常州双炬智能科技有限公司 Urban environment detection and comprehensive evaluation method based on remote sensing data
CN117132883B (en) * 2023-05-08 2024-03-19 江苏商贸职业学院 GIS-based intelligent agricultural disaster discrimination method and system
CN117132883A (en) * 2023-05-08 2023-11-28 江苏商贸职业学院 GIS-based intelligent agricultural disaster discrimination method and system
CN117153291B (en) * 2023-10-31 2024-01-02 水利部交通运输部国家能源局南京水利科学研究院 Method and system for calculating carbon exchange value of paddy field in irrigation area
CN117153291A (en) * 2023-10-31 2023-12-01 水利部交通运输部国家能源局南京水利科学研究院 Method and system for calculating carbon exchange value of paddy field in irrigation area
CN117151354A (en) * 2023-11-01 2023-12-01 北京建工环境修复股份有限公司 Farmland restoration and improvement management method and system
CN117151354B (en) * 2023-11-01 2024-01-02 北京建工环境修复股份有限公司 Farmland restoration and improvement management method and system

Also Published As

Publication number Publication date
CN111727443B (en) 2023-10-31
CN111727443A (en) 2020-09-29

Similar Documents

Publication Publication Date Title
WO2021226976A1 (en) Soil available nutrient inversion method based on deep neural network
Sun et al. Multilevel deep learning network for county-level corn yield estimation in the us corn belt
CA3129617C (en) Yield forecasting using crop specific features and growth stages
CN110751094B (en) Crop yield estimation method based on GEE comprehensive remote sensing image and deep learning method
CN110909933B (en) Agricultural drought rapid diagnosis and evaluation method coupling crop model and machine learning language
WO2019224347A1 (en) Training a machine learning algorithm and predicting a value for a weather data variable, especially at a field or sub-field level
Shi et al. Impacts of cultivated land reclamation on the climate and grain production in Northeast China in the future 30 years
CN113570273B (en) Spatialization method and system for irrigation farmland statistical data
Lou et al. An effective method for canopy chlorophyll content estimation of marsh vegetation based on multiscale remote sensing data
Chaudhari et al. Spatial wheat yield prediction using crop simulation model, GIS, remote sensing and ground observed data
Zhang et al. Integrating a novel irrigation approximation method with a process-based remote sensing model to estimate multi-years' winter wheat yield over the North China Plain
Ji et al. Integrating the temperature vegetation dryness index and meteorology parameters to dynamically predict crop yield with fixed date intervals using an integral regression model
Huang et al. Digital image analysis technologies for decision support systems in agricultural
LU501248B1 (en) Inversion method for readily available nutrients of soil based on deep neural network
CN115527108A (en) Method for rapidly identifying water and soil loss artificial disturbance plots based on multi-temporal Sentinel-2
CN111724073B (en) Modularized agricultural machine emission calculation and pollution identification management and control system
CN115688955A (en) Vegetation growth prediction method, vegetation growth prediction device and storage medium
Gulyaev et al. On the development of a system for digital remote monitoring of agricultural land
Jiang et al. Application and evaluation of an improved LSTM model in the soil moisture prediction of southeast chinese tobacco-producing areas
Patil et al. Role of remote sensing in precision agriculture
Maitra et al. Application of Remote Sensing & Geographic Information Systems for Agriculture-A Review
Liu et al. Research on apple orchard classification and yield estimation model based on GF-1 and Sentinel-2
Skocir et al. Analysis of Open Access Data Sources for Application in Precision Agriculture
AU2021107439A4 (en) A Novel method- Crop Health Monitoring System Using IoT and Machine Learning
AU2021100538A4 (en) Crop Health Monitoring System Using IoT and Machine Learning

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20935454

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20935454

Country of ref document: EP

Kind code of ref document: A1