TWI703529B - A method for calculating a growth stage of a crop and computer program product - Google Patents

Abstract

A method for calculating a growth stage of a crop includes: establishing a basic growth model including a plurality of environmental ecological data and a number of days required for a crop in a growth stage; using a machine learning algorithm using a plurality of historical data corresponding to the plurality of environmental ecological data and the number of days in the growth stage, a plurality of features of the basic growth model are automatically adjusted, and an optimized growth model is established; and a predicted number of days of the growth stage of the crop is calculated by the optimized growth model based on a plurality of current data corresponding to the plurality of environmental ecological data.

Description

Calculation method of crop growth stage and computer program product

The present invention relates to a calculation method and computer program product for crop growth stages, in particular to a calculation method and computer program product for crop growth stages for predicting growth days.

Traditional agriculture mostly relies on the personal experience of agricultural workers and the inheritance of masters and apprentices to determine the pros and cons of the growth environment and the impact on the growth cycle of crops, or some scholars have calculated a generalized standard model of crops based on accumulated data under standard environmental ecology The theoretical growth cycle.

However, in fact, the environmental and ecological conditions such as soil properties, temperature, humidity, and rainfall of the production areas of different latitudes and longitudes are different, and they are different from the standard environmental ecological conditions. Therefore, the aforementioned crop standard model cannot accurately calculate the specific growth cycle and the time course of each stage for the crops in each producing area.

In view of this, some embodiments of the present invention provide a method for calculating the growth stage of a crop and a computer program product.

A method for calculating the growth stage of a crop according to an embodiment of the present invention includes: establishing a basic growth model including plural environmental and ecological data required for a crop in a growth stage and the number of days in a stage; using plural corresponding to the plural environmental and ecological data and the number of days in the stage historical data, A machine learning algorithm is used to automatically adjust the multiple characteristic parameters of the basic growth model, and an optimized growth model is established; and based on the current complex environmental ecological data, the optimized growth model is used to calculate the predicted number of days for the corresponding growth stage of the crop.

A computer program product according to another embodiment of the present invention includes a set of instructions. When the computer loads and executes the set of instructions, the method for calculating the crop growth stage according to any embodiment of the present invention can be completed.

The following detailed descriptions are provided with specific embodiments in conjunction with the accompanying drawings to make it easier to understand the purpose, technical content, features, and effects of the present invention.

S1~S4, S31~S33: steps

1: Electronic device

10: Operation unit

20: storage unit

30: Image capture unit

40: Communication unit

FIG. 1 is a schematic diagram of the steps of a method for calculating the growth stage of a crop according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of the structure of an electronic device that implements an embodiment of the crop growth stage calculation method of FIG. 1.

3 is a schematic diagram of the steps of a method for calculating crop growth stages according to an embodiment of the present invention.

Hereinafter, each embodiment of the present invention will be described in detail, with drawings as an example. In the description of the specification, in order to enable the reader to have a more complete understanding of the present invention, many specific details are provided; however, the present invention may still be implemented under the premise that some or all of the specific details are omitted. The same or similar elements in the drawings will be represented by the same or similar symbols. It is particularly important to note that the drawings are for illustrative purposes only, and do not represent the actual size or quantity of the components. Some details may not be completely drawn in order to keep the drawings concise.

Please refer to FIGS. 1 and 2 together, which show that the method for calculating the growth stage of crops according to any embodiment of the present invention can be implemented by a computer program, so that it can be used as a computer (that is, with a computing unit 10 Any electronic device 1 of the storage unit 20, such as a server, a tablet computer, or a smart phone, can complete the crop growth stage calculation method of any embodiment after the program is loaded and executed.

Since the existing standard models of crops cannot be integrated into different production areas, different varieties of the same crop grown in different periods or after improvements in the same production area, through the crop growth stage calculation method of this embodiment, the existing model is used as the basis and through the environment Ecological data training and machine learning algorithms will automatically modify the existing model parameters and corresponding weight values to establish a specific model for adaptive adjustment. The detailed operation principle is described below.

In at least one embodiment, the crops can be, but are not limited to: edible crops, special crops (ie, craft crops, horticultural crops, commercial crops, trade crops), miscellaneous crops, and the like. The growth cycle of a crop includes at least one growth stage. For example, from the initial rooting to the final harvest, the pineapple crop will have different stage traits in sequence, as shown in the following table 1. The pineapple growth cycle includes the seedling and planting period, Growth stages such as the plant development period, the flowering period, the flowering period, the fruit development period, and the fruit maturity period. For example, the period from the beginning of the root to the first leaf is the seedling and planting period.

In this embodiment, through step S1, the arithmetic unit 10 establishes a basic growth model corresponding to the crop as a basis for subsequent machine learning.

For example, the computing unit 10 collects past environmental and ecological data of a place of production, including but not limited to: atmospheric temperature (℃), atmospheric humidity (%), atmospheric pressure (hPa), wind speed (m/s), accumulated rainfall ( mm), dew point temperature (℃), solar radiation (W/m ² ), degree of ultraviolet index, soil temperature (℃), soil water content (%), soil electrical conductivity (ds/m), soil pH, Carbon dioxide concentration (ppm), nitrogen concentration (ppm) and a combination of above, etc. For example: According to historical experimental data, the optimal temperature range for pineapple crops during the fruit development period is 28°C to 32°C, which has an important influence on the growth rate of the crop and the number of days in the stage. Therefore, the temperature is the growth of pineapple crops during the fruit development stage. One of the required environmental and ecological data. In some embodiments, the environmental ecological data can be obtained by a sensing device at the location of the crop, and then transmitted to the electronic device 1 through wired or wireless communication technology. At the same time, the arithmetic unit 10 uses the plural stage traits that appear in sequence to define the corresponding growth stage, and calculates the number of days corresponding to the growth stage. For example, the arithmetic unit 10 finds the appearance time of the hair root trait and the long leaf trait, and defines For the seedling picking and planting period, subtract the two times to calculate the number of days in the seedling picking and planting growth stage.

Then, the computing unit 10 calculates a basic growth model that can be used for machine learning through, for example, but not limited to, regression analysis methods based on the environmental ecological data and the number of days in the stage. Among them, the basic growth model includes environmental and ecological data required for the growth of the crop in the growth stage and the number of days in the stage.

In another embodiment, the basic growth model can be taken from an existing standard crop model, which includes the approximate number of days required for the growth stage and the most suitable environmental and ecological conditions, that is, the basic growth model includes the crops required during the growth stage Plural environmental and ecological data and the number of days in the stage. Lift For example, the computing unit 10 selectively receives a basic growth model from the outside through a communication unit 40 and stores it in the storage unit 20, as shown in FIG. 2, whereby the computing unit 10 can read and create a corresponding crop The basic growth model is used as the basis for subsequent machine learning.

From the above description, it can be seen that whether the basic growth model is taken from existing crop models or experimental models obtained through historical experimental data analysis, etc., they cannot target crops in different regions and different climates (for example, pineapples are harvested twice in three years, each The environment and climate are different during the period), and the growth cycle of different improved varieties is customized to calculate, and accurately provide the specific growth phase forecast days.

In this embodiment, through step S2, the arithmetic unit 10 uses the plural historical data corresponding to the plural environmental ecological data and the number of days to automatically adjust the plural characteristic parameters of the basic growth model through a machine learning algorithm, and establishes an optimized growth model . Then, through step S3, the computing unit 10 calculates the predicted number of days for the corresponding growth stage of the crop through the optimized growth model based on the current complex environmental and ecological data. The predicted number of days will change with the regional climate and ecological environment, corresponding to the current crop The ecological conditions of the growth environment are used to calculate the remaining days to the next growth stage.

Take a basic growth model with a single growth stage (e.g., seedling and planting period) as an example. The growth stage is defined as the stage character from the root to the first leaf. Through the observation of the character, the beginning and the beginning of the stage can be understood. The end date, calculate the number of days in a stage, as the label of the machine learning calculation model, and organize the environmental and ecological data of the same period. The horizontal axis is the time, the unit is day, and the vertical axis is the normalization value of each item. , Organized into a set of two-dimensional data array as the feature parameters of the model. With the technology of Convolutional Neural Network (CNN, Convolutional Neural Network), through the feeding of parameters and label learning again and again, the model is automatically adjusted through back propagation and/or gradient descent For each neuron parameter, one input is environmental ecological data, and the output is a deep machine learning model that predicts the number of days. That is, the computing unit 10 uses the number of days in the stage as the tag parameter and the complex environmental ecological data as the complex feature parameter, and automatically adjusts the complex weight value of the complex feature parameter in the convolutional neural network through the back propagation technology. In other words, the method of automatically adjusting the weight value of characteristic parameters is to use the technology of data analysis to automatically weight and judge the environmental and ecological data, and calculate the amount of change in the environmental and ecological data in each stage for the production of the growth traits corresponding to the next growth stage. The amount of time affects, the model is generated and the model is continuously optimized.

For example, in a planting record of pineapple, the date of rooting is November 15, 2017, and the date of the first leaf is December 15, 2017, then the convolutional neural network is used in this training data The label of is 30 days. Starting from the first day, the environmental ecological data of the period will be sorted, and the corresponding advanced environmental data will be generated through data fusion and data transformation technologies to form Environmental and ecological data for subsequent calculations. Among them, advanced environmental data includes but is not limited to: daily maximum atmospheric temperature (℃), daily minimum atmospheric temperature (℃), cumulative atmospheric temperature (℃), cumulative insolation value of the day (MJ), and progressive insolation (MJ) , Daily progressive rainfall (mm), daily maximum soil temperature (℃), daily minimum soil temperature (℃), cumulative soil temperature (℃), maximum nitrogen concentration of the day (ppm), maximum carbon dioxide of the day (ppm) ) And a combination of the above. In some embodiments, by analyzing the farming records in the field, the farming work can be sorted from the farming work records into, for example, but not limited to: whether there is a binary value for weeding or not, or a continuous amount such as the amount of watering (mm) Agricultural records information such as numerical values; it can also be obtained from the trait records, such as but not limited to: plant height (cm), plant weight (g), leaf number, fruit weight and other trait information. Therefore, the machine learning calculation model can selectively integrate agricultural Record information and trait information are used as characteristic parameters of the machine learning algorithm to effectively improve the accuracy of prediction.

In one embodiment, the plural environmental ecological data can be light insolation, soil water content, daily maximum atmospheric temperature, and daily minimum atmospheric temperature. Good training effects can be obtained in subsequent machine learning algorithms to avoid excessive simulation. Together, but not limited to this. Taking the first day as an example, the model will organize the information in the record into a multivariate two-dimensional array of training data. The two-dimensional horizontal axis includes the 14 items mentioned above, such as the amount of sunlight and soil moisture in the environmental ecological data. Basic environmental ecological data, daily maximum temperature and minimum temperature obtained by data conversion, the aforementioned 11 advanced environmental data, the aforementioned 55 agricultural record information such as weeding and watering amount analyzed in field records, and plant height , Plant weight, number of leaves, etc. of the aforementioned 10 traits; while the two-dimensional vertical axis is the time unit. On the first day, there will be only the first day’s information, so the input will be a 1*95(14+11+ 55+10) two-dimensional matrix, the label at this time is 29 (the end of 29 days), and so on, the input and output parameters obtained on the last day are 30*1=30 rows, and the vertical axis is a two-dimensional with 95 columns Array, label with number 0.

Then, after inputting the plural environmental and ecological data currently obtained, the predicted number of days remaining in the growth stage of the seedling and planting period can be obtained. The data on the first day of the growth period in different regions (1*25 dimensional array) may be Because the environmental and ecological data of the two places are different, the results of different days such as 26 to 29 will be obtained, so as to realize customized calculation and accurately provide the specific growth stage forecast days. At the same time, the newly obtained environmental ecological data and the judgment of the number of days in the stage can be used as a new training sample to give feedback to the model, and subsequently used to optimize the accuracy of the model's judgment.

According to the above description, the crop growth stage calculation method of some embodiments of the present invention can be customized for the growth cycle of crops under different environmental and ecological conditions, different periods, and different improved varieties, and accurately provide specific growth stage prediction days. For example, the pineapple varieties in Taiwan have undergone various improvements and are different from the standard model of existing pineapple crops, such as: Tainong No. 17 Golden Diamond Pineapple is an improved variety. The growth requirements of the pineapple before the improvement are different, and its overall growth cycle is also different, and it is different from the existing model. The standard growth cycle of the existing model is 18 months. After learning the growth cycle of the golden pineapple in our actual field, the specific growth cycle of the golden pineapple is 16 and a half months, which is more in line with the actual crop growth. Significantly improve the technical defects of existing models.

The calculation method of crop growth stage according to some derivative embodiments of the present invention is described below. In one embodiment, the basic growth model includes plural environmental and ecological data required by a crop in plural growth stages and the number of days in plural stages. Because the environmental and ecological data of pineapples at different growth stages are not the same, for example, during the development of the pineapple plant, temperature will be the most important factor affecting the growth, that is, slight changes in temperature will affect the growth time of the pineapple. Soil moisture changes have little effect. Relatively speaking, the soil moisture content has the greatest impact on plants during the harvest period. Therefore, by training different machine learning models at each growth stage, it can be tailored to each growth stage. The stage focuses on different weights for optimization without affecting the learning of the parameters. In other words, the same feature parameter will have different weight values in different growth stages under the machine learning model, and a feature parameter will have more than two weights. Value instead of a single fixed value, that is, each characteristic parameter has a different complex weight value in different growth stages. When the model is facing new crops and the training is not sufficient, the output of the model will be different from the actual situation. In order to make up for this gap, the automatic calculation of the growth stage will simultaneously analyze the agricultural records to deduce the current plant belongs to Growth period.

Please refer to FIG. 2 and FIG. 3 together. In this embodiment, through step S31, the electronic device 1 receives the current plural environmental ecological data and a farming record. For example, the electronic device 1 optionally includes an image capturing unit 30, and the image capturing unit 30 is electrically connected to the computing unit 10. The image capturing unit 30 captures the agricultural record images and transmits them to the computing unit 10. For example, the electronic device 1 can be a mobile internet device such as a wireless camera or a smartphone, but not limited to this; or, the electronic device 1 is optional A communication unit 40 is included, and the communication unit 40 is electrically connected to the computing unit 10. For example, the communication unit 40 can be a wireless communication interface that establishes a connection with a remote device through a wireless communication protocol. The communication unit 40 receives the farming record images and sends them to the computing unit 10. That is, the electronic device 1 can receive one or more farming record images from the mobile internet device installed in the crop production area through wired and wireless network communication. In this way, the electronic device 1 can be uploaded to a server or cloud system for subsequent data calculation and prediction. For example, the electronic device 1 can be a server or a desktop computer, but not limited to this.

In one embodiment, the multiple growth traits and growth stages of the crops in sequence can be judged through agricultural records, and then the number of days in the stage can be calculated, which will then be fed back to the model as a new training label for deep learning. For example, agricultural records are the daily operations records of agricultural workers on crops, including the time of application, the operation method of the action, the equipment used and other records related to the daily agricultural work, as well as the records of plant traits. Example records are shown in Table 2.

With text analysis technology, including Chinese word segmentation (word segmentation), part of speech tagging (PoS, Part of Speech), such a pen record can be analyzed. For example, like growing "leaves", combined with the description of the traits and the definition of the growth stage, it can be judged that the current plant should belong to the developmental stage; and the more important "wearing cap" is extracted from "wearing a cap for pineapple" Actions, combined with a defined vocabulary set, can be used to determine the basis. For example, the agricultural action of "wearing a hat" will only occur when the pineapple has fruited, so the scope can be narrowed to the fruit development period and the fruit mature period. From the word "fertilization", it can be judged that the current growth stage of the crop is the fruit development stage. In other words, through step S32, the agricultural records are automatically analyzed, and the crop growth stage is determined according to the character look-up table (such as but not limited to the one shown in Table 1). In addition, through the analysis of field farming records, labels can be set for each growth stage as feedback to the basic growth model learning; it can also be converted into word vectors by converting text records into word vectors through principal component analysis ( PCA, Principal Component Analysis) and association rule learning (association rule learning), automatically expand the vocabulary set of each growth stage.

Then, through step S33, the computing unit 10 can more accurately calculate the predicted number of days for the corresponding growth stage of the crop based on the current growth stage and the current multiple environmental ecological data parsed from the agricultural records. It is supplemented that in addition to using the currently collected environmental and ecological data, future forecasts or simulated environmental and ecological data can also be used to improve the accuracy of the forecast and estimation. For example, the computing unit 10 can simulate the future environmental and ecological data through weather forecasting or combining with historical environmental data of the crop area to estimate the subsequent growth schedule. For example, the current date is 3/15. In the fourth growth stage, it is judged that the remaining days to the next growth stage is 20 days. In the near future, the weather forecast can be used to determine the remaining days of the day. For example, the forecast of 3/22 will be combined Enter the parameters to get the next distance on the day of 3/22 The remaining days of each growth stage may be 12 days. In the long-term future, the historical environmental data of the crop area is subjected to operations such as climate translation. For example, the weather changes in May last year are more similar to those in April this year. The data at the same time last year was panned and zoomed, and then combined with environmental ecological data to simulate the remaining days at a future point in time, and then calculate the overall growth cycle to find out the expected harvest date of the crop.

In addition, in some embodiments, by giving the current growth stage and current environmental ecological data, the computing unit 10 can determine which pests and diseases are likely to occur in the current environment through, for example, but not limited to, an expert system, and then generate corresponding agricultural behavior recommendations. As shown in Table 3, for example, provide qualified recommended spraying agents and their dilution ratios (percentages), and provide recommendations for the implementation of certain agents that require special attention. Take the seedling and planting period as an example. This is the growth stage where parasitic nematodes are prone to occur. Several mixed spraying agents can be recommended. Seven days before planting, 15 cm deep in the planting ditch, and after application Cover the soil immediately and compact it, or for example: Take the plant development period as an example, if it does not rain within two days, you need to water to 30% of the water content. In short, through step S4, the computing unit 10 provides a corresponding agricultural behavior suggestion based on the predicted number of days.

In some embodiments, the computer program that implements the crop growth stage calculation method according to any embodiment of the present invention is composed of a set of instructions, and the computer program can be stored in a computer storage medium or a computer program product.

In some embodiments, the arithmetic unit 10 may be one or more such as microprocessors, microcontrollers, digital signal processors, microcomputers, central processing units, field programming gate arrays, programmable logic devices, state machines, logic circuits , Analog circuit, digital circuit and/or any processing element based on operation command operation signal (analog and/or digital).

In some embodiments, the storage unit 20 may be implemented by one or more memories.

Based on the above, some embodiments of the present invention provide a method for calculating the growth stage of crops and computer program products, which use environmental ecology such as but not limited to: solar radiation, soil moisture content, daily maximum atmospheric temperature, and daily minimum atmospheric temperature. Data, through the machine learning algorithm to establish an automatic learning optimized growth model to predict the subsequent crop growth period, and calculate the predicted days of each growth stage. Therefore, customized calculations can be made for the growth cycle of crops in different environmental and ecological conditions and different improved varieties, and precise specific growth prediction days can be provided, which significantly improves the technical defects of existing standard crop models. In addition, by training different machine learning models at each different growth stage, it is possible to optimize the different weights of each growth stage, so that each feature parameter has a different complex weight value in different growth stages. It will not affect the learning of characteristic parameters.

The above-mentioned embodiments are only to illustrate the technical ideas and features of the present invention, and their purpose is to enable those who are familiar with the art to understand the content of the present invention and implement them accordingly. If the scope of the patent of the present invention cannot be limited by this, That is, all equal changes or modifications made in accordance with the spirit of the present invention should still be covered by the patent scope of the present invention.

Claims (9)

A method for calculating the growth stage of a crop, comprising: an arithmetic unit establishes a basic growth model that includes a plurality of environmental and ecological data required by a crop in a growth stage and a number of days in a stage, wherein the basic growth model includes a plurality of the growth stages; The complex environmental ecological data and the complex historical data of the number of days in the stage are automatically adjusted by a machine learning algorithm to automatically adjust the complex weight values of the complex feature parameters of the basic growth model, and an optimized growth model is established, wherein each feature parameter is different The growth stage has a different complex weight value; and the arithmetic unit calculates a predicted number of days in the growth stage corresponding to the crop through the optimized growth model based on the current complex environmental ecological data, wherein the growth stage is calculated The step of predicting the number of days includes: calculating according to the current growth stage and the current plural environmental ecological data. The method for calculating the growth stage of the crop as described in claim 1, further includes: the arithmetic unit automatically analyzes a farming record, and determines the growth stage according to a character look-up table. The method for calculating the growth stage of the crop according to claim 2, further comprising: the arithmetic unit receives the current plural environmental ecological data and the agricultural record. The method for calculating the growth stage of a crop according to claim 1, wherein the plural environmental and ecological data include light solar radiation, soil water content, daily maximum atmospheric temperature, and daily minimum atmospheric temperature. The crop growth stage calculation method according to claim 1, wherein the step of automatically adjusting the basic growth model includes: using the number of days in the stage as a label parameter and using the plural environmental and ecological data as the plural feature parameter , Automatically adjust the complex weight value of the complex feature parameter in a convolutional neural network through a back propagation technique. The method for calculating the growth stage of a crop according to claim 5, wherein the plurality of characteristic parameters includes a plurality of two-dimensional array parameters formed by the accumulation of the plurality of environmental and ecological data along with a number of growth days in the growth stage. The method for calculating the growth stage of the crop as described in claim 1 further includes: the arithmetic unit provides a corresponding agricultural behavior suggestion based on the predicted number of days. The crop growth stage calculation method according to claim 1, wherein the step of establishing the basic growth model includes: the computing unit defines the number of days in the stage according to the two stage traits appearing in sequence. A computer program product includes a set of instructions. When the computer loads and executes the set of instructions, the calculation method of the crop growth stage as described in any one of the request items 1 to 8 can be completed.

Images