CN117556956A - Rice unit area yield estimation method and device - Google Patents
Rice unit area yield estimation method and device Download PDFInfo
- Publication number
- CN117556956A CN117556956A CN202311583027.3A CN202311583027A CN117556956A CN 117556956 A CN117556956 A CN 117556956A CN 202311583027 A CN202311583027 A CN 202311583027A CN 117556956 A CN117556956 A CN 117556956A
- Authority
- CN
- China
- Prior art keywords
- scale
- dimensional
- rice
- layer
- time sequence
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 235000007164 Oryza sativa Nutrition 0.000 title claims abstract description 81
- 235000009566 rice Nutrition 0.000 title claims abstract description 81
- 238000000034 method Methods 0.000 title claims abstract description 41
- 240000007594 Oryza sativa Species 0.000 title 1
- 241000209094 Oryza Species 0.000 claims abstract description 81
- 230000015654 memory Effects 0.000 claims abstract description 37
- 238000013519 translation Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 230000007787 long-term memory Effects 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 3
- 238000013136 deep learning model Methods 0.000 abstract description 2
- 238000004364 calculation method Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 5
- 230000010287 polarization Effects 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 241000283898 Ovis Species 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 239000002028 Biomass Substances 0.000 description 2
- 235000013339 cereals Nutrition 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 235000003642 hunger Nutrition 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 239000002689 soil Substances 0.000 description 2
- 241000196324 Embryophyta Species 0.000 description 1
- 241001464837 Viridiplantae Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 229930002875 chlorophyll Natural products 0.000 description 1
- 235000019804 chlorophyll Nutrition 0.000 description 1
- ATNHDLDRLWWWCB-AENOIHSZSA-M chlorophyll a Chemical compound C1([C@@H](C(=O)OC)C(=O)C2=C3C)=C2N2C3=CC(C(CC)=C3C)=[N+]4C3=CC3=C(C=C)C(C)=C5N3[Mg-2]42[N+]2=C1[C@@H](CCC(=O)OC\C=C(/C)CCC[C@H](C)CCC[C@H](C)CCCC(C)C)[C@H](C)C2=C5 ATNHDLDRLWWWCB-AENOIHSZSA-M 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000005388 cross polarization Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000009355 double cropping Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 238000009396 hybridization Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 102000039446 nucleic acids Human genes 0.000 description 1
- 108020004707 nucleic acids Proteins 0.000 description 1
- 150000007523 nucleic acids Chemical class 0.000 description 1
- 235000003715 nutritional status Nutrition 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- Molecular Biology (AREA)
- Economics (AREA)
- Computing Systems (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Mining & Mineral Resources (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Algebra (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- Primary Health Care (AREA)
- Marine Sciences & Fisheries (AREA)
- Animal Husbandry (AREA)
- Agronomy & Crop Science (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
Abstract
The invention discloses a rice yield per unit area estimation method and a device, wherein the method comprises the following steps: the spatial resolution and the time resolution of SAR time sequence data are unified; calculating radar vegetation indexes by taking county level as a scale; and constructing a multi-scale one-dimensional convolution-long-short-time memory network model for estimating the yield value of the rice in unit area. The invention uses a plurality of radar vegetation indexes reflecting the growth state of crops; the multi-scale one-dimensional convolution-long-short-term memory network deep learning model is provided, the high-dimensional characteristics of each time step and the characteristics among the high-dimensional channels can be combined, the long-short-term memory layer keeps the high-dimensional characteristics of the model output in each time step, the contribution of the high-dimensional characteristics of each time step to the unit area yield estimation result is synthesized, the translation invariant characteristics among the high-dimensional characteristic channels are extracted through multi-scale one-dimensional convolution, the multi-scale characteristics of the high-dimensional characteristic channels of each time step are fused, and the estimation accuracy of the unit area yield of rice is effectively improved.
Description
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to a rice unit area yield estimation method and device.
Background
Target 2 in the united nations sustainable development target (Sustainable Development Goals, SDGs): the core of zero hunger is to eliminate hunger, realize grain safety, improve nutrient status and promote sustainable agriculture. Rice is the staple food for more than half of the world population. Reliable rice yield information is critical to global grain safety, agricultural management and import and export strategic planning.
Remote sensing estimates have the advantage of accurate and rapid yield measurements compared to traditional field surveys of yield. The optical vegetation index (Optical vegetation indices, OVIs) directly reflects the structure (e.g., chlorophyll) and status (e.g., water content) of the green plant leaves, as well as the spectral characteristics of the vegetation canopy. OVIs dominate the field of crop yield estimation. Rice mainly grows in tropical or subtropical areas with cloudiness and raininess. Optical remote sensing is susceptible to cloud and rain weather and may not be effectively monitored during critical growth periods of rice. Synthetic Aperture Radar (SAR) has unique all-day and all-weather monitoring capabilities, and has been increasingly applied to crop yield predictions.
The main data source of remote sensing rice yield estimation is optical data, and research on rice yield estimation by using SAR data is limited. There are three main classes of rice yield estimation models based on multi-phase SAR data, including process-based crop models, semi-empirical models, and empirical models.
The process-based crop model can quantitatively reproduce the growth and development process of rice. The vegetation biochemical parameters obtained by SAR data are generally used as assimilation observation, and initial input parameters of a model are adjusted to optimize a crop model, simulate a rice growth process and further estimate the yield of rice. Leaf Area Index (LAI) is one of the most commonly used assimilation parameters. ORYZA rice models are widely used to simulate rice growth. Process-based crop models require the input of a variety of parameters (including crop management, weather conditions, soil and various environmental factors) that are continuously observed over a long period of time.
The semi-empirical model can invert the biological parameters of rice. And the yield prediction of the rice is realized by establishing a regression model of biological parameters and yield. Representative semi-empirical models include rice canopy scattering models and traditional or modified water cloud models. The semi-empirical model requires fewer parameters than the crop model, but still requires measurement of a certain number of crop structural parameters.
Compared with the first two models, the principle of the empirical model is simple, and the complex rice yield forming mechanism is not involved. The rice yield estimation is realized by establishing a regression model between SAR data and rice yield data. Empirical models are widely used because they are independent of crop growth parameters and environmental parameters. The regression models include conventional regression models and machine learning regression models.
The conventional regression model in common use is based on a multiple linear regression model. There are also studies using linear regression models, exponential regression and gaussian kernel regression. Most of the currently available statistical-based models are poorly generalizable and have large limitations in dealing with nonlinear relationships between independent and dependent variables.
The machine learning model can process complex relations between independent variables and dependent variables, and improves generalization capability of the regression model. Common machine learning regression models are random forest regression models and neural networks.
The estimation method based on the empirical model mainly utilizes the time sequence backscattering coefficient of the crop growth and development period to estimate the rice yield. In addition, the simple differential indices of time sequences and the interference coherence values of time sequences obtained by mathematical operations of various polarization channels are also related to the yield of the regression model.
In recent years, studies have shown that there is a theoretical relationship between radar vegetation index (radar vegetation indices, RVIs) and OVIs. In particular, RVIs have a relationship with normalized vegetation index (NDVI) and LAI. And RVIs can be inverted to obtain reliable crop biophysical parameters (plant area index and vegetation water content), and crop growth in different climatic periods is monitored. In summary, the performance of RVIs in rice yield estimation is worth exploring. However, studies introducing RVIs into crop yield estimates are still in the preliminary stage. The number of documents for estimating crop yield using RVIs is limited. Thus, the full potential of RVIs in yield estimation remains to be further demonstrated.
In SAR data rice yield estimation research based on an empirical model, a machine learning regression model cannot learn time sequence characteristics, so that yield prediction accuracy is limited. A long-short-time memory (LSTM) network is a special recurrent neural network, good at feature extraction of time-series data. Due to the cyclic structure of LSTM and the gating mechanism that regulates the flow of information into and out of the cell, it can remember information for longer periods of time and can capture complex nonlinear relationships. In the field of crop yield estimation based on vegetation indexes and LSTM, researchers usually extract the output of the last moment of the LSTM model, and obtain a yield regression value through dimension reduction of the full-link layer. However, this approach has certain drawbacks. Firstly, the optimal regression result can be obtained after learning the data of all time steps by default. The contribution of the output of the model at each time step to the final regression result is not considered in the final output of the yield prediction value. And the yield value may be most relevant to a certain time step in the time series data. The presence of too much data over this time step may interfere with the regression results. Secondly, only the extraction of time sequence features is concerned, and translational invariance among high-dimensional feature channels is ignored, namely the relation between the features among the high-dimensional feature channels and the yield is ignored.
Disclosure of Invention
In order to solve the problems in the prior art, the embodiment of the invention provides a rice yield per unit area estimation method and device, which can solve the problems in the prior art.
The first aspect of the embodiment of the invention provides a rice yield per unit area estimation method, which comprises the following steps:
step 1, unifying the spatial resolution and the time resolution of SAR time sequence data;
step 2, calculating various time sequence radar vegetation indexes by taking county level as a scale;
and 3, constructing a multi-scale one-dimensional convolution-long-short-time memory network model for estimating the yield of rice in unit area.
Further, step 1 includes: unifying the spatial resolution of SAR time sequence data; and unifying the time resolution of the SAR time sequence data.
Further, step 2 includes:
step 2.1, calculating various time sequence radar vegetation indexes;
and 2.2. Converting various time sequence radar vegetation indexes into the index taking county level as a scale.
Further, in the multi-scale one-dimensional convolution-long-short-time memory network model in step 3:
the first layer is an input layer and is used for stacking various time sequence radar vegetation indexes in a channel dimension so as to meet the input format requirement of a long-short-term memory network layer of the second layer;
the second layer is a long-short-term memory network layer and is used for extracting time sequence high-dimensional characteristics of the time sequence radar vegetation index;
the third layer is a multi-scale one-dimensional convolution layer, takes the time sequence high-dimensional characteristics output by the long-short-time memory network layer at all time steps as input, and extracts translation invariance among all time step high-dimensional characteristic channels by utilizing a multi-scale one-dimensional convolution kernel;
the fourth layer is a full-connection layer and is used for taking the result of the multi-scale one-dimensional convolution layer as input, and obtaining the estimated value of the yield of the rice in unit area through dimension reduction of the full-connection layer.
Further, the long-short-time memory network layer includes:
and the two LSTM units are used for extracting the high-dimensional characteristics of the vegetation index of the time sequence radar.
Further, the multi-scale one-dimensional convolution layer includes:
three one-dimensional convolution kernels with different scales are convolved in the high-dimensional characteristic channel dimension; and splicing the convolution results of the three convolution kernels.
Further, the fully-connected layer comprises:
reducing the dimension of the result of the multi-scale one-dimensional convolution layer through the full connection layer;
and obtaining the estimated value of the yield of the rice in unit area by using a Tanh activation function.
A second aspect of an embodiment of the present invention provides an apparatus, comprising: comprising the following steps:
the SAR time sequence data processing module is used for unifying the spatial resolution and the time resolution of the SAR time sequence data;
the time sequence radar vegetation index processing module is used for calculating the time sequence radar vegetation index by taking county level as a scale;
and the rice unit area yield estimation processing module is used for constructing a multi-scale one-dimensional convolution-long-short-time memory network model and estimating the rice unit area yield.
A third aspect of an embodiment of the present invention provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the steps of the above-described rice unit area yield estimation method using multiple radar vegetation indexes and a multi-scale one-dimensional convolution-long-short-term memory network.
The invention has the following beneficial technical effects:
the invention realizes a novel rice yield estimation method in unit area, uses a plurality of radar vegetation indexes capable of reflecting the growth state of crops for rice yield estimation in unit area, and verifies the yield estimation potential of the radar vegetation indexes; the novel multi-scale one-dimensional convolution-long-short-time memory network deep learning model is provided, the high-dimensional characteristics of each time step and the characteristics among high-dimensional channels can be combined, the multi-scale one-dimensional convolution layer keeps the high-dimensional characteristics of the model output in each time step, and the contribution of the high-dimensional characteristics of each time step to the unit area yield estimation result is synthesized; and the translation invariant features among the high-dimensional feature channels are extracted by utilizing multi-scale one-dimensional convolution, and the multi-scale features of each time-step high-dimensional feature channel are fused, so that the higher-level and more abstract expression of the time sequence radar vegetation index data is realized, and the estimation precision of the estimated value of the yield of the unit area of the rice is improved.
Drawings
FIG. 1 is a flowchart of a rice yield per unit area estimation method according to an embodiment of the present invention;
fig. 2 is a block diagram of a multi-scale one-dimensional convolution-long-short-time memory network according to an embodiment of the present invention.
Detailed Description
The present invention will be described with respect to particular embodiments and with reference to certain drawings but the invention is not limited thereto but only by the claims. Any reference signs in the claims shall not be construed as limiting the scope. The drawings described are only schematic and are non-limiting. In the drawings, the size of some of the elements may be exaggerated and not drawn on scale for illustrative purposes. The term "comprising" when used in the specification and claims does not exclude other elements or steps. When an indefinite or definite article such as "a" or "an" is used when referring to a singular noun, this includes a plural of that noun unless something else is specifically stated. Furthermore, the terms first, second, third and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the invention described herein are capable of operation in other sequences than described or illustrated herein.
Unless defined otherwise herein, technical and scientific terms and phrases used herein should have the meaning commonly understood by one of ordinary skill in the art. In general, the molecular and cellular biology, genetics, and protein and nucleic acid chemistry and hybridization techniques described herein and the terms used in such techniques are all techniques and terms that are well known and commonly used in the art. Unless otherwise indicated, it is generally done according to conventional methods well known in the art and described in various general and more specific references cited and discussed in this specification.
So that the manner in which the features and objects of the present invention can be understood in more detail, a more particular description of the invention, briefly summarized below, may be had by reference to the embodiments thereof which are illustrated in the appended drawings.
FIG. 1 is a flowchart of a rice yield per unit area estimation method according to an embodiment of the present invention. The embodiment of the invention uses a Sentinel-1 IW mode VH/VV polarized GRD data of 2017-2021 in a 2268 scene, a rice area map (https:// zenodo. Org/record/5555721) and county-grade annual rice early-late rice unit area yield statistical data as a research area for explanation.
As shown in fig. 1, a first aspect of the embodiment of the present invention provides a method for estimating yield per unit area of rice, including the steps of:
step S1: the spatial resolution and the temporal resolution of the SAR time series data are unified.
Specifically, the method comprises the steps of unifying the spatial resolution of SAR data and unifying the time resolution of SAR data.
In the step, the Sentinel-1GRD time sequence SAR data is required to be processed, so that the functions of unifying the spatial resolution and the time resolution of the Sentinel-1GRD time sequence SAR are realized.
To ensure subsequent processing, the spatial resolution of the Sentinel-1GRD data needs to be kept consistent with that of the rice area map. To ensure that the time series data of different frames covering the investigation region have the same time series length, the time resolution of the Sentinel-1GRD data needs to be unified.
And unifying the spatial resolution of the Sentinel-1GRD data to 500 meters for subsequent processing.
And unifying the time resolution of the Sentinel-1GRD data so as to carry out subsequent processing. Specifically, each month is divided into two half months: the mean values of the time series SAR data in the first half month (the first 15 days) and the second half month (the remaining days) were calculated, respectively.
GRD (Ground Range Detected) data is data which is processed by multiple views and projected to ground distance by adopting WGS84 ellipsoids. The GRD image is stored as a real-valued array, the pixel information representing the amplitude information of the monitored area, and the phase information being lost.
Step S2: and calculating a time sequence radar vegetation index by taking the county level as a scale.
Specifically, the method comprises the steps of calculating a time sequence radar vegetation index, and converting the time sequence radar vegetation index into a county scale.
The calculation of the time sequence radar vegetation index refers to calculating 3 representative radar vegetation indexes by using Sentinel-1 time sequence SAR data. Radar vegetation index RVI (Radar Vegetation Index), sentinel-1 radar vegetation index RVI4S1 (Radar Vegetation Index for Sentinel-1), and polarized radar vegetation index PRVI (Polarimetric Radar Vegetation Index), respectively. The three time sequence radar vegetation indexes have a simple calculation formula, and are convenient to calculate by utilizing Sentinel-1 dual-polarized data.
The radar vegetation index RVI is used to measure the randomness of the scattering in the microwave signal. For a smooth bare surface, its value is low, and as the vegetation grows, the value increases until the end of the crop growth cycle, it is affected by the moisture content of the vegetation and is sensitive to biomass. The calculation formula is shown in equation (1).
The Sentinel-1 radar vegetation index RVI4S1 is a radar vegetation index RVI index created from dual polarized radar images of Sentinel-1 satellites for monitoring crop growth. The Sentinel-1 radar vegetation index RVI4S1 index of bare soil or pure primary target is very low. In contrast, a fully developed crop canopy has a higher value. The calculation formula is shown in equation (2).
Based on a volume scattering theoretical model and a semi-empirical model, a polarized radar vegetation index PRVI using polarization degree and cross polarization backscattering coefficient is provided. The correlation of the polarized radar vegetation index PRVI with biomass is higher than other radar polarization parameters, in particular the existing radar vegetation index RVI. The calculation formula is shown in equation (3).
Where VH represents the backscatter coefficients of the vertically transmitted and horizontally received polarization data, and VV represents the backscatter coefficients of the vertically transmitted and vertically received polarization data.
The county scale radar vegetation index conversion refers to masking the radar vegetation index by using the rice planting area, and removing non-rice pixels. The county average of the rice pixels is then calculated using the county administrative division provided by the county vector file.
Step S3: and constructing a multi-scale one-dimensional convolution-long-short-time memory network model for estimating the yield of rice in unit area.
In step S3, multiple time sequence radar vegetation indexes are stacked in a channel dimension through a first input layer, time sequence high-dimension characteristics of the time sequence radar vegetation indexes are extracted through a second long-short-time memory network layer, high-dimension characteristics output by the long-short-time memory network in all time steps are taken as input through a third multi-dimension one-dimensional convolution layer, translational invariance among channels of all the time steps of the high-dimension characteristics is extracted through a multi-dimension one-dimensional convolution kernel, a result of the multi-dimension one-dimensional convolution layer is taken as input through a fourth full-connection layer, and the yield estimated value of the rice in unit area is obtained through dimension reduction of the full-connection layer.
FIG. 2 illustrates a block diagram of a multi-scale one-dimensional convolution-long and short-term memory network model provided by an embodiment of the present invention. The model fully considers the contribution of the output of the model to the final regression result and the relation between the characteristics and the yield of the high-dimensional characteristic channels in each time step. The method specifically comprises the following steps:
step S301: and stacking a plurality of time sequence radar vegetation indexes in a channel dimension so as to meet the input format requirement of a second layer long-short time memory network layer. The time sequence length of the three time sequence radar vegetation indexes is 10, and the channel dimension is 1, namely [10,1]. After stacking the three time series radar vegetation indexes in the channel dimension, the dimension of the data becomes [10,3], as shown in fig. 2. The radar vegetation index after stacking is then input into a long short time memory network layer, i.e., an LSTM layer.
Step S302: and extracting the time sequence high-dimensional characteristics of the time sequence radar vegetation index by using the long-short time memory network layer. The long-short-time memory network layer has two long-short-time memory network units, namely LSTM units, and the dimension of a hidden layer of each LSTM unit is 64.
LSTM is a special recurrent neural network that can handle arbitrary length input sequences, capturing complex nonlinear relationships. It is one of the mainstream methods in the field of classification, processing and prediction tasks based on time series data. The LSTM layer abstracts the time sequence characteristics of the radar vegetation index to obtain high-dimensional characteristics. The layer outputs high-dimensional features in dimensions 10, 64.
Step S303: and extracting translation invariance among all time-step high-dimensional characteristic channels by using a multi-scale one-dimensional convolution kernel.
One-dimensional convolution is a variation of two-dimensional convolution, and the one-dimensional convolution kernel can only move in one direction to extract features, so that the method is more suitable for processing one-dimensional signal data. Because one-dimensional convolution requires less computational complexity and time than two-dimensional convolution, one-dimensional convolution is widely used in various fields such as rotating machinery fault detection, structural damage detection, real-time electrocardiogram monitoring, and the like.
The multi-scale one-dimensional convolution layer consists of three one-dimensional convolution kernels, and the sizes of the convolution kernels are 3,5 and 7 respectively. The LSTM obtained high-dimensional features of all time steps are input into a multi-scale one-dimensional convolution layer. The three convolution kernels extract the translational invariance of the high-dimensional features in the channel dimension, respectively. The multi-scale one-dimensional convolution layer keeps the high-dimensional characteristics of the model output at each time step, and the contribution of the model output at each time step to the final regression result is considered. The output dimension of convolution kernel 1 is [1,62], the output dimension of convolution kernel 2 is [1,60], and the output dimension of convolution kernel 3 is [1,58].
Step S304: and splicing the three one-dimensional convolved outputs, and sending the spliced outputs into the full-connection layer.
And splicing the output results of the three convolutions, wherein the output dimension is [1,180]. And fusing the multi-scale characteristics of each high-dimensional characteristic channel of each time step.
Step S305: and reducing the dimension of the result of the multi-scale one-dimensional convolution layer by using the full connection layer.
And taking the result of the multi-scale one-dimensional convolution layer as input, and reducing the dimension of the multi-scale one-dimensional convolution layer by the full-connection layer, wherein the dimension is [1,1].
Step S306: and (3) acting the Tanh activation function on the result of the last step to obtain a yield predicted value.
And (3) applying the Tanh activation function to the result of the last step, and increasing the expression capacity of the nonlinear relation between the time-lapse radar vegetation index and the rice yield of the model to obtain the rice yield estimation value in unit area with the dimension of [1,1]. The model can support the expression of higher level and more abstract of the vegetation index data of the timing radar, and improves the yield regression accuracy.
The planting time of double cropping rice in Guangdong province in the experimental area is early rice (4 months-8 months) and late rice (8 months-12 months). The total sample amount of early/late rice in 2017-2021 was 315, with 252 samples as training set (80%), and 63 samples as test set (20%). The early rice and the late rice are respectively trained and evaluated in precision.
Using a decision coefficient R 2 And unbiased RMSE (unbiased root mean square error, ubRMSE) to evaluate the performance of the regression model. R is R 2 The method is used for predicting the data to the extent that the measurement model can predict, and the value of the data is between 0 and 1.R is R 2 The closer to 1, the higher the predictive performance of the model. ubRMSE is the standard deviation of the residual. The smaller ubRMSE value indicates a smaller difference between the statistical yield and the predicted yield. The calculation formula is shown in equations (4) and (5).
Where n (i=1, 2, …, n) is the number of samples, y i Is the statistical data of the yield per unit area of the rice,is the corresponding average value, f i Is the estimated value of the yield per unit area of the predicted rice, < >>Is the corresponding average value.
Inputting three time sequence radar vegetation indexes into an improved multi-scale one-dimensional convolution-long-short time memory model to obtain unit area yield estimation results of early rice and late rice, and utilizing R 2 And ubRMSE for precision evaluation as shown in table 1.
As shown in Table 1, R in the estimated yield accuracy of early rice was obtained by the proposed method 2 0.67, ubRMSE of 217.77kg/ha. In the yield estimation precision of late rice, R 2 Is 0.61 and ubRMSE is 456.54kg/ha. The early rice yield and late rice yield estimation results show that the provided rice yield estimation method utilizing the radar vegetation index and the multi-scale one-dimensional convolution-long-short-time memory network can realize excellent yield estimation effect, and has great application prospect.
Table 1 timing radar vegetation index and multi-scale one-dimensional convolution-long and short term memory network model early-late rice unit area yield estimation results.
The rice yield per unit area estimation method utilizing the multiple radar vegetation indexes and the multi-scale one-dimensional convolution-long-short-term memory network provided by the embodiment of the invention fully excavates the relation between the radar vegetation indexes and the rice yield per unit area, and achieves a satisfactory yield estimation effect. And taking a plurality of representative radar vegetation indexes as input data, and fully exerting the yield estimation potential of the time sequence radar vegetation indexes. The complex nonlinear relation between the vegetation index of the time sequence radar and the yield of the unit area is abstracted by using the proposed multi-scale one-dimensional convolution-long-short-term memory network model. The multi-scale one-dimensional convolution layer reserves the high-dimensional characteristics of the model output in each time step, and considers the contribution of the model output in each time step to the final regression result; and extracting translation invariant features among the high-dimensional feature channels by utilizing multi-scale one-dimensional convolution, and fusing the multi-scale features of each time-step high-dimensional feature channel. Finally, the vegetation index data of the timing radar is expressed in a higher level and more abstract mode, and the yield regression accuracy is improved.
A second aspect of the embodiments of the present invention provides a rice unit area yield estimation device using a plurality of radar vegetation indexes and a multi-scale one-dimensional convolution-long-short-term memory network, including: the SAR time sequence data processing module is configured to unify the spatial resolution and the time resolution of the SAR time sequence data; a time-sequential radar vegetation index processing module configured to county scale radar vegetation index calculation; and the rice unit area yield estimation processing module is used for estimating the rice unit area yield by a multi-scale one-dimensional convolution-long-short-time memory network model.
A third aspect of the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of an embodiment of the above-described rice unit area yield estimation method using multiple radar vegetation indexes and a multi-scale one-dimensional convolution-long-short-term memory network.
The computer readable storage medium of embodiments of the present invention may include any entity or device capable of carrying computer program code, recording medium, such as ROM/RAM, magnetic disk, optical disk, flash memory, and so on.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (9)
1. A method for estimating the yield per unit area of rice, comprising the steps of:
step 1, unifying the spatial resolution and the time resolution of SAR time sequence data;
step 2, calculating various time sequence radar vegetation indexes by taking county level as a scale;
and 3, constructing a multi-scale one-dimensional convolution-long-short-time memory network model for estimating the yield of rice in unit area.
2. The method of claim 1, wherein step 1 comprises:
unifying the spatial resolution of SAR time sequence data;
and unifying the time resolution of the SAR time sequence data.
3. The method of claim 1, wherein step 2 comprises:
step 2.1, calculating various time sequence radar vegetation indexes;
and 2.2. Converting various time sequence radar vegetation indexes into the index taking county level as a scale.
4. The method of claim 1, wherein in the multi-scale one-dimensional convolution-long and short-term memory network model of step 3,
the first layer is an input layer and is used for stacking various time sequence radar vegetation indexes in a channel dimension so as to meet the input format requirement of a long-short-term memory network layer of the second layer;
the second layer is a long-short-term memory network layer and is used for extracting time sequence high-dimensional characteristics of various time sequence radar vegetation indexes;
the third layer is a multi-scale one-dimensional convolution layer, takes the time sequence high-dimensional characteristics output by the long-short-time memory network layer at all time steps as input, and utilizes the convolution kernel of the multi-scale one-dimensional convolution layer to extract translation invariance among all time steps high-dimensional characteristic channels;
the fourth layer is a full-connection layer and is used for taking the result of the multi-scale one-dimensional convolution layer as input, and obtaining the estimated value of the yield of the rice in unit area through dimension reduction of the full-connection layer.
5. The method of claim 4, wherein the long-short-time memory network layer comprises:
and the two long-short-time memory units are used for extracting time sequence high-dimensional characteristics of various time sequence radar vegetation indexes.
6. The method of claim 4, wherein the multi-scale one-dimensional convolution layer comprises:
three one-dimensional convolution kernels with different scales are convolved in the high-dimensional characteristic channel dimension; and splicing the convolution results of the three convolution kernels.
7. The method of claim 5, wherein the fully-connected layer comprises:
reducing the dimension of the result of the multi-scale one-dimensional convolution layer through the full connection layer;
and obtaining the estimated value of the yield of the rice in unit area by using a Tanh activation function.
8. An apparatus for use in the rice yield per unit area estimation method according to any one of claims 1 to 7, comprising:
the SAR time sequence data processing module is used for unifying the spatial resolution and the time resolution of the SAR time sequence data;
the time sequence radar vegetation index processing module is used for calculating the time sequence radar vegetation index by taking county level as a scale;
and the rice unit area yield estimation processing module is used for constructing a multi-scale one-dimensional convolution-long-short-time memory network model and estimating the rice unit area yield.
9. A computer-readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the rice yield per unit area estimation method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311583027.3A CN117556956A (en) | 2023-11-24 | 2023-11-24 | Rice unit area yield estimation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311583027.3A CN117556956A (en) | 2023-11-24 | 2023-11-24 | Rice unit area yield estimation method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117556956A true CN117556956A (en) | 2024-02-13 |
Family
ID=89823002
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311583027.3A Pending CN117556956A (en) | 2023-11-24 | 2023-11-24 | Rice unit area yield estimation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117556956A (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011102520A1 (en) * | 2010-02-22 | 2011-08-25 | 株式会社パスコ | Method of generating paddy rice crop yield forecasting model, and method of forecasting crop yield of paddy rice |
WO2021098472A1 (en) * | 2019-11-19 | 2021-05-27 | 浙江大学 | Crop yield estimation method based on deep temporal and spatial feature combined learning |
CN115222100A (en) * | 2022-06-23 | 2022-10-21 | 郑州大学 | Crop yield prediction method based on three-dimensional cyclic convolution neural network and multi-temporal remote sensing image |
CN115578637A (en) * | 2022-10-17 | 2023-01-06 | 中国科学院空天信息创新研究院 | Winter wheat yield estimation analysis method and system based on long-term and short-term memory network |
KR20230040765A (en) * | 2021-09-16 | 2023-03-23 | 한국항공우주연구원 | Crop yield prediction method and apparatus using the same |
CN116994144A (en) * | 2023-08-30 | 2023-11-03 | 河南大学 | High-resolution winter wheat yield estimation method, system, storage medium and electronic equipment based on synthetic aperture radar image |
-
2023
- 2023-11-24 CN CN202311583027.3A patent/CN117556956A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2011102520A1 (en) * | 2010-02-22 | 2011-08-25 | 株式会社パスコ | Method of generating paddy rice crop yield forecasting model, and method of forecasting crop yield of paddy rice |
WO2021098472A1 (en) * | 2019-11-19 | 2021-05-27 | 浙江大学 | Crop yield estimation method based on deep temporal and spatial feature combined learning |
KR20230040765A (en) * | 2021-09-16 | 2023-03-23 | 한국항공우주연구원 | Crop yield prediction method and apparatus using the same |
CN115222100A (en) * | 2022-06-23 | 2022-10-21 | 郑州大学 | Crop yield prediction method based on three-dimensional cyclic convolution neural network and multi-temporal remote sensing image |
CN115578637A (en) * | 2022-10-17 | 2023-01-06 | 中国科学院空天信息创新研究院 | Winter wheat yield estimation analysis method and system based on long-term and short-term memory network |
CN116994144A (en) * | 2023-08-30 | 2023-11-03 | 河南大学 | High-resolution winter wheat yield estimation method, system, storage medium and electronic equipment based on synthetic aperture radar image |
Non-Patent Citations (4)
Title |
---|
CHUNLING SUN ET AL: "RICE YIELD PREDICTION USING SENTINEL-1 RADAR VEGETATION INDICES AND XGBOOST", 《2023 SAR IN BIG DATE ERA》, 1 November 2023 (2023-11-01) * |
DIPANKAR MANDAL ET AL: "Dual polarimetric Radar Vegetation Index for Crop Growth Monitoring Using Sentinel-1 SAR Data", 《2021 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM IGARSS》, 12 October 2021 (2021-10-12) * |
潘信亮;张震;江涛;隋百凯;孙玉超;: "基于卷积长短期记忆网络的NDVI预测方法研究", 地理信息世界, no. 02, 25 April 2020 (2020-04-25) * |
薛利红, 曹卫星, 罗卫红: "基于冠层反射光谱的水稻产量预测模型", 遥感学报, no. 01, 10 February 2005 (2005-02-10) * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Dawood et al. | Deep-PHURIE: deep learning based hurricane intensity estimation from infrared satellite imagery | |
Wang et al. | SSRNet: In-field counting wheat ears using multi-stage convolutional neural network | |
CN113095409B (en) | Hyperspectral image classification method based on attention mechanism and weight sharing | |
Miao et al. | Classification of farmland images based on color features | |
Su et al. | LodgeNet: Improved rice lodging recognition using semantic segmentation of UAV high-resolution remote sensing images | |
Sun et al. | Wheat head counting in the wild by an augmented feature pyramid networks-based convolutional neural network | |
CN115861858B (en) | Small sample learning crop canopy coverage calculating method based on background filtering | |
Guo et al. | Multitemporal hyperspectral images change detection based on joint unmixing and information coguidance strategy | |
CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
CN114202643A (en) | Apple leaf disease identification terminal and method based on multi-sensor fusion | |
Farooque et al. | Red-green-blue to normalized difference vegetation index translation: a robust and inexpensive approach for vegetation monitoring using machine vision and generative adversarial networks | |
Islam et al. | Improving hyperspectral image classification through spectral-spatial feature reduction with a hybrid approach and deep learning | |
Farooque et al. | A dual attention driven multiscale-multilevel feature fusion approach for hyperspectral image classification | |
CN112464172A (en) | Growth parameter active and passive remote sensing inversion method and device | |
CN116863341A (en) | Crop classification and identification method and system based on time sequence satellite remote sensing image | |
CN114708307B (en) | Target tracking method, system, storage medium and device based on correlation filter | |
CN117556956A (en) | Rice unit area yield estimation method and device | |
Varma et al. | HSIS-Net: Hyperspectral Image Segmentation Using Multi-view Active Learning Based FCSN. | |
Li et al. | Early drought plant stress detection with bi-directional long-term memory networks | |
Ramachandran et al. | Tiny Criss-Cross Network for segmenting paddy panicles using aerial images | |
CN115147727A (en) | Method and system for extracting impervious surface of remote sensing image | |
CN114519402A (en) | Citrus disease and insect pest detection method based on neural network model | |
Shukla et al. | Predictive modeling of vegetative drought using ML/DL approach on temporal satellite data | |
Stanski et al. | Flower detection using object analysis: new ways to quantify plant phenology in a warming tundra biome | |
Rong et al. | Grape cluster detection based on spatial-to-depth convolution and attention mechanism |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |