Deep learning-based intelligent breeding collaborative optimization system and dynamic decision-making method

Abstract

The invention relates to the technical field of data recognition, in particular to an intelligent breeding collaborative optimization system based on deep learning and a dynamic decision-making method. The method comprises the steps of collecting characteristics of plants based on a collection device; screening an abnormal period of the plant according to the collected characteristics and the abnormality of other plants in the period; determining the self-influence index of the plant according to the characteristic anomaly continuous condition in the historical period and the number of abnormal plants in combination with the abnormal conditions of other plants of the plant in the current period; determining an excellent index based on the actual and standard difference of the plant growth stage, the number of abnormal periods and the self-influence plant, and obtaining the collection frequency; and acquiring data based on the acquisition frequency, and acquiring a recommendation index in combination with the excellent index to complete breeding. According to the invention, the recognition precision of the periodic growth stage is improved.

Description

Technical Field

[0001] This application relates to the field of data recognition technology, and specifically to an intelligent breeding collaborative optimization system and dynamic decision-making method based on deep learning. Background Art

[0002] In the process of intelligent plant breeding, accurate identification of the plant's growth cycle is crucial to breeding efficiency. Taking cotton as an example, accurate identification of the key growth stages of cotton is conducive to screening the target traits of cotton plants in advance, screening out invalid groups, and thus shortening the breeding cycle.

[0003] Currently, the combination of drones or ground robots with deep learning methods is a key means in field monitoring and precision breeding. For example, in the process of smart cotton breeding, drones are usually used to carry out hyperspectral imaging combined with RGB cameras on ground robots to obtain cotton field images, and the obtained image data is input into a convolutional neural network to identify the growth stage of cotton, thereby improving the efficiency of smart cotton breeding.

[0004] In the actual detection process, when using existing methods to identify the growth stage of plants, there is a situation where the phenotypic characteristics of plants in adjacent growth stages overlap, making it impossible to correctly determine whether it is a change in growth stage or an identification error, resulting in misjudgment or multiple fluctuations in the identification results. Summary of the Invention

[0005] In order to solve the technical problem of misjudgment caused by misjudgment of growth stages, this application provides an intelligent breeding collaborative optimization system and dynamic decision-making method based on deep learning. The technical solutions adopted are as follows:

[0006] In the first aspect, this application proposes a dynamic decision-making method for intelligent breeding based on deep learning, which includes the following steps:

[0007] Characteristics of plants collected based on the collection device;

[0008] Cluster the plants within a cycle based on all their characteristics; obtain the characteristic outlier value of each characteristic of the plant within a cycle based on the proportion of the number of clusters to which the plants belong and the difference between the plant characteristics and the overall characteristics of all plants within the cycle; and filter out abnormal cycles based on the characteristic outlier values;

[0009] The plants with abnormalities within a cycle are merged to obtain the abnormal area; the long-term abnormal value is determined based on the maximum continuous abnormal value of each plant's characteristics in all cycles before each cycle; the number of plant abnormalities is obtained based on the number of plants in the maximum abnormal area in all cycles before each cycle; the number of abnormalities is obtained based on the number of times all the characteristics of the plant are abnormal; the plant's own influence index is calculated based on the number of plant abnormalities, long-term abnormal value and number of plant abnormalities;

[0010] The standard duration of the plant growth stage is preset, and the plant's excellent index in each cycle is obtained based on the difference between the actual duration and the standard duration of all growth stages before each cycle, the number of abnormal cycles, and its own impact index; based on the excellent index of the previous cycle, the sampling frequency of the previous cycle is adjusted to obtain the sampling frequency of the current cycle;

[0011] According to the collection frequency of the current cycle, the inspection route is adjusted to collect plant data, the plant growth rate is obtained based on the plant data, and the recommended index is obtained in combination with the plant's excellent index. The recommended index is used to screen excellent plants to complete breeding.

[0012] In the above scheme, this application monitors the characteristics of the plants, first screens for abnormal cycles, and then analyzes the differences in plant characteristics in different regions to determine whether the abnormalities are caused by internal factors or external factors, thereby solving errors caused by self-influences, and adjusting the data collection frequency of excellent plants based on the comparison results with the standard growth cycle, thereby improving the recognition accuracy of its cyclical growth stage.

[0013] In one embodiment, the plant characteristics include plant height, stem diameter, and NDVI value.

[0014] In one embodiment, the clustering adopts the k-means clustering method, the cluster distance is the Euclidean distance between data points, and the cluster number is the ratio of the number of plants to the maximum cluster distance rounded down.

[0015] In one embodiment, the characteristic outlier is negatively correlated with the proportion of the number of data points in the cluster where the plant is located, and is negatively correlated with the difference between the plant characteristics and the overall characteristics of all plants in the period.

[0016] In one embodiment, the abnormal period is a period in which the plant has one or more abnormal characteristics; the abnormal characteristics are characteristics in which the abnormal value of the characteristic is greater than the abnormal threshold.

[0017] In one embodiment, the self-influence index is positively correlated with the long-term abnormal value, and negatively correlated with the abnormal times and the abnormal quantity of the plants.

[0018] In one embodiment, the method for obtaining the number of long-term abnormal values ​​and the number of abnormal plants is as follows:

[0019] For a feature of a plant, the features of all cycles before each cycle are combined into an abnormal sequence in time sequence; the feature abnormal values ​​in the abnormal sequence that are greater than the abnormal threshold are marked; the maximum continuous number of all marks is counted; and the maximum continuous number in the abnormal sequence of all features is recorded as a long-term abnormal value;

[0020] Count the number of plants in the abnormal area corresponding to each feature of the plant, and take the maximum value as the maximum abnormal area of ​​the plant; calculate the average of the maximum abnormal areas of all cycles before each cycle as the number of plant abnormalities;

[0021] The area where the abnormal areas corresponding to all the characteristics of a plant overlap is recorded as the core abnormal area; the number of times the core abnormal area exists in the plant is counted and recorded as the abnormal number.

[0022] In one embodiment, the good index is positively correlated with the difference between the actual duration of the growth phase and the standard duration, and negatively correlated with the number of abnormal cycles and the self-influence index.

[0023] In one embodiment, the method for adjusting the sampling frequency of the previous cycle based on the excellence index of the previous cycle to obtain the sampling frequency of the current cycle is:

[0024] p T+1,i =p T,i ×(1+v T,i ), v T,i represents the quality index of the i-th plant in the T-th cycle; p T,i represents the sampling frequency of the i-th plant in the T-th period, p T+1,i represents the sampling frequency of the i-th plant in the T+1-th cycle. The initial sampling frequency is once per cycle.

[0025] On the other hand, an embodiment of the present application also provides an intelligent breeding collaborative optimization system based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-mentioned intelligent breeding dynamic decision-making methods based on deep learning.

[0026] The beneficial effects of this application are:

[0027] This application monitors the characteristics of plants, first screening for abnormal cycles, and then analyzing the differences in plant characteristics in different regions to determine whether the abnormalities are caused by internal factors or external factors, thereby resolving errors caused by internal influences, and adjusting the data collection frequency of excellent plants based on the comparison results with the standard growth cycle, thereby improving the recognition accuracy of their growth cycles. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0029] Figure 1 This is a flow chart of a deep learning-based intelligent breeding dynamic decision-making method provided in one embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to further illustrate the technical means and effects adopted by this application to achieve the predetermined invention purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation methods, structures, features and effects of a smart breeding collaborative optimization system and dynamic decision-making method based on deep learning proposed in this application. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics of one or more embodiments may be combined in any suitable form.

[0031] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0032] Examples of intelligent breeding collaborative optimization system and dynamic decision-making method based on deep learning:

[0033] The following describes in detail a specific scheme of a dynamic decision-making method for intelligent breeding based on deep learning provided by this application with reference to the accompanying drawings.

[0034] See also Figure 1 , which shows a flow chart of a dynamic decision-making method for intelligent breeding based on deep learning provided by an embodiment of the present application, the method comprising the following steps:

[0035] Step S001: collecting plant characteristics using a collection device.

[0036] The drone and ground robot collaborate to monitor plant characteristics. Each monitoring cycle is one day, with plant monitoring performed during a fixed time period each day. In this example, the fixed time period is from 12:00 PM to 2:00 PM. The collection frequency within the fixed time period is set. The initial collection frequency is once, meaning that the drone and ground robot perform one inspection within the fixed time period.

[0037] The characteristics of the plants are obtained through inspections by drones and ground robots. In this embodiment, the characteristics include plant height, stem diameter, and NDVI value.

[0038] UAV-mounted LiDAR generates point cloud data. UAV-mounted hyperspectral cameras are used during inspections to monitor plant diseases and perform image preprocessing (including radiation correction). Orthophotos are generated from IMU data, preserving spatial detail at the individual plant level. Individual plants are segmented using LiDAR point cloud data to obtain the plant height H. The NDVI (normalized difference vegetation index) for each pixel is obtained from the hyperspectral detection results. The NDVI for a single plant is calculated as the average of the NDVIs for all pixels in the corresponding area (the larger the value, the less likely the plant is to suffer from pests and diseases). A laser caliper mounted on a ground robot is used to collect plant stem diameters.

[0039] At this point, the characteristics of the plant were collected.

[0040] Step S002, screening abnormal cycles of plants by using the collected features and combining them with the abnormalities of other plants in the cycle.

[0041] Due to differences in disease resistance between plants, soil fertility, pH, temperature, and other factors, plants in the same area may have different characteristics. For example, plants at the edge of an area may receive more sunlight, absorb more nutrients, and have access to more water during irrigation, resulting in taller plants and thicker stems than other plants. Therefore, by comparing these differences in characteristics between plants, we analyze abnormalities in these characteristics. These abnormalities refer to situations where a plant has a problem.

[0042] For each cycle, a sample space is established using the data measured therein, wherein each characteristic of the collected plants corresponds to a dimension, wherein the positive direction of the dimension is the direction in which the characteristic data corresponding to the dimension increases.

[0043] All plants in each period were clustered using a clustering algorithm, with each plant serving as a data point. In this embodiment, the clustering algorithm used was k-means clustering, the cluster distance was the Euclidean distance between data points, and the number of clusters was the ratio of the number of plants to the maximum cluster distance, rounded down.

[0044] If a feature corresponding to each plant is significantly different from the same feature of the other plants, it means that an anomaly is more likely to occur. The smaller the proportion of data points in the cluster where the plant is located, the more abnormal the cluster is. Based on this, the characteristic anomaly value of each feature of the plant is obtained.

[0045] The characteristic outlier is negatively correlated with the proportion of the number of data points in the cluster where the plant is located, and is negatively correlated with the difference between the plant characteristics and the overall characteristics.

[0046] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large; the specific relationship is determined by actual application and this application does not impose any special restrictions.

[0047] It should be noted that negative correlation means that when one variable increases, the other variable decreases accordingly, and the two variables change in opposite directions. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small. The specific relationship is determined by actual application and this application does not impose any special restrictions.

[0048] Preferably, in this embodiment, the expression of the characteristic abnormal value is:

[0049] n T,j (i) represents the number of data points in the cluster corresponding to the i-th data point in the T-th period, n T represents the number of all data points in the Tth period, H T (i) represents the characteristic abnormal value of the i-th data point in the T-th period, represents the mean of the characteristic outliers in the cluster where the i-th data point is located in the T-th period, δ T,i Represents a feature outlier of the i-th data point in the T-th period.

[0050] The above calculations are performed for each feature of each data point to obtain the feature anomaly value for each feature of each data point. The feature anomaly values ​​are normalized using a normalization method. Plants with normalized values ​​greater than the anomaly threshold are designated as feature anomaly plants, and their corresponding features are designated as abnormal features. For example, if the plant height is abnormal, it is designated as a plant with abnormal plant height. In this embodiment, the anomaly threshold is set to 0.5.

[0051] If a plant has at least one abnormal characteristic in a cycle, the cycle is recorded as the abnormal cycle of the plant; that is, if the plant is a plant with abnormal characteristics in the cycle, the cycle is an abnormal cycle.

[0052] At this point, the characteristic abnormal values ​​of different characteristics of the plants in different periods have been obtained and the abnormal periods of the plants have been confirmed.

[0053] Step S003, determining the plant's own impact index based on the continuity of characteristic abnormalities in historical cycles and the number of abnormal plants combined with the abnormal conditions of other plants in the current cycle.

[0054] When a plant abnormality is caused by external factors, multiple abnormalities are often present in the corresponding area. For example, uneven soil fertility or water distribution in a region can cause all plants in that area to experience temporary abnormalities (with the degree of abnormality changing rapidly). This can lead to multiple abnormalities in the corresponding area. However, when a plant's abnormality is caused by internal factors, only a single plant may experience abnormalities. This can be caused by minor genetic mutations.

[0055] Therefore, the causes of abnormalities were distinguished based on the difference in the degree of abnormality between a single plant and the surrounding plants.

[0056] In any cycle, for any feature, the abnormal plants with adjacent features corresponding to the feature are merged, and the merged area is recorded as the abnormal area; multiple abnormal areas are obtained for each feature; if a plant has abnormalities in all features, that is, the plant corresponds to multiple abnormal areas, one of which corresponds to an abnormal area; the area where the abnormal areas corresponding to all features of a plant overlap is recorded as the core abnormal area; if the plant has features but no abnormal area, there is no core abnormal area.

[0057] If the characteristics of a plant are in an abnormal state for a long time, it is believed that it is more likely to be affected by itself; if there are fewer abnormalities in multiple characteristics of a plant, it means that it is more likely to be affected by itself; if the number of plants in the abnormal area is small, it means that it is less affected by the environment and more likely to be affected by itself.

[0058] Based on the above analysis, calculate your own impact index.

[0059] For each characteristic of any plant, the characteristic outlier values ​​for that characteristic in all cycles preceding the current cycle are counted and an anomaly sequence is constructed in chronological order, i.e., each characteristic corresponds to one anomaly sequence. In the anomaly sequence, characteristic outlier values ​​greater than an anomaly threshold are marked, and the maximum number of consecutive outliers across all marked characteristics is counted. This maximum consecutive outlier value across all characteristic anomaly sequences is recorded as the long-term outlier value. The larger the value, the more abnormal the characteristic of the plant has been for a long time. In this embodiment, the anomaly threshold is 0.5.

[0060] In a cycle, a plant may have multiple abnormal characteristics, so it may have multiple abnormal areas. The number of plants in the corresponding abnormal area is counted, and the maximum value is taken as the maximum abnormal area of ​​the plant. For all cycles before the current cycle, the average of the maximum abnormal areas of all cycles is calculated and recorded as the number of abnormal plants. The smaller the value, the fewer plants in the abnormal area and the greater the impact of the plant itself.

[0061] In all cycles before the current cycle, the number of times the plant has core abnormal areas is counted and recorded as the number of abnormalities. The smaller the number, the fewer times all the characteristics of the plant are abnormal at the same time, and the greater the impact it is on itself.

[0062] Therefore, the plant's self-influence index is calculated based on the plant's abnormality frequency, long-term abnormality value, and plant abnormality quantity. The self-influence index is positively correlated with the long-term abnormality value and negatively correlated with the plant's abnormality frequency and plant abnormality quantity.

[0063] Preferably, in this embodiment, the expression of the self-influence index is:

[0064] represents the number of abnormal plants in the ith plant at the Tth cycle, c T,i represents the number of abnormalities of the i-th plant in all cycles before the T-th cycle, L T,i represents the long-term abnormal value of the i-th plant in the T-th period, u T,i Represents the self-influence index of the i-th plant in the T-th period.

[0065] Normalize the self-influence index. If the normalized value is less than the influence threshold, the plant in the current cycle is marked as a plant whose current abnormal state is affected by external factors and is recorded as an environmentally abnormal plant.

[0066] At this point, the self-influence index of each plant in each cycle was obtained, and the degree to which each plant was affected by the environment was determined.

[0067] Step S004, based on the difference between the actual and standard growth stages of the plant, the number of abnormal cycles and the self-influence of the plant, a good index is determined, and a collection frequency is obtained.

[0068] Plant growth is a continuous process. Adjacent growth stages may result in overlapping features, blurring the CNN classification boundaries. This can lead to misjudgments and even multiple fluctuations in the recognition results when identifying the plant's growth stage. Therefore, to improve the accuracy of plant growth stage recognition, it is necessary to appropriately increase the data collection frequency for the corresponding plant when two recognition results are inconsistent, thereby improving the accuracy of identifying the plant's growth state. For example, the bud and flowering stages of cotton plants are immediately followed by the flowering stage.

[0069] Among them, the better the growth status of the plant, the more conducive it is to improving breeding efficiency, and the more accurate the identification of its growth stage needs to be improved. Therefore, it is necessary to evaluate the quality of each plant based on its growth status.

[0070] Through the above steps, it can be seen that the growth status of the plant is affected by many factors (such as the external environment and its own genes), which makes the growth cycle of a single plant different from the standard state. Therefore, based on the abnormality of the plants in each monitoring cycle obtained in the above steps, combined with the difference between the growth cycle of each plant and the standard growth cycle, the quality of each plant is evaluated.

[0071] Each plant undergoes CNN recognition once per cycle. The input to CNN recognition is each plant's features, and the output is the plant's growth stage. If the CNN recognition results differ between two consecutive cycles, two scenarios may occur: the first is a correct recognition, indicating that the two cycles are at different growth stages; the second is a misidentification, indicating that the two cycles are at the same growth stage but different stages are identified. To address inaccurate recognition, the acquisition frequency is adjusted to collect data multiple times within the same cycle to improve recognition accuracy.

[0072] As of the current cycle, if a single plant has experienced fewer abnormal cycles and is less likely to have inherent abnormalities in the current cycle (i.e., the likelihood that the plant's phenotypic abnormalities are affected by its own factors is lower), and the duration of the plant's growth phase is shorter than the standard duration, then the plant is considered to be of higher quality. The standard duration is the average growth duration for this type of plant at this growth stage. This is set based on the growth stages of different plants; in this example, the flowering and boll-forming stage of cotton is used as an example, and its duration is 60 days.

[0073] For a plant, if it reaches the next stage in multiple consecutive growth stages, it means that it has indeed reached the next stage. At this time, the length of all growth stages of the plant before the cycle is counted to obtain the length of the plant in the previous growth stage.

[0074] For a plant, the difference between the duration of each growth stage before the current cycle and the standard duration of the corresponding growth stage is taken as the duration difference of each growth stage;

[0075] The quality index of each plant in the current cycle is obtained based on the difference in the duration of all growth stages before the current cycle, the number of abnormal cycles and its own influence index.

[0076] The excellent index is positively correlated with the difference in the duration of all growth stages, and negatively correlated with the number of abnormal cycles and the self-influence index.

[0077] Preferably, in this embodiment, the expression of the good index is:

[0078] u T,i represents the self-influence index of the i-th plant in the T-th period, d T,irepresents the number of abnormal cycles of the i-th plant before the T-th cycle, t T,i represents the sum of the duration differences of all growth stages of the i-th plant before the T-th cycle, sigmoid() represents the normalization function, v T,i It represents the quality index of the i-th plant in the T-th cycle.

[0079] Based on the quality index of each plant in the current cycle, the sampling frequency of the next cycle can be adjusted as follows:

[0080] p T+1,i =p T,i ×(1+v T,i ), v T,i represents the quality index of the i-th plant in the T-th cycle; p T,i represents the sampling frequency of the i-th plant in the T-th period, p T+1,i represents the sampling frequency of the i-th plant in the T+1 period. The initial sampling frequency is once per period. The rounding method is used to calculate the sampling frequency of the next period. For example, p T+1,i If it is 1.5, the next cycle will collect data twice. T+1,i If it is 1.4, the data will be collected once in the next cycle.

[0081] At this point, the acquisition frequency of the next cycle is obtained through the acquisition frequency of the previous cycle.

[0082] Step S005: collect data based on the collection frequency, and obtain the recommended index in combination with the excellent index to complete the breeding.

[0083] After adjusting the collection frequency, the drone and ground robot need to be adjusted, and different collection frequencies need to be used for different plants.

[0084] The collection frequency is used as the pheromone, and the pheromone is input into the ant colony algorithm to obtain the optimal path for the inspection of drones and ground robots; data is collected once during each inspection, and the next inspection (for example, the w+1th inspection) is carried out after the collection. Among them, the nodes that do not need to be inspected for the w+1th time are deleted, and their inspection paths are planned using the ant colony algorithm center.

[0085] The growth curve model was used to obtain the growth curve of each plant and its growth rate β at different times through the collected plant data. i .

[0086] The recommended index of the plant is obtained based on the growth rate of the current cycle and the excellent index of the current cycle.

[0087] In this embodiment, the recommended index is a normalized value of the product of the growth rate and the quality index.

[0088] Finally, the output device is used to output the final recommendation index in descending order, and the plants that finally meet the breeding requirements are regarded as the final excellent plants, thereby realizing dynamic recommendation decisions for intelligent breeding. In this embodiment, when the recommendation index is greater than 0.7, it is considered to be an excellent plant.

[0089] Based on the same inventive concept as the above method, an embodiment of the present invention also provides a deep learning-based intelligent breeding collaborative optimization system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-mentioned deep learning-based intelligent breeding dynamic decision-making methods.

[0090] It should be noted that the above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and should all be included in the scope of protection of the present application.

[0091] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A dynamic decision-making method for intelligent breeding based on deep learning, characterized by: The method comprises the following steps: Characteristics of plants collected based on the collection device; Cluster the plants within a cycle based on all their characteristics; obtain the characteristic outlier value of each characteristic of the plant within a cycle based on the proportion of the number of clusters to which the plants belong and the difference between the plant characteristics and the overall characteristics of all plants within the cycle; and filter out abnormal cycles based on the characteristic outlier values; The plants with abnormalities within a cycle are merged to obtain the abnormal area; the long-term abnormal value is determined based on the maximum continuous abnormal value of each plant's characteristics in all cycles before each cycle; the number of plant abnormalities is obtained based on the number of plants in the maximum abnormal area in all cycles before each cycle; the number of abnormalities is obtained based on the number of times all the characteristics of the plant are abnormal; the plant's own influence index is calculated based on the number of plant abnormalities, long-term abnormal value and number of plant abnormalities; The standard duration of the plant growth stage is preset, and the plant's excellent index in each cycle is obtained based on the difference between the actual duration and the standard duration of all growth stages before each cycle, the number of abnormal cycles, and its own impact index; based on the excellent index of the previous cycle, the sampling frequency of the previous cycle is adjusted to obtain the sampling frequency of the current cycle; According to the collection frequency of the current cycle, the inspection route is adjusted to collect plant data, the plant growth rate is obtained based on the plant data, and the recommended index is obtained in combination with the plant's excellent index. The recommended index is used to screen excellent plants to complete breeding.

2. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The characteristics of the plants include plant height, stem diameter, and NDVI value.

3. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The clustering adopts the k-means clustering method, the cluster distance is the Euclidean distance between data points, and the cluster number is the ratio of the number of plants to the maximum cluster distance rounded down.

4. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The characteristic outlier is negatively correlated with the proportion of the number of data points in the cluster where the plant is located, and is negatively correlated with the difference between the plant characteristics and the overall characteristics of all plants in the period.

5. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The abnormal period refers to a period in which the plant has one or more abnormal characteristics; the abnormal characteristic refers to a characteristic whose abnormal value is greater than an abnormal threshold.

6. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The self-influence index is positively correlated with the long-term abnormal value, and negatively correlated with the abnormal times and the abnormal number of plants.

7. The deep learning-based intelligent breeding dynamic decision-making method according to claim 6, characterized in that: The method for obtaining long-term abnormal values, abnormal times and abnormal number of plants is as follows: For a feature of a plant, the features of all cycles before each cycle are combined into an abnormal sequence in time sequence; the feature abnormal values ​​in the abnormal sequence that are greater than the abnormal threshold are marked; the maximum continuous number of all marks is counted; and the maximum continuous number in the abnormal sequence of all features is recorded as a long-term abnormal value; Count the number of plants in the abnormal area corresponding to each feature of the plant, and take the maximum value as the maximum abnormal area of ​​the plant; calculate the average of the maximum abnormal areas of all cycles before each cycle as the number of plant abnormalities; The area where the abnormal areas corresponding to all the characteristics of a plant overlap is recorded as the core abnormal area; the number of times the core abnormal area exists in the plant is counted and recorded as the abnormal number.

8. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The excellent index is positively correlated with the difference between the actual duration and the standard duration of the growth stage, and is negatively correlated with the number of abnormal cycles and the self-influence index.

9. The deep learning-based intelligent breeding dynamic decision-making method according to claim 1, characterized in that: The method for adjusting the sampling frequency of the previous cycle based on the excellent index of the previous cycle to obtain the sampling frequency of the current cycle is: p T+1,i =p T,i ×(1+v T,i ), v T,i represents the quality index of the i-th plant in the T-th cycle; p T,i represents the sampling frequency of the i-th plant in the T-th period, p T+1,i Represents the sampling frequency of the i-th plant in the T+1-th cycle; the initial sampling frequency is once per cycle.

10. A smart breeding collaborative optimization system based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the deep learning-based intelligent breeding dynamic decision-making method as described in any one of claims 1 to 9 are implemented.

Images