A greenhouse crop irrigation method based on deep neural network

Abstract

The present invention provides a greenhouse crop irrigation method based on a deep neural network. First, a crop transpiration rate prediction model, a greenhouse environmental parameter prediction model, a crop water state detection model and a crop water stress recovery time prediction model based on a deep neural network are constructed; then the crop planting date, the current date, the current time and the greenhouse environmental parameter prediction value are input into the crop transpiration rate prediction model to obtain the crop transpiration rate prediction value, and the time is integrated within the interval of the next irrigation time to obtain the crop transpiration amount prediction value; the planting date, the current date, the current time, the greenhouse environmental parameters, the RGB image, the depth image and the near infrared image of the crop canopy leaves are input into the crop water state detection model to judge the crop water state, and irrigation is performed according to the crop water state and the crop transpiration amount prediction value. The present invention can dynamically adjust the irrigation time and irrigation amount, realize accurate irrigation of crops, and save irrigation water.

Description

Technical Field

[0001] The present invention belongs to the technical field of greenhouse crop irrigation, and in particular relates to a greenhouse crop irrigation method based on a deep neural network. Background Art

[0002] Irrigation is the only source of water for greenhouse crops. Too much or too little water supply will have a certain impact on the yield and quality of crops. Too much water supply will harm the root system of crops, and the air humidity in the greenhouse will be too high, leading to an increase in pests and diseases, and causing water waste; too little water supply will reduce the photosynthetic rate of crops and cause a reduction in yield. The daily water requirement of crops will vary at different growth stages. Crops require less water in the seedling stage. As the crops grow, the water requirement of crops gradually increases. In order to control the quality, the water requirement will decrease during the harvest period. Changes in temperature, humidity and light in the greenhouse at different times of the day will also affect the water requirement of crops. The water requirement of crops increases under high temperature and high light intensity; the water requirement of crops decreases under low temperature and low light.

[0003] A Chinese patent (CN106258855A) discloses an intelligent irrigation system based on light radiation. When the accumulated energy of light radiation is greater than the predetermined critical energy, the irrigation system performs quantitative irrigation. In 2018, Gu Zhe et al. disclosed in the 23rd issue of Volume 34 of the Journal of the Chinese Society of Agricultural Engineering a real-time and precise irrigation decision-making for solar greenhouses based on evapotranspiration and water balance. The time scale is days. When the total amount of evaporation in the field is greater than the amount of water available for crops in the soil, irrigation is triggered. The irrigation amount is equal to the sum of evapotranspiration since the last irrigation period. In 2019, Chen Shiwang et al. disclosed in the supplement to Volume 50 of the Journal of the Chinese Society of Agricultural Machinery a prediction model for daily transpiration of a single tomato plant with inputs of changes in substrate moisture content, air temperature, air humidity and light intensity. In 2020, Xu Lihong et al. published a tomato matrix cultivation irrigation model based on greenhouse environment and crop growth in Volume 36, Issue 10 of the Journal of the Chinese Society of Agricultural Engineering. The original expression of the Penman-Monteith model was modified to remove the soil evaporation part, and the TOMGRO model was introduced to simulate the growth of the tomato canopy to correct the impedance parameters, and a new transpiration model was obtained; and the influence of solar radiation on the water consumption below the crop canopy was considered, and a model for the daily irrigation amount of a single tomato plant was established. In 2020, Li Li et al. published a study on strawberry irrigation strategy based on the K-means clustering algorithm in Volume 51, Issue 1 of the Journal of the Chinese Society of Agricultural Machinery, and quantitative irrigation was carried out according to the changes in matrix moisture content and the daily average temperature range. The existing technology mainly considers greenhouse environmental parameters, matrix moisture content, etc. to make irrigation decisions, without considering the actual growth of crops, and only predicts in units of days. It cannot dynamically adjust the irrigation time and irrigation amount according to the actual environmental parameters of the day and the actual water demand of crops, and it is difficult to meet the precise irrigation needs of greenhouse crops. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a greenhouse crop irrigation method based on deep neural network, which can timely detect the moisture status of crops, and dynamically adjust the irrigation time and irrigation amount according to the moisture status of crops and environmental parameters, so as to achieve precise irrigation of crops.

[0005] The present invention achieves the above technical objectives through the following technical means.

[0006] A greenhouse crop irrigation method based on deep neural network, specifically:

[0007] Every m minutes, the greenhouse environmental parameters and the RGB images, depth images, and near-infrared images of the crop canopy leaves are collected and input into the crop moisture status detection model to determine the moisture status of the crop;

[0008] If the crop water status is irrigation deficit, irrigation is immediately carried out; and the next crop water status detection is carried out after the crop water stress recovery time has passed;

[0009] If the crop water status does not show irrigation deficit within n hours after irrigation, the crop water status is determined by the crop water status detection model n hours after irrigation, and irrigation is performed; wherein n is the irrigation interval.

[0010] Furthermore, the crop moisture status detection model is specifically as follows: taking the planting date, current date, current time, greenhouse environmental parameters, and RGB images, depth images, and near-infrared images of crop canopy leaves as input, and taking the crop moisture status as output, a deep neural network model for crop moisture status detection is constructed.

[0011] Furthermore, the irrigation deficit is a crop moisture state where the irrigation level is less than F(1-P), the irrigation level with the highest net photosynthetic rate is defined as the irrigation benchmark, denoted by F; and the allowable relative deviation is defined as the threshold, denoted by P.

[0012] Furthermore, the crop moisture status when the irrigation level is greater than F(1+P) is judged as over-irrigation, and the crop moisture status in the interval [F(1-P), F(1+P)] is judged as appropriate irrigation.

[0013] Furthermore, the crop water stress recovery time is predicted based on a crop water stress recovery time prediction model. Specifically, the crop water stress recovery time prediction model takes the crop planting date, current date, current time, greenhouse environmental parameters and irrigation amount as inputs and the crop water stress recovery time as output to establish a deep neural network model for predicting the crop water stress recovery time.

[0014] Furthermore, the irrigation amount is determined by the following formula:

[0015]

[0016] Where: Q is the irrigation amount per plant, Q pred is the predicted value of transpiration per plant in the next n hours, Q' is the irrigation amount per plant at the last irrigation, and ET is the transpiration per plant from the last irrigation to the current moment.

[0017] Furthermore, the predicted value of transpiration per plant Q pred The acquisition method is as follows: input the greenhouse environmental parameters every minute of the first 24 hours, the weather forecast information at the hour in the next n hours, and the control parameter setting values ​​of the greenhouse environmental regulation system into the greenhouse environmental parameter prediction model to obtain the greenhouse environmental prediction information in the next n hours; input the crop planting date, current date, current time and greenhouse environmental prediction information into the crop transpiration rate prediction model to calculate the crop transpiration rate; integrate the crop transpiration rate in the next n hours to obtain the crop transpiration amount prediction value Q pred .

[0018] Furthermore, the crop transpiration rate prediction model is specifically as follows: the crop planting date, current date, current time and greenhouse environmental parameters are used as the input of the model, and the crop transpiration rate is used as the output of the model to construct a deep neural network model for crop transpiration rate prediction.

[0019] Furthermore, the greenhouse environmental parameter prediction model is specifically as follows: the greenhouse environmental parameters every minute in the previous 24 hours, the weather forecast information at the hour in the next n hours, and the control parameter setting values ​​of the greenhouse environmental regulation system are taken as input, and the greenhouse environmental parameters every minute in the next n hours are taken as output to construct a deep neural network model for greenhouse environmental parameter prediction.

[0020] Furthermore, m and n satisfy that (n×60) is divisible by m.

[0021] The beneficial effects of the present invention are as follows: the present invention can timely judge the moisture status of crops and irrigate them according to the planting date, current date, current time, greenhouse environmental parameters and RGB images, depth images and near-infrared images of crop canopy leaves in one day, and consider the impact of greenhouse environmental regulation on the greenhouse environment, and then dynamically adjust the irrigation amount according to the greenhouse environmental parameters and the current time to achieve precise irrigation of crops, while also saving irrigation water and reducing waste of water resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of the greenhouse crop irrigation method based on deep neural network described in the present invention. DETAILED DESCRIPTION

[0023] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0024] This embodiment takes the greenhouse tomato irrigation method based on deep neural network as an example to further illustrate the implementation process.

[0025] S1, building a crop transpiration rate prediction model based on deep neural network

[0026] Cultivate the crops needed to establish the crop transpiration rate prediction model, irrigate the crops sufficiently, record the planting date of the crops, and collect the current date, current time, greenhouse environmental parameters and crop transpiration rate every minute.

[0027] The planting date, current date, current time and greenhouse environmental parameters of the crop are used as the input of the model, and the transpiration rate of the crop is used as the output of the model to construct a deep neural network model for predicting the transpiration rate of the crop. In this example, the tomato transpiration rate prediction model has 3 layers and 6 hidden layer neurons. The above collected data is used for training to obtain a crop transpiration rate prediction model based on a deep neural network.

[0028] In this embodiment, greenhouse environmental parameters are collected by a weather station in the greenhouse, and tomato transpiration rate is measured using a wrapped stem flow system.

[0029] S2, build a greenhouse environmental parameter prediction model based on deep neural network

[0030] The greenhouse environment control system automatically adjusts and records the greenhouse environment parameters, current weather forecast and greenhouse environment control system control parameter settings every minute. The time interval between two crop irrigations is defined as the irrigation interval, recorded as n, in hours. In this example, n=1. The greenhouse environment parameters every minute in the first 24 hours, the weather forecast information at the hour in the next n hours and the greenhouse environment control system control parameter settings are used as input, and the greenhouse environment parameters every minute in the next n hours are used as output to construct a deep neural network model for greenhouse environment parameter prediction. In this embodiment, the greenhouse environment parameter prediction model has 3 layers and 45 hidden layer neurons. Use the recorded greenhouse environment parameters, weather forecast information and greenhouse environment control system control parameter settings for training to obtain a greenhouse environment parameter prediction model based on a deep neural network.

[0031] S3, building a crop moisture status detection model based on deep neural network

[0032] Cultivate the crops required to establish the crop water status detection model, set different irrigation levels, and make the crops in different water status. Collect the planting date, current date, current time, greenhouse environmental parameters, and RGB images, depth images, near-infrared images, and net photosynthetic rate of crop canopy leaves every minute; define the irrigation level with the highest net photosynthetic rate as the irrigation benchmark, denoted by F; define the allowable relative deviation as the threshold, denoted by P, in this example, P = 5%; the crop water status with an irrigation level greater than F(1+P) is judged as excessive irrigation, the crop water status in the interval [F(1-P), F(1+P)] is judged as appropriate irrigation, and the crop water status less than F(1-P) is judged as insufficient irrigation.

[0033] The planting date, current date, current time, greenhouse environmental parameters, and RGB images, depth images, and near-infrared images of crop canopy leaves were used as input, and the crop moisture status was used as output to construct a deep neural network model for crop moisture status detection. The collected greenhouse environmental parameters, RGB images, depth images, near-infrared images of crop canopy leaves, and crop moisture status were used for training to obtain a crop moisture status detection model based on a deep neural network.

[0034] The crop canopy leaf RGB image, depth image and near-infrared image are collected using RealSense's D435i depth camera.

[0035] The net photosynthetic rate is measured using a photosynthetic instrument.

[0036] S4, Construction of a crop water stress recovery time prediction model based on deep neural network

[0037] Cultivate the crops required to establish the crop water stress recovery time prediction model, and input the planting date, current date, current time, greenhouse environmental parameters, and RGB images, depth images, and near-infrared images of crop canopy leaves into the crop water state detection model to obtain the crop water state. When the crop is in an irrigation deficit state, fully irrigate the crop, and record the planting date, current date, current time, greenhouse environmental parameters, and irrigation amount of the crop. Thereafter, detect the crop water state once a minute until the crop water state is no longer an irrigation deficit, and the elapsed time is recorded as the crop water stress recovery time.

[0038] The planting date, current date, current time, greenhouse environmental parameters and irrigation amount of the crop are used as inputs, and the crop water stress recovery time is used as output to establish a deep neural network model for predicting the crop water stress recovery time. In this embodiment, the tomato water stress time prediction model has 3 layers and 6 hidden layer neurons. The above-mentioned collected data is used for training to obtain a crop water stress recovery time prediction model based on a deep neural network.

[0039] S5. Develop greenhouse crop irrigation strategies

[0040] Greenhouse environmental parameters are collected every minute throughout the day. The first irrigation is carried out after sunrise at the location of the greenhouse, and the irrigation amount is calculated according to the formula (1) where the water state is the appropriate state for irrigation;

[0041] The time interval between two crop water status detections is defined as the crop water status detection time interval, denoted as m, in minutes. In this embodiment, m=10. Every m minutes, the greenhouse environmental parameters and the RGB image, depth image and near infrared image of the crop canopy leaves are collected, and the above information is input into the crop water status detection model to determine the water status of the crop. If the crop water status is irrigation deficit, irrigation is immediately carried out, and the irrigation amount is calculated by formula (1), and the next crop water status detection is carried out after the crop water stress recovery time has passed.

[0042] If the crop water status does not show irrigation deficit (i.e., excessive irrigation or appropriate irrigation) within n hours after irrigation, the crop water status detection model is used to determine the crop water status n hours after irrigation and the fertigation machine is started for irrigation. The irrigation amount is calculated by formula (1):

[0043]

[0044] Where, Q is the irrigation volume per plant, unit: mL / plant;

[0045] Q pred The predicted value of transpiration per plant in the next n hours, unit: mL / plant;

[0046] Q' is the irrigation amount per plant during the last irrigation, unit: mL / plant;

[0047] ET is the transpiration of a single plant from the last irrigation to the current moment, in mL / plant. The calculation method is to first calculate the crop transpiration rate from the last irrigation to the current moment according to the crop transpiration rate prediction model, and then integrate the crop transpiration rate in this time period. The obtained value is the crop transpiration in this time period.

[0048] The predicted value of transpiration per plant Q predThe determination is made by the following method: the greenhouse environmental parameters of each minute in the first 24 hours, the weather forecast information at the hour in the next n hours, and the control parameter setting values ​​of the greenhouse environmental regulation system are input into the greenhouse environmental parameter prediction model established by S2 to obtain the greenhouse environmental prediction information in the next n hours; the crop planting date, current date, current time and greenhouse environmental prediction information are input into the crop transpiration rate prediction model established by S1 to calculate the crop transpiration rate; and the crop transpiration rate is integrated and calculated in the next n hours to obtain the crop transpiration amount prediction value Q pred .

[0049] The crop moisture status detection time interval of m minutes and the irrigation time interval of n hours are adjusted according to the crop type, and (n×60) can be divided by m.

[0050] S6, irrigating the crops according to the greenhouse crop irrigation strategy. In this example, the fertigation machine irrigates the crops according to the greenhouse crop irrigation strategy.

[0051] The embodiments are preferred implementations of the present invention, but the present invention is not limited to the above-mentioned implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A greenhouse crop irrigation method based on deep neural network, characterized in that: Every m minutes, the greenhouse environmental parameters and the RGB images, depth images, and near-infrared images of the crop canopy leaves are collected and input into the crop moisture status detection model to determine the moisture status of the crop; If the crop water status is irrigation deficit, irrigation is immediately carried out; and the next crop water status detection is carried out after the crop water stress recovery time has passed; If the crop water status does not show irrigation deficit within n hours after irrigation, the crop water status is determined by the crop water status detection model n hours after irrigation, and irrigation is performed; where n is the irrigation interval; The irrigation deficit is the crop water state where the irrigation level is less than F(1-P), the irrigation level with the highest net photosynthetic rate is defined as the irrigation benchmark, denoted by F; the allowed relative deviation is defined as the threshold, denoted by P; The crop water status when the irrigation level is greater than F(1+P) is judged as over-irrigation, and the crop water status in the interval [F(1-P), F(1+P)] is judged as appropriate irrigation; The irrigation amount is determined by the following formula: Where: Q is the irrigation amount per plant, Q pred is the predicted value of transpiration per plant in the next n hours, Q' is the irrigation amount per plant at the last irrigation, and ET is the transpiration per plant from the last irrigation to the current moment.

2. The greenhouse crop irrigation method based on deep neural network according to claim 1, characterized in that: The crop moisture status detection model specifically takes the planting date, current date, current time, greenhouse environmental parameters, and RGB images, depth images, and near-infrared images of crop canopy leaves as inputs, and takes the crop moisture status as output to construct a deep neural network model for crop moisture status detection.

3. The greenhouse crop irrigation method based on deep neural network according to claim 1, characterized in that: The crop water stress recovery time is predicted based on a crop water stress recovery time prediction model. Specifically, the crop water stress recovery time prediction model takes the crop planting date, current date, current time, greenhouse environmental parameters and irrigation amount as inputs and the crop water stress recovery time as output to establish a deep neural network model for predicting the crop water stress recovery time.

4. The greenhouse crop irrigation method based on deep neural network according to claim 1, characterized in that: The predicted value of transpiration per plant Q pred The acquisition method is as follows: input the greenhouse environmental parameters every minute of the first 24 hours, the weather forecast information at the hour in the next n hours, and the control parameter setting values ​​of the greenhouse environmental regulation system into the greenhouse environmental parameter prediction model to obtain the greenhouse environmental prediction information in the next n hours; input the crop planting date, current date, current time and greenhouse environmental prediction information into the crop transpiration rate prediction model to calculate the crop transpiration rate; integrate the crop transpiration rate in the next n hours to obtain the crop transpiration amount prediction value Q pred .

5. The greenhouse crop irrigation method based on deep neural network according to claim 4, characterized in that: The crop transpiration rate prediction model specifically takes the crop planting date, current date, current time and greenhouse environmental parameters as the input of the model, takes the crop transpiration rate as the output of the model, and constructs a deep neural network model for crop transpiration rate prediction.

6. The greenhouse crop irrigation method based on deep neural network according to claim 5, characterized in that: The greenhouse environmental parameter prediction model specifically takes the greenhouse environmental parameters every minute in the first 24 hours, the weather forecast information at the hour in the next n hours, and the control parameter setting values ​​of the greenhouse environmental regulation system as input, and takes the greenhouse environmental parameters every minute in the next n hours as output, to construct a deep neural network model for greenhouse environmental parameter prediction.

7. The greenhouse crop irrigation method based on deep neural network according to claim 1, characterized in that: The m and n satisfy that (n×60) is divisible by m.

Images