CN111034596B - Water culture flower rooting induction cultivation method based on reinforcement learning - Google Patents
Water culture flower rooting induction cultivation method based on reinforcement learning Download PDFInfo
- Publication number
- CN111034596B CN111034596B CN201911212757.6A CN201911212757A CN111034596B CN 111034596 B CN111034596 B CN 111034596B CN 201911212757 A CN201911212757 A CN 201911212757A CN 111034596 B CN111034596 B CN 111034596B
- Authority
- CN
- China
- Prior art keywords
- flowers
- root
- rooting induction
- root system
- hydroponic
- 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.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
- A01G31/00—Soilless cultivation, e.g. hydroponics
Abstract
The invention relates to a water culture flower rooting induction cultivation method based on reinforcement learning, which comprises the following steps: 1) constructing a root activity detection mechanism of the hydroponic flowers through a convolutional neural network; 2) the method comprises the following steps of modeling the rooting induction of the hydroponic flowers into a Markov chain with 4 discrete time stages, representing four stages of the rooting induction of the hydroponic flowers respectively: cutting roots, disinfecting roots, domesticating, and breeding roots in water; 3) and performing combined optimization on the 4 modeled Markov chains in discrete time stages to finally obtain a four-stage combined strategy. The invention has the beneficial effects that: the rooting induction cultivation strategy of the hydroponics flowers generated by the method is better than the common expert strategy, so that under the condition that common personnel do not have cultivation experience, the rooting induction cultivation method of the hydroponics flowers based on reinforcement learning can also realize the rooting induction cultivation of the hydroponics flowers, and the method has universality and can be popularized and applied to various crops.
Description
Technical Field
The invention relates to a rooting induction culture method for water-cultured flowers, in particular to a rooting induction culture method for water-cultured flowers based on reinforcement learning.
Background
In recent years, with the rapid development of economy in China and the continuous improvement of the living standard of people, the demands of people on living diversity and green environmental protection are increasingly strengthened, and water culture flowers are more and more attracted and favored by people as a novel type in ornamental plants. Compared with the traditional soil culture technology, the water culture flower is not limited by land, time and space, has the characteristics of strong ornamental value, convenient combination, cleanness, sanitation, convenient maintenance, various forms and the like, and has very wide development prospect. Although hydroponic flowers are considered to have great development potential, cultivation of the hydroponic flower cultivation technique in the rooting induction period is still insufficient, resulting in difficulty in widespread application in practice. At present, how to better perform rooting induction cultivation of water-cultured flowers becomes the focus of attention of people.
In the aspect of the rooting induction technology of the hydroponics flowers, most of the existing research works discuss the influence of single factors (such as the concentration of a rooting agent hormone, a root cutting mode, the pH value of a cultivation environment, temperature and illumination and the like) on the rooting induction of the hydroponics flowers, and the rooting of the hydroponics flowers is a complex dynamic physiological process of interaction between plants and the environment and is influenced by various factors, so that the rooting induction cultivation method optimized in combination cannot be obtained by the existing rooting induction technical scheme of the hydroponics flowers. In addition, the existing research mostly adopts single-factor experiments or orthogonal experiments to determine the rooting induction of the hydroponic flowers, has the defects of more experiment times and larger workload, and is not suitable for the rooting induction research of the hydroponic flowers with complicated experiment work and high cost.
Disclosure of Invention
The invention aims to overcome the defects and provide a water culture flower rooting induction cultivation method based on reinforcement learning.
The water culture flower rooting induction cultivation method based on reinforcement learning comprises the following steps:
1) constructing a root activity detection mechanism of the hydroponic flowers through a convolutional neural network to obtain the growth state of the root of the current hydroponic flowers;
2) the method comprises the following steps of modeling the rooting induction of the hydroponic flowers into a Markov chain with 4 discrete time stages, representing four stages of the rooting induction of the hydroponic flowers respectively: cutting roots, disinfecting roots, domesticating, and breeding roots in water;
3) and performing combined optimization on the Markov chains of the 4 discrete time stages of modeling to finally obtain a combined strategy of the four stages, wherein the combined strategy is used as an optimal scheme for rooting induction cultivation of the hydroponic flowers.
Preferably, the method comprises the following steps: in the step 1), the mechanism for detecting the root activity of the hydroponic flowers comprises the following steps:
1.1) defining the root activity of the hydroponic flowers as the sum of the root elongation of k straight roots before the length ranking of the hydroponic flowers, and recording as:
wherein RER represents the root activity of the hydroponic flowers and RERt-1,t(i) The calculation is shown in the following formula:
wherein, RERt-1,t(i) Representing the root elongation len of the hydroponic flower root system i from the date t-1 to the date tt(i) And lent-1(i) Respectively representing the length of the root system i measured on the date t-1 and the date t, Pt-1,tThe interval from date t-1 to date t;
1.2) detecting the root system edge of the water culture flowers; jointly mining the characteristics of image color, brightness, gradient and the like to carry out root system edge detection, dividing the root system image into a plurality of local slices by adopting an N4-Fields method, calculating the characteristics of each local slice based on a convolutional neural network, searching in a preset characteristic dictionary, searching for edges similar to the local slices, and integrating the similar edge information to obtain a root system edge detection result;
1.3) evaluating the root length of the water-cultured flowers; after root system edge detection, calculating the length of the root system by using the size of a reference object image; firstly, selecting a reference object based on the position or the unique color shape characteristic, calibrating the number of pixels in unit length by using the reference object, and recording as a 'pixel/measurement' ratio; secondly, comparing the root system image obtained by edge detection with a reference object image obtained by shooting at the same time, and estimating the length of the root system based on the pixel/measurement ratio;
1.4) calculating the root elongation sum of k straight roots before ranking according to the length of the root system, namely the activity of the root system of the water culture flowers.
Preferably, the method comprises the following steps: in step 3), the combined optimization of the modeled markov chains in 4 discrete time phases includes the following steps:
3.1) determining the optimized targets as the root activity RER and the root color C of the hydroponic flowers;
3.2) adopting a Model-free reinforcement learning (Model-free reinforcement learning) method to solve the Markov chain decision problem.
Preferably, the method comprises the following steps: in step 3.2), solving the markov chain decision problem comprises the following steps:
3.2.1) obtaining a cultivation state target formula of rooting induction according to a limited time domain dynamic programming method:
wherein
SiAnd DiRespectively representing a state variable space and a decision variable space of the rooting induction stage i, wherein the state s in the state variable space is represented by a two-dimensional vector to be ═ RER, C]，P(s′I s, d) represents the state s applying the decision d to transition to the state s′Probability of ri(s,d,s′) Indicating the transition of the application decision d to the state s′Local reward of (1);
according to the state objective formula, the combined optimization strategy for the rooting induction of the hydroponics flowers can be represented by the following formula:
3.2.2) the state transition probability and the local return in the combined optimization strategy formula are unknown, the state transition continuously changes during decision optimization, and a reinforced learning algorithm DDPG (deep dependent Policy gradient) Based on strategy (Policy-Based) is adopted to solve the problem.
Preferably, the method comprises the following steps: in step 3.2.2), four neural networks are trained by adopting a strategy-based reinforcement learning algorithm DDPG, and the optimized strategy pi is obtained through network convergence in the updating modeiFinally, a combination strategy pi of four stages is obtained1，π2，π3，π4And the method is the optimal scheme for rooting induction culture of the water culture flowers.
The method has the advantages that the rooting induction cultivation strategy of the hydroponic flowers generated by the method is better than the common expert strategy, so that under the condition that common personnel do not have cultivation experience, the rooting induction cultivation method of the hydroponic flowers based on reinforcement learning can realize the rooting induction cultivation of the hydroponic flowers, has universality and can be popularized and applied to various crops.
Drawings
FIG. 1 is a flow chart of rooting induction culture of water-cultured flowers;
FIG. 2 is a flow chart of the detection of the root activity of the hydroponic flowers;
FIG. 3 is a schematic diagram of a convolutional neural network-based root activity detection model;
FIG. 4 is a flow chart of rooting induction culture of water-cultured flowers based on reinforcement learning;
FIG. 5 is a schematic diagram of a rooting induction culture model of water-cultured flowers based on reinforcement learning;
FIG. 6 is a diagram of the experimental simulation result of rooting induction culture of water-cultured flowers based on reinforcement learning.
Detailed Description
The present invention will be further described with reference to the following examples. The following examples are set forth merely to aid in the understanding of the invention. It should be noted that, for those skilled in the art, it is possible to make various improvements and modifications to the present invention without departing from the principle of the present invention, and those improvements and modifications also fall within the scope of the claims of the present invention.
First, the overall idea of the invention:
firstly, analyzing the actual situation of the root system of the water culture flowers through a convolutional neural network, and establishing a root system activity detection mechanism; and then constructing a Markov chain decision simulation Model according to the obtained root system data, and solving the Markov chain decision problem by using Model-free reinforcement learning (Model-free reinforcement learning) method with the root system activity and the root system color of the water culture flowers as optimization targets to obtain an optimal water culture flower rooting induction scheme.
Secondly, the rooting induction cultivation scheme of the hydroponic flowers comprises the following steps (as shown in figure 1):
1. the detection mechanism of the root activity of the hydroponic flowers is constructed by a convolutional neural network (as shown in figure 2)
Defining the root activity of the hydroponic flowers as the sum of the root elongation of k straight roots before the length ranking of the hydroponic flowers, and recording as:
wherein, RERt-1,t(i) The calculation is shown in the following formula:
wherein, RERt-1,t(i) Representing the root elongation len of the hydroponic flower root system i from the date t-1 to the date tt(i) And lent-1(i) Respectively representing the length of the root system i measured on the date t-1 and the date t, Pt-1,tThe interval from date t-1 to date t.
Regularly shooting the hydroponic flower root system by using a high-resolution digital camera, calculating the length of the root system, and comprising the following steps of:
1) and detecting root system edges. Aiming at the phenomena that fine roots in a water culture flower root system image are irregular and have inclination angles generally, the root system edge detection is carried out by jointly mining the characteristics of the image such as color, brightness, gradient and the like, an N4-Fields method is adopted to divide the root system image into a plurality of local slices, as shown in figure 3, the characteristics of each local slice are calculated based on a convolutional neural network, retrieval is carried out in a preset characteristic dictionary, edges similar to the local slices are searched, and the similar edge information is integrated to obtain a root system edge detection result;
2) and (5) evaluating the root system length. After root system edge detection, the root system length is calculated using the size of the reference object image. Firstly, selecting a reference object based on the position or the unique color shape characteristic, calibrating the number of pixels in unit length by using the reference object, and recording as a 'pixel/measurement' ratio; and secondly, comparing the root system image obtained by edge detection with a reference object image shot at the same time, and estimating the length of the root system based on the pixel/measurement ratio.
And calculating the root elongation sum of the front k straight roots according to the estimated length, namely the root activity of the hydroponic flowers.
2. A method for carrying out scheme optimization on the basis of reinforcement learning and inducing rooting of hydroponic flowers (as shown in figure 4) comprises the following steps:
1) the method comprises the following steps of modeling the rooting induction of the hydroponic flowers into a Markov chain with 4 discrete time stages, representing four stages of the rooting induction of the hydroponic flowers respectively:
firstly, cutting roots. After the flowers needing water culture are subjected to soil removal and root washing, root cutting treatment is carried out on the roots. Wherein, the root cutting degree is 0 to indicate that the root is not cut, and 1 to indicate that the root cutting treatment is carried out completely;
② sterilizing the root system. The root system of the soil-borne flowers has a plurality of germs, which easily causes the infection at the cut and needs to disinfect the root;
and thirdly, domestication. The rooting induction domestication is to artificially control the whole physiological and morphological development process of a root system and induce the formation of an aeration tissue under the anaerobic action. The domestication process is that the disinfected flower roots are put into a rooting agent solution for root promotion treatment, and then the root promotion is carried out on a root promoting seedbed;
and fourthly, cultivating roots by water. The purpose of the water breeding root is to enable the domesticated flowers to adapt to the growth of the water environment, wherein the oxygen concentration of the nutrient solution is the key to whether the water breeding root is successful.
2) And performing combined optimization on the decisions of the four stages, wherein the steps are as follows:
determining an optimization target:
I. root activity RER. The root activity is defined as the sum of the root elongation rates of k straight roots before the length ranking of the cultivated flowers after the domestication is finished and the aquatic root cultivation is finished;
root color C, with a larger value indicating a lighter root color. The lighter the color of the root system is, the better the water adaptability of the root system is, the more developed the ventilating tissue is, and the judgment is carried out through the root system image.
Secondly, a Model-free reinforcement learning (Model-free reinforcement learning) method is adopted to solve the Markov chain decision problem.
The cultivation optimization goal of the rooting induction of the water culture flowers is to find a group of four stagesCombined strategy of pi ═ pi1，π2，π3，π4Maximizing the optimization goal of rooting induction, and formally expressing: vΠE (RER + C | Π). According to the finite time domain dynamic programming method, the cultivation target of rooting induction is shown as the following formula:
wherein
SiAnd DiRespectively representing a state variable space and a decision variable space of the rooting induction stage i, wherein the state s in the state variable space is represented by a two-dimensional vector to be ═ RER, C]P (s '| s, d) represents the probability that state s applies decision d to transition to state s', ri(s, d, s ') represents the local return of applying decision d to transition to state s'.
According to the formula, the combination optimization strategy for the rooting induction of the hydroponic flowers can be obtained by the following formula:
thirdly, the formula is realized by adopting a Policy-Based reinforcement learning algorithm
Since the state transition probability and the local return in the formula are unknown, considering that the state transition is continuous change in decision optimization, if a Value-Based reinforcement learning algorithm is adopted, the problem of excessive states occurs.
Therefore, a Policy-based reinforcement learning algorithm ddpg (deep Deterministic Policy gradient) is adopted to solve the problem, and the implementation framework is shown in fig. 5.
Two neural networks are respectively created for the strategy network and the target network in the solving process, wherein one neural network is an online learning network (online) and the other neural network is a target optimization network (target).
The policy network is as follows:
the Q network is as follows:
by such updating, the network converges faster to obtain an optimized strategy pii。
And then, sampling learning is carried out from the obtained experience database, so that the selection correlation of each step can be reduced, better learning is carried out, and finally the combination strategy pi of the four stages is obtained1，π2，π3，π4The method is the optimal scheme of rooting induction culture of the water culture flowers.
Thirdly, verifying the result:
in order to verify the effect of the method, a simulation experiment is carried out on an open-source crop simulation platform provided by the university of wageninggen, plant leaf area index LAI is taken as an optimization target during simulation, a reinforcement learning algorithm DDPG is used for optimizing the plant leaf area index LAI, training is continuously carried out until the algorithm converges to obtain an optimized strategy, and the results of cultivation by the optimized strategy and a common strategy are shown in FIG. 6.
Fourthly, experimental conclusion:
the experimental results show that the rooting induction cultivation strategy of the hydroponic flowers generated by the method is better than the common expert strategy, so that under the condition that common personnel do not have cultivation experience, the rooting induction cultivation strategy of the hydroponic flowers obtained by calculation can also realize the rooting induction cultivation of the hydroponic flowers, and the method has universality and can be popularized and applied to various crops.
Claims (2)
1. A water culture flower rooting induction cultivation method based on reinforcement learning is characterized by comprising the following steps:
1) constructing a root activity detection mechanism of the hydroponic flowers through a convolutional neural network to obtain the growth state of the root of the current hydroponic flowers;
2) the method comprises the following steps of modeling the rooting induction of the hydroponic flowers into a Markov chain with 4 discrete time stages, representing four stages of the rooting induction of the hydroponic flowers respectively: cutting roots, disinfecting roots, domesticating, and breeding roots in water;
3) performing combined optimization on the Markov chains of the 4 discrete time stages of modeling to finally obtain a combined strategy of the four stages, wherein the combined strategy is used as an optimal scheme for rooting induction cultivation of the hydroponic flowers;
in the step 1), the mechanism for detecting the root activity of the hydroponic flowers comprises the following steps:
1.1) defining the root activity of the hydroponic flowers as the sum of the root elongation of k straight roots before the length ranking of the hydroponic flowers, and recording as:
whereinRERThe activity of the root system of the water culture flowers is shown,RER t-1,t (i)the calculation is shown in the following formula:
wherein the content of the first and second substances,RER t-1,t (i)indicating hydroponic flower root systemiFrom datet-1Date of arrivaltThe root elongation of (a) is high,len t (i)andlen t-1 (i)respectively represent root systemsiOn the datetDate and timet-1The length of the measured length is measured,P t-1,t is a datet-1Date of arrivaltThe interval of (a);
1.2) detecting the root system edge of the water culture flowers; jointly mining image color, brightness and gradient characteristics to carry out root system edge detection, dividing a root system image into a plurality of local slices by adopting an N4-Fields method, calculating the characteristics of each local slice based on a convolutional neural network, searching in a preset characteristic dictionary, searching for edges similar to the local slices, and integrating the similar edge information to obtain a root system edge detection result;
1.3) evaluating the root length of the water-cultured flowers; after root system edge detection, calculating the length of the root system by using the size of a reference object image; firstly, selecting a reference object based on the position or the unique color shape characteristic, calibrating the number of pixels in unit length by using the reference object, and recording as the ratio of pixels to measurement; secondly, comparing the root system image obtained by edge detection with a reference object image obtained by shooting at the same time, and estimating the length of the root system based on the ratio of pixels to measurement;
1.4) calculating the root elongation sum of k straight roots before ranking according to the length of the root system, namely the activity of the root system of the water culture flowers.
2. The reinforced learning-based rooting induction culture method for hydroponics flowers according to claim 1, characterized in that: in step 3), the combined optimization of the modeled markov chains in 4 discrete time phases includes the following steps:
3.1) determining the optimized target as the root activity of the water-cultured flowersRERAnd root color C;
3.2) solving the Markov chain decision problem by adopting a model-free reinforcement learning method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911212757.6A CN111034596B (en) | 2019-12-02 | 2019-12-02 | Water culture flower rooting induction cultivation method based on reinforcement learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911212757.6A CN111034596B (en) | 2019-12-02 | 2019-12-02 | Water culture flower rooting induction cultivation method based on reinforcement learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111034596A CN111034596A (en) | 2020-04-21 |
| CN111034596B true CN111034596B (en) | 2021-06-18 |
Family
ID=70234371
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911212757.6A Active CN111034596B (en) | 2019-12-02 | 2019-12-02 | Water culture flower rooting induction cultivation method based on reinforcement learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111034596B (en) |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8160248B2 (en) * | 2009-04-02 | 2012-04-17 | Broadcom Corporation | Authenticated mode control |
| CN102823480A (en) * | 2012-09-11 | 2012-12-19 | 江苏省农业科学院宿迁农科所 | Terrestrial flower water culture domestication method |
| JP2016036329A (en) * | 2014-08-06 | 2016-03-22 | 高嶋 康豪 | Full recycling hydroponic (hydroponic soil culture) system using composite fermentation of composite microorganisms |
| US20190050742A1 (en) * | 2017-12-29 | 2019-02-14 | Intel Corporation | Compatibility prediction technology in shared vehicles |
| WO2019163057A1 (en) * | 2018-02-22 | 2019-08-29 | ヤンマーグリーンシステム株式会社 | Cultivation system and cultivation method |
| CN110348562B (en) * | 2019-06-19 | 2021-10-15 | 北京迈格威科技有限公司 | Neural network quantization strategy determination method, image identification method and device |
-
2019
- 2019-12-02 CN CN201911212757.6A patent/CN111034596B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111034596A (en) | 2020-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lososová et al. | Biotic homogenization of Central European urban floras depends on residence time of alien species and habitat types | |
| Morton et al. | Plant–pollinator interactions under climate change: The use of spatial and temporal transplants | |
| US20220075344A1 (en) | A method of finding a target environment suitable for growth of a plant variety | |
| CN106373022B (en) | BP-GA-based greenhouse crop planting efficiency condition optimization method and system | |
| CN106292802A (en) | A kind of Intelligent Prediction Control System for fish and vegetable symbiotic system and method | |
| Kolmanič et al. | An algorithm for automatic dormant tree pruning | |
| CN107103554A (en) | The intelligent monitoring method and system of a kind of milk cow | |
| CN113158750A (en) | Self-feedback learning evaluation method of plant growth model based on convolutional neural network | |
| CN110414115A (en) | A kind of wavelet neural network tomato yield prediction technique based on genetic algorithm | |
| CN115529987A (en) | Air port regulating and controlling method, device, equipment and storage medium for crop facility | |
| KR20200044604A (en) | Smart Farm Control System Using Smart Farm Control History | |
| Cao et al. | Recognition of common insect in field based on deep learning | |
| CN108181814A (en) | Plant growth environment monitoring method and device, computer storage media | |
| Spencer et al. | Estimating Arundo donax ramet recruitment using degree-day based equations | |
| US11864552B2 (en) | Digital detection method and system for predicting drug resistance of transgenic maize | |
| CN113435825B (en) | Intelligent management method, system and storage medium based on soil-borne disease control | |
| CN114637351A (en) | Greenhouse environment regulation and control method and system for facility crops | |
| CN111034596B (en) | Water culture flower rooting induction cultivation method based on reinforcement learning | |
| CN112861655A (en) | Point cloud model plant and analysis algorithm for pest and disease damage identification and early warning thereof | |
| CN112783228A (en) | Neural network-based large-scale greenhouse PD control system and method | |
| CN113010529A (en) | Crop management method and device based on knowledge graph | |
| ROMDHONAH et al. | Empirical model for the estimation of whole-plant photosynthetic rate of cherry tomato grown in a commercial greenhouse | |
| Chen et al. | A water-saving irrigation decision-making model for greenhouse tomatoes based on genetic optimization TS fuzzy neural network | |
| CN115643956A (en) | High-efficiency comprehensive prevention and control method for gynostemma pentaphylla nematode | |
| CN108244003A (en) | A kind of aquatic products plant disease epidemic trend prediction and methods of exhibiting and system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220721 Address after: 310015 No. 51, Huzhou street, Hangzhou, Zhejiang Patentee after: Zhejiang University City College Address before: 310015 No. 50 Huzhou Street, Hangzhou City, Zhejiang Province Patentee before: Zhejiang University City College |