An intelligent plant breeding monitoring method based on image processing

Abstract

The present application relates to the technical field of breeding monitoring, and more particularly to an intelligent plant breeding monitoring method based on image processing, comprising: obtaining target breeding monitoring data and a plurality of historical breeding data; determining an extraction interval according to a growth comparison value corresponding to the target breeding monitoring data, and under a preset abnormal condition, reducing the extraction interval based on a comparison difference to extract a plurality of initial extraction images; determining whether to perform extraction optimization according to a comparison abnormality degree of the initial extraction images, and when performing extraction optimization, determining to perform feature paragraph image supplementation or supplementary analysis according to an abnormal deviation degree to obtain a plurality of supplementary extraction images; and when an abnormal reference value corresponding to the extraction image is greater than a preset abnormal reference value, performing early warning. The present application can improve the accuracy of breeding monitoring.

Description

Technical Field

[0001] This invention relates to the field of breeding monitoring technology, and in particular to an intelligent plant breeding monitoring method based on image processing. Background Technology

[0002] Traditional plant breeding monitoring methods rely heavily on manual observation and recording. This approach is not only costly in terms of manpower and time, but also highly susceptible to subjective experience, making it prone to errors and hindering continuous, real-time, and accurate monitoring of the breeding process. In recent years, with the rapid development of technologies such as computer vision and image processing, non-contact and automated monitoring can be achieved by collecting plant images or video data and extracting plant feature parameters using image processing algorithms. However, during periods of slow plant growth, a fixed collection frequency generates a large amount of redundant data, increasing the burden of data storage and processing. Conversely, during critical growth periods or when anomalies occur, insufficient collection frequency may lead to the omission of important information. Therefore, how to extract images to improve the accuracy of breeding monitoring is a problem that urgently needs to be solved by those skilled in the art.

[0003] Chinese Patent Publication No. CN116820167A discloses an IoT-based control system for a vegetable breeding isolation shed, comprising: a breeding video monitoring module, a breeding environment parameter acquisition module, a central control module, an IoT communication module, a temperature regulation module, a timed ventilation module, a growth prediction module, and a display module. The temperature regulation module adjusts the temperature of the vegetable breeding isolation shed according to the appropriate temperature required for growth, thereby enabling energy recycling and effectively saving energy consumption and reducing operating costs. Simultaneously, the growth prediction module enables comprehensive prediction of the growth characteristics of vegetable seedlings. By extracting the spectrum of each pixel in the leaves and stems for component measurement, errors caused by non-uniformity in component measurement are overcome. However, the above solution has the following problems: it cannot extract images based on the actual growth state, resulting in high data redundancy and easy omission of key mutations, leading to poor accuracy in breeding monitoring. Summary of the Invention

[0004] To address this, the present invention provides an intelligent plant breeding monitoring method based on image processing, which overcomes the problems in the prior art where images cannot be extracted based on the actual growth status, resulting in high data redundancy, easy omission of key mutations, and poor accuracy of breeding monitoring.

[0005] To achieve the above objectives, the present invention provides an intelligent plant breeding monitoring method based on image processing, comprising:

[0006] Acquire target breeding monitoring data and some historical breeding data;

[0007] The extraction interval is determined based on the growth comparison value corresponding to the target breeding monitoring data. Under preset abnormal conditions, the extraction interval is adjusted to be reduced based on the comparison difference in order to extract a number of initial extraction images.

[0008] Whether to perform extraction optimization is determined based on the anomaly degree of the initial extracted image. When performing extraction optimization, the feature segment image is supplemented or supplementary analysis is performed to obtain several supplementary extracted images based on the anomaly deviation degree.

[0009] In the supplementary analysis, the selection parameters are determined based on the associated parameter group, and the feature points are determined based on the feature monitoring values ​​determined by the numerical comparison of each selection parameter. The abnormal state is determined based on the average change of the feature points and the average feature length. Based on the abnormal state, the unstable point is supplemented or the abnormal segment is uniformly extracted.

[0010] An alert is issued when the abnormal reference value corresponding to the extracted image exceeds the preset abnormal reference value;

[0011] The average feature length is the average of the feature lengths corresponding to the historical monitoring data with the label "abnormal".

[0012] Furthermore, the extraction interval is determined based on the growth comparison values ​​corresponding to the target breeding monitoring data;

[0013] The extraction interval and the growth comparison value are negatively correlated.

[0014] Furthermore, under preset abnormal conditions, the extraction interval is adjusted to be reduced based on the comparison difference;

[0015] The preset abnormal condition is that the abnormal coefficient is greater than the preset abnormal coefficient, and the comparison difference is determined based on the difference between the abnormal coefficient and the preset abnormal coefficient.

[0016] Furthermore, if the anomaly score of the initially extracted image is greater than or equal to the preset anomaly score, extraction optimization is performed.

[0017] Furthermore, if the abnormal deviation is greater than or equal to the preset abnormal deviation, then feature segment image supplementation is performed;

[0018] In the feature segment image supplementation, based on the monitoring period corresponding to the target breeding monitoring data for each extraction point, several segmentation segments can be obtained. The segmentation segment with a fluctuation anomaly degree greater than the preset fluctuation anomaly degree is recorded as the feature segment, and the number of supplemented extraction images corresponding to each feature segment is determined based on the anomaly degree difference between the fluctuation anomaly degree and the preset fluctuation anomaly degree.

[0019] Furthermore, if the abnormal deviation is less than the preset abnormal deviation, supplementary analysis will be performed.

[0020] Furthermore, the selection parameters determined based on the associated parameter group include:

[0021] The correlation parameter group is determined based on the correlation coefficient of the feature monitoring parameters, and parameters are selected for each correlation parameter group. When selecting parameters for a single correlation parameter group, the feature monitoring parameter with the largest parameter comparison value in that correlation parameter group is selected as the selection parameter.

[0022] The feature monitoring parameter is a monitoring parameter whose parameter comparison value is greater than a preset parameter comparison value.

[0023] Furthermore, if the abnormal state is that the average change of feature points is greater than or equal to the preset average change of feature points or the average feature length is greater than or equal to the preset average feature length, then unstable points are supplemented and extracted.

[0024] In the process of extracting additional instability points, the instability points corresponding to the target breeding monitoring data are determined based on the frequency of feature points, and images are extracted at each instability point.

[0025] Furthermore, if the abnormal state is that the average change of feature points is less than the preset average change of feature points and the average feature length is less than the preset average feature length, then the abnormal segment is uniformly extracted.

[0026] Furthermore, the methods for confirming the abnormal reference value include:

[0027] Based on the sub-comparison values ​​corresponding to each selected parameter, the abnormal comparison value corresponding to each extracted image is determined. The fluctuation reference value is determined according to the standard deviation of the abnormal comparison value corresponding to each extracted image. The abnormal reference value is determined according to the absolute value of the difference between the fluctuation reference value and the preset fluctuation reference value.

[0028] The extracted images include an initial extracted image and supplementary extracted images.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: In the technical solution of the present invention, the growth comparison value corresponding to the target breeding monitoring data effectively reflects the degree of deviation between the current wheat growth status and the qualified historical breeding data at the same growth stage. Then, the extraction interval is determined according to the growth comparison value corresponding to the target breeding monitoring data, which can match the extraction frequency with the growth difference and improve the accuracy of data collection. The anomaly coefficient effectively reflects the prevalence of anomalies in the current growth stage. When the anomaly coefficient is greater than the preset anomaly coefficient, it indicates that the frequency of anomalies in the current growth stage is relatively high, and more refined monitoring is required. Then, the extraction interval is reduced and adjusted, which is conducive to capturing more key images in the high-incidence stage of anomalies and avoiding the omission of important anomaly information due to sparse extraction.

[0030] Furthermore, in this invention, the anomaly degree is used to effectively reflect the overall abnormal deviation level of the growth stage corresponding to the currently extracted image. If the anomaly degree corresponding to the extracted image is greater than or equal to the preset anomaly degree, it indicates that the growth state reflected by the current image deviates significantly from the standard level. In this case, extraction optimization can be performed to track abnormal details more meticulously. If the anomaly degree is less than the preset anomaly degree, it indicates that the current growth state deviates slightly from the standard level, and the existing extraction density can meet the monitoring requirements without additional optimization, thus avoiding data redundancy.

[0031] Furthermore, this invention effectively reflects the dispersion of the extracted images in terms of abnormal deviation degree through the abnormal deviation degree, and then determines to supplement the feature segment images or conduct supplementary analysis based on the abnormal deviation degree, so that the image acquisition is more in line with the actual application scenario, avoiding the problem of missing key details due to fixed extraction strategies, which is conducive to optimizing data acquisition efficiency while ensuring monitoring accuracy, thereby providing more accurate and efficient image data support for breeding monitoring and improving the timeliness of anomaly identification.

[0032] Furthermore, in this invention, the average deviation and average duration of abnormal states during plant breeding are effectively reflected by the average change of feature points and the average length of feature points. Then, the abnormal state is determined based on the average change of feature points and the average length of feature points. By adaptively extracting unstable points or uniformly extracting abnormal segments through abnormal states, the image acquisition can accurately match the actual characteristics of the abnormality, which is beneficial to improving the accuracy and efficiency of breeding monitoring. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the intelligent plant breeding monitoring method based on image processing according to the present invention;

[0034] Figure 2 This is a flowchart of the present invention for determining whether to perform extraction optimization based on the comparison anomaly degree corresponding to the extracted image;

[0035] Figure 3 This is a flowchart illustrating the process of supplementing or analyzing feature segment images based on abnormal deviation in this invention.

[0036] Figure 4 This is a flowchart illustrating the process of supplementing and extracting unstable points or uniformly extracting abnormal segments based on abnormal conditions, according to the present invention. Detailed Implementation

[0037] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0038] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0039] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0040] Please see Figures 1 to 4 As shown, this invention provides an intelligent plant breeding monitoring method based on image processing, comprising:

[0041] Acquire target breeding monitoring data and some historical breeding data;

[0042] The extraction interval is determined based on the growth comparison value corresponding to the target breeding monitoring data. Under preset abnormal conditions, the extraction interval is adjusted to be reduced based on the comparison difference in order to extract a number of initial extraction images.

[0043] Whether to perform extraction optimization is determined based on the anomaly degree of the initial extracted image. When performing extraction optimization, the feature segment image is supplemented or supplementary analysis is performed to obtain several supplementary extracted images based on the anomaly deviation degree.

[0044] In the supplementary analysis, the selection parameters are determined based on the associated parameter group, and the feature points are determined based on the feature monitoring values ​​determined by the numerical comparison of each selection parameter. The abnormal state is determined based on the average change of the feature points and the average feature length. Based on the abnormal state, the unstable point is supplemented or the abnormal segment is uniformly extracted.

[0045] An alert is issued when the abnormal reference value corresponding to the extracted image exceeds the preset abnormal reference value;

[0046] The average feature length is the average of the feature lengths corresponding to the historical monitoring data with the label "abnormal".

[0047] The application scenario of this invention is plant breeding monitoring. This invention has several historical records, each of which records the growth comparison value, sampling interval, anomaly coefficient, and comparison anomaly degree of at least one plant breeding monitoring process. Each historical record also has a corresponding qualified mark, which records whether the plant breeding monitoring process meets the user's needs. The qualified mark can be recorded manually. It is understood that the user can determine whether the plant breeding monitoring process meets the requirements based on self-defined indicators. Self-defined indicators can be, but are not limited to, the number of errors, which will not be elaborated here. The number of errors refers to the number of times when abnormal plant growth occurs during the breeding process without warning.

[0048] The present invention has a continuous cyclic monitoring cycle, the duration of which can be set according to the user's needs. The greater the user's need to improve the accuracy of breeding monitoring, the shorter the duration of the monitoring cycle. A monitoring cycle value of 24 hours is provided.

[0049] The target breeding monitoring data consists of images of the target plant captured in real-time by a high-definition digital camera at a frequency of 1 frame per minute at various shooting points within the target monitoring area during a single monitoring cycle. The target plant species is wheat. This invention sets up several shooting points to photograph the target plant. One shooting point is set up every 2 square meters within a single monitoring area. Each shooting point corresponds to monitoring one wheat plant within its field of view. It is important to note that the image captured at each shooting point must completely cover the entire structure of the target wheat plant. Vertically, it should extend from the point of contact between the root and the soil (base) to the top of the plant (growing point in the seedling stage, ear of wheat after heading); horizontally, it should include the complete morphology of all leaves and stems. The shooting height and angle of all shooting points must be completely consistent and will be adjusted synchronously according to the wheat's growth stage. For example, during the wheat seedling stage (plant height is usually less than 30 cm), all shooting points are uniformly set at 40 cm above the ground. The height is set at a 60° downward angle. When the wheat enters the jointing stage (plant height exceeds 50 cm), all shooting points will be simultaneously adjusted to a height of 100 cm above the ground, using a 30° side-down angle. The target monitoring area is the monitoring area currently undergoing breeding monitoring, which is the experimental area for breeding. This invention includes several historical breeding data, each historical breeding data corresponding to a label, which includes qualified and abnormal labels. Each historical breeding data contains the monitoring parameters corresponding to each time point monitored in real time after the start of sowing in a single monitoring area. The monitoring parameters include, but are not limited to, the average number of leaves, the average plant height, and the leaf area index. A method for setting the time points corresponding to historical breeding data is provided. For a single historical breeding data, the start time of sowing of the historical breeding data is taken as the starting point, and an interval point is set every 1 minute. The starting point and each interval point are recorded as time points.

[0050] It should be noted that the environmental parameters corresponding to the breeding data and the target breeding monitoring data are the same. These environmental parameters include, but are not limited to, temperature, humidity, light intensity and carbon dioxide concentration. This is something that is easy for those skilled in the art to understand, and will not be elaborated on in detail.

[0051] The value of a single monitoring parameter corresponding to a single moment is determined by the images captured at each shooting point at that moment. For a single moment, this moment is recorded as the target moment, and the images captured at each shooting point at the target moment are recorded as target images. The average number of leaves corresponding to the target moment is the average number of leaves corresponding to each target image, and the plant height corresponding to the target moment is the average plant height corresponding to each target image. The number of leaves, plant height, and leaf area index can be determined based on the captured images through image processing. Using a semantic segmentation model (such as U-Net), wheat plants are separated from the background such as soil and weeds. Gaussian filtering is used to eliminate image noise to obtain a preprocessed image. Contrast enhancement is used to highlight the leaf edges, and the Canny algorithm is used to extract the leaf edges. Combining the morphological characteristics of wheat leaves, such as "slender, with obvious midribs, and slightly serrated edges," regions that conform to the leaf contour are selected. The number of each segmented independent leaf contour is counted, which is the number of leaves corresponding to the target image. In the preprocessed image, wheat plants are found... The boundary line with the soil is defined by the pixel position corresponding to the midpoint of the bottom of the plant outline, denoted as (x1, y1). The top position is determined according to the wheat growth stage: the highest point of the leaf tip during the seedling stage, and the top of the ear after heading. The pixel position of the highest point of the top of the plant outline is defined as (x2, y2). The pixel distance Δy between the top and the base is: Δy = |y2 - y1|. Plant height = Δy × length scale calibration coefficient, which is 0.1 cm / pixel. The total leaf pixel area corresponding to a single target image is the sum of the pixel areas of all segmented leaf outlines in the target image (the total number of pixels contained in the outline). The actual total leaf area = the average of the total leaf pixel areas corresponding to each target image × the number of target plants in the target monitoring area × the area calibration coefficient, which is 0.01 cm² / pixel. The leaf area index corresponding to a single moment in a single monitoring area = actual total leaf area / area of ​​the monitoring area.

[0052] Specifically, the extraction interval is determined based on the growth comparison value corresponding to the target breeding monitoring data;

[0053] The extraction interval and the growth comparison value are negatively correlated.

[0054] Among them, the growth comparison value is the maximum value among the sub-comparison values ​​corresponding to each monitoring parameter at the end of the current monitoring cycle;

[0055] For any given moment, that moment is recorded as the target moment, and the moment when sowing begins in a single monitored area is recorded as the sowing moment; the duration between the sowing moment and the target moment is recorded as the reference duration.

[0056] The target time corresponds to a reference time for each historical breeding data point. The reference time for the target time in a single historical breeding data point is the time from the sowing time corresponding to that historical breeding data point to the reference time. It should be noted that the reference time for a single historical breeding data point is later than the sowing time corresponding to that historical breeding data point.

[0057] For a single monitoring parameter, the average value of the monitoring parameter corresponding to the reference time of each historical breeding data that is labeled as qualified is denoted as a. At the target time, the sub-comparison value of the monitoring parameter is (a - the value of the monitoring parameter corresponding to the target time of the target breeding data) / a.

[0058] The extraction interval is the number of images between two adjacent extracted images;

[0059] Historical records that can meet user needs and whose extraction intervals are determined based on the growth comparison values ​​corresponding to the target breeding monitoring data will be recorded as reference historical records.

[0060] The extraction interval is the smallest integer greater than or equal to w1, where w1 = the average of the growth comparison values ​​corresponding to each reference historical record / the growth comparison value × the reference interval. The reference interval is the average of the extraction intervals determined based on the growth comparison values ​​in each reference historical record.

[0061] Specifically, under preset abnormal conditions, the extraction interval is adjusted to be reduced based on the comparison difference;

[0062] The preset abnormal condition is that the abnormal coefficient is greater than the preset abnormal coefficient, and the comparison difference is determined based on the difference between the abnormal coefficient and the preset abnormal coefficient.

[0063] Among them, the monitoring period in which the time from the sowing time in the historical records that can meet the user's needs is used as the reference time is recorded as the comparison monitoring period;

[0064] Anomaly coefficient = Number of comparison monitoring cycles in which warnings are sent to users / Total number of comparison monitoring cycles;

[0065] The user can determine the value of the preset anomaly coefficient according to the actual application scenario. The greater the user's need to improve the monitoring accuracy, the smaller the value of the preset anomaly coefficient. A preset anomaly coefficient value is provided. The historical records of the sampling interval are adjusted based on the anomaly coefficient. The average value of the anomaly coefficients corresponding to the historical records that meet the user's needs is recorded as the preset anomaly coefficient.

[0066] Comparison difference = anomaly coefficient - preset anomaly coefficient;

[0067] The reduction value of the extraction interval is the smallest integer greater than or equal to w2, where w2 = (comparison difference / preset anomaly coefficient) × extraction interval determined based on growth comparison value × ratio coefficient, and the ratio coefficient is 0.5.

[0068] When sampling at a predetermined interval, the starting point of the current monitoring cycle is taken as the starting point, and images are extracted according to the sampling interval. Each extracted image is recorded as the initial sampled image. It should be noted that when extracting images for a single moment, images captured by all shooting points at that moment must be extracted.

[0069] Specifically, if the anomaly score of the initially extracted image is greater than or equal to the preset anomaly score, extraction optimization is performed.

[0070] If the anomaly score of the initially extracted image is less than the preset anomaly score, then no extraction optimization is required.

[0071] The anomaly rate is the average of the comparison reference values ​​corresponding to each initial extracted image. The comparison reference value corresponding to a single initial extracted image is the maximum value among the sub-comparison values ​​corresponding to each monitoring parameter at the acquisition time corresponding to that initial extracted image.

[0072] The preset value of the comparison anomaly degree can be determined by the user according to the actual application scenario. The greater the user's demand for improving the accuracy of breeding monitoring, the smaller the preset value of the comparison anomaly degree should be. A method for setting the preset comparison anomaly degree is provided, which detects the user's historical extraction optimization history and records the average value of the comparison anomaly degree corresponding to the historical history that meets the user's needs as the preset comparison anomaly degree.

[0073] Specifically, if the abnormal deviation is greater than or equal to the preset abnormal deviation, then feature segment image supplementation is performed;

[0074] In the feature segment image supplementation, based on the monitoring period corresponding to the target breeding monitoring data for each extraction point, several segmentation segments can be obtained. The segmentation segment with a fluctuation anomaly degree greater than the preset fluctuation anomaly degree is recorded as the feature segment, and the number of supplemented extraction images corresponding to each feature segment is determined based on the anomaly degree difference between the fluctuation anomaly degree and the preset fluctuation anomaly degree.

[0075] Wherein, the anomaly deviation is the standard deviation of the comparison reference value corresponding to each initially extracted image;

[0076] The user can determine the value of the preset anomaly deviation degree according to the actual application scenario. The larger the value of the preset anomaly deviation degree, the greater the user's need for supplementary analysis. A method for determining the value of the preset anomaly deviation degree is provided, which detects the user's historical records of supplementing feature segment images, and records the average value of the anomaly deviation degree corresponding to the historical records that can meet the user's needs as the preset anomaly deviation degree.

[0077] The extraction point is the acquisition time corresponding to each initial extracted image, and a single segment is the time segment between two adjacent extraction points;

[0078] The fluctuation anomaly degree corresponding to a single segment = the absolute value of the difference between the comparison reference values ​​corresponding to the two initial extracted images of the segment / the larger value among the comparison reference values ​​corresponding to the two initial extracted images of the segment;

[0079] The user can determine the value of the preset fluctuation anomaly degree according to the actual application scenario. The greater the user's demand for improving the accuracy of breeding monitoring, the smaller the value of the preset fluctuation anomaly degree. A method for determining the value of the preset fluctuation anomaly degree is provided, which detects the user's historical records of supplementing feature segment images, and records the average value of the fluctuation anomaly degree corresponding to each feature segment in the historical records that can meet the user's needs as the preset fluctuation anomaly degree.

[0080] The anomaly difference for a single feature segment = the fluctuation anomaly corresponding to that feature segment - the preset fluctuation anomaly;

[0081] The number of supplementary extracted images corresponding to a single feature segment is the smallest integer greater than or equal to g, where g = (the anomaly difference corresponding to the feature segment / the preset fluctuation anomaly) × the total number of images in the feature segment × the ratio coefficient, and the ratio coefficient is 0.5;

[0082] The images in a single feature segment include two initial extracted images corresponding to the feature segment and images to be extracted. The images to be extracted are the images in the feature segment located between the two initial extracted images.

[0083] When extracting supplementary images for a single feature segment, the first supplementary image is extracted starting from the earliest time of the feature segment. Then, other supplementary images are extracted sequentially at fixed intervals until the total number of extractions reaches g. The fixed interval is the smallest integer less than or equal to k, where k = the number of images corresponding to the feature segment / the number of supplementary images.

[0084] Specifically, if the abnormal deviation is less than the preset abnormal deviation, supplementary analysis will be performed.

[0085] It is understandable that the abnormal deviation degree can effectively reflect the dispersion of the abnormal deviation of the initial extracted image. When the abnormal deviation degree is greater than or equal to the preset abnormal deviation degree, it indicates that the difference in the degree of abnormal deviation of different initial extracted images is more significant, the data stability is poor, and there is a possibility of local drastic fluctuations. It is necessary to supplement with feature segment images to capture the detailed changes between the two extraction points and avoid the omission of key fluctuation information caused by the extraction interval.

[0086] When the abnormal deviation is less than the preset abnormal deviation, it means that the abnormal deviation of the initially extracted image is close and the overall change is gradual, which cannot reflect the subtle local abnormal features. Therefore, supplementary analysis is required to accurately identify potential abnormal features and ensure that no gradual but key abnormal information is missed.

[0087] Specifically, determining the selection parameters based on the associated parameter group includes:

[0088] The correlation parameter group is determined based on the correlation coefficient of the feature monitoring parameters, and parameters are selected for each correlation parameter group. When selecting parameters for a single correlation parameter group, the feature monitoring parameter with the largest parameter comparison value in that correlation parameter group is selected as the selection parameter.

[0089] The feature monitoring parameter is a monitoring parameter whose parameter comparison value is greater than a preset parameter comparison value.

[0090] The comparison interval for a single historical breeding data point is defined as [the initial time of the current monitoring period is the reference time corresponding to the historical breeding data, and the end time of the current monitoring period is the reference time corresponding to the historical breeding data];

[0091] For a single monitoring parameter, the monitoring mean value for that monitoring parameter in a single historical breeding data set is the average value of that monitoring parameter at each time point within the comparison interval of that historical breeding data set;

[0092] The parameter comparison value corresponding to a single monitoring parameter = the average of the monitoring mean values ​​corresponding to that monitoring parameter in each historical breeding data set that is labeled as qualified - the average of the monitoring mean values ​​corresponding to that monitoring parameter in each historical breeding data set that is labeled as abnormal;

[0093] The user can determine the value of the preset parameter comparison value according to the actual application scenario. The smaller the value of the preset parameter comparison value, the greater the user's need to identify the monitoring parameter as a feature monitoring parameter. A method for determining the preset parameter comparison value is provided, which records the average value of the parameter comparison values ​​corresponding to each feature monitoring parameter in the historical records that can meet the user's needs as the preset parameter comparison value.

[0094] The correlation parameter group is determined based on the correlation coefficient of the feature monitoring parameters, including: performing correlation analysis on each feature monitoring parameter; when performing correlation analysis on a single feature monitoring parameter, the feature monitoring parameter is recorded as the target feature monitoring parameter; other feature monitoring parameters besides the target feature monitoring parameter are recorded as reference feature monitoring parameters; the reference feature monitoring parameters whose correlation coefficient with the target feature monitoring parameter is greater than the preset correlation coefficient and the target feature monitoring parameter are recorded as a correlation parameter group; and the correlation analysis continues for feature monitoring parameters not recorded in the correlation parameter group until all feature monitoring parameter groups are recorded in the corresponding correlation parameter group, at which point the correlation analysis stops.

[0095] The correlation coefficient between any two feature monitoring parameters is the average of the sub-correlation coefficients between the two feature monitoring parameters in each historical breeding data where the label is qualified;

[0096] The formula for calculating the sub-correlation coefficient r between two characteristic monitoring parameters in a single historical breeding dataset is as follows:

[0097]

[0098] Where m represents the number of time points in the comparison interval of the historical breeding data. It is the value of a feature monitoring parameter corresponding to the k-th time point in the comparison interval of the historical breeding data. It is the value of another feature monitoring parameter corresponding to the k-th time point in the comparison interval of the historical breeding data. for The corresponding characteristic monitoring parameter is the average value of the characteristic monitoring parameter at each time point within the comparison interval of the historical breeding data. for The corresponding characteristic monitoring parameter is the average value of the characteristic monitoring parameter at each time point in the comparison interval of the historical breeding data, k = 1, 2, 3, ..., m;

[0099] The user can determine the value of the preset correlation coefficient according to the actual application scenario. The greater the user's need to improve the accuracy of the correlation between feature monitoring parameters in the correlation parameter group, the larger the value of the preset correlation coefficient will be. One preset correlation coefficient value is provided, which is 0.75.

[0100] Specifically, if the abnormal state is that the average change of feature points is greater than or equal to the preset average change of feature points or the average feature length is greater than or equal to the preset average feature length, then unstable points will be extracted.

[0101] In the process of extracting additional instability points, the instability points corresponding to the target breeding monitoring data are determined based on the frequency of feature points, and images are extracted at each instability point.

[0102] The images extracted at each instability point are referred to as supplementary extracted images.

[0103] The abnormal state includes a first abnormal state and a second abnormal state. The first abnormal state is when the average change of feature points is greater than or equal to the preset average change of feature points or the average feature length is greater than or equal to the preset average feature length. The second abnormal state is when the average change of feature points is less than the preset average change of feature points and the average feature length is less than the preset average feature length.

[0104] When determining feature points based on the feature monitoring values ​​determined by the numerical comparison degree of each selected parameter, the numerical comparison degree corresponding to a single selected parameter = |average value of the selected parameter at the reference time point corresponding to each historical monitoring data with a qualified label - value of the selected parameter at the reference time point corresponding to each historical monitoring data with an abnormal label| / average value of the selected parameter at the reference time point corresponding to each historical monitoring data with a qualified label; the feature monitoring value corresponding to a single time point in a single historical breeding data with an abnormal label is the maximum value among the numerical comparison degrees corresponding to each selected parameter; each historical breeding data with an abnormal label corresponds to several feature points, and the feature point corresponding to a single historical breeding data with an abnormal label is the time point where the feature monitoring value is greater than the preset feature monitoring value;

[0105] The mean change of feature points is the average of the sub-change values ​​corresponding to each historical monitoring data with an abnormal label. The sub-change value corresponding to a single historical monitoring data with an abnormal label is equal to |the average of the feature monitoring values ​​corresponding to each feature point in the historical monitoring data with an abnormal label - the average of the feature monitoring values ​​corresponding to each time point in the historical monitoring data with an abnormal label|.

[0106] The mean feature length is the average of the feature lengths corresponding to each historical monitoring data point labeled as anomaly; the feature length corresponding to a single historical monitoring data point labeled as anomaly is the time length between the earliest and latest feature points in that historical monitoring data point labeled as anomaly.

[0107] The user can determine the values ​​of the preset mean change of feature points and the preset mean length of feature points according to the actual application scenario. The smaller the values ​​of the preset mean change of feature points and the preset mean length of feature points, the greater the user's need for instability point supplementation extraction. A method for determining the values ​​of the preset mean change of feature points and the preset mean length of feature points is provided. The historical records of the user's instability point supplementation extraction are detected, and the average value of the mean change of feature points and the average value of the mean length of feature points corresponding to the historical records that meet the user's needs are respectively recorded as the preset mean change of feature points and the preset mean length of feature points.

[0108] A method for setting time points corresponding to target breeding monitoring data is provided, taking the start time of the current monitoring cycle as the starting point, setting an interval point every 1 minute, and recording the starting point and each interval point as time points;

[0109] For a single time point in the target breeding monitoring data, this time point is recorded as the target time point, and the time points in the comparison interval of the historical breeding data that are in the same order as the target time point are recorded as reference time points. It can be understood that each historical breeding data has a corresponding reference time point.

[0110] If the frequency of feature points corresponding to the target time point is greater than the preset frequency of feature points, then the target time point is recorded as the instability point;

[0111] The frequency of feature points corresponding to the target time point = the number of feature points in the reference time points corresponding to the historical breeding data with the label of anomalous / the number of reference time points corresponding to the historical breeding data with the label of anomalous;

[0112] The user can determine the preset feature point frequency value according to the actual application scenario. The greater the user's need for more precise anomaly capture, the smaller the preset feature point frequency value should be. One preset feature point frequency value is provided, which is 50%.

[0113] Specifically, if the abnormal state is that the average change of feature points is less than the preset average change of feature points and the average feature length is less than the preset average feature length, then the abnormal segment is uniformly extracted.

[0114] Among them, the detection label is the starting feature point corresponding to each historical monitoring data that is abnormal, and the starting feature point corresponding to a single historical monitoring data that is abnormal is the earliest feature point in that historical monitoring data.

[0115] The time points in the target breeding monitoring data corresponding to each starting feature point are detected and recorded as matching time points. The time point in the target breeding monitoring data corresponding to a single starting feature point is the time point in the target breeding monitoring data that is in the same order as the starting feature point. Each starting feature point corresponds to a matching time point. The smallest time segment that can contain all matching time points is recorded as an abnormal segment.

[0116] The extraction interval corresponding to the abnormal segment is the smallest integer greater than or equal to w3, where w3 = number of images in the abnormal segment / average number × average interval.

[0117] The mean number is the average number of images in the historical records corresponding to the abnormal segments that are uniformly extracted from each abnormal segment and can meet the user's needs.

[0118] The mean interval is the average of the extraction intervals corresponding to the historical records that are uniformly extracted from each abnormal segment and can meet the user's needs.

[0119] When uniformly extracting abnormal segments, the first frame of supplementary extraction image is extracted starting from the beginning time of the abnormal segment. Then, other supplementary extraction images are extracted sequentially according to the extraction interval corresponding to the abnormal segment.

[0120] Understandably, the average deviation and average duration of anomalies can be effectively reflected by the average change of feature points and the average length of feature points. In the first abnormal state, it indicates that the deviation of the anomaly is large or the duration is long, which is a relatively significant anomaly. It is necessary to supplement the extraction by unstable points, focus on the key time points where the abnormal features appear frequently, and accurately capture the details of the anomaly.

[0121] In the second abnormal state, it indicates that the deviation of the abnormality is small and the duration is short. It is a relatively gentle abnormal situation. It is necessary to extract the abnormal segments evenly to cover the entire time period including the potential abnormality starting point and to record the development process of the gentle abnormality completely.

[0122] Specifically, the methods for confirming the abnormal reference values ​​include:

[0123] Based on the sub-comparison values ​​corresponding to each selected parameter, the abnormal comparison value corresponding to each extracted image is determined. The fluctuation reference value is determined according to the standard deviation of the abnormal comparison value corresponding to each extracted image. The abnormal reference value is determined according to the absolute value of the difference between the fluctuation reference value and the preset fluctuation reference value.

[0124] The extracted images include an initial extracted image and supplementary extracted images.

[0125] Among them, the anomaly comparison value corresponding to a single extracted image is the average of the sub-comparison values ​​corresponding to each selected parameter at the acquisition time corresponding to the extracted image;

[0126] Anomaly reference value = |fluctuation reference value - preset fluctuation reference value|, where the fluctuation reference value is the standard deviation of the anomaly comparison value corresponding to each extracted image, and the preset fluctuation reference value is the average value of the fluctuation reference value corresponding to the historical records that do not require warning and can meet the user's needs.

[0127] The user can determine the preset anomaly reference value based on the actual application scenario. The greater the user's need to improve the accuracy of anomaly detection, the smaller the preset anomaly reference value should be. A method for setting the preset anomaly reference value is provided, which detects the user's historical warning records and records the minimum anomaly reference value among the historical records that meet the user's needs as the preset anomaly reference value.

[0128] When the abnormal reference value is greater than the preset abnormal reference value, it indicates that the fluctuation of abnormal characteristics of plant growth is far beyond expectations, suggesting that there may be significant instability in the growth status. It is necessary to promptly remind the management personnel to intervene through early warning to prevent the abnormality from further expanding and affecting the breeding effect.

[0129] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A smart plant breeding monitoring method based on image processing, characterized in that, include: Acquire target breeding monitoring data and some historical breeding data; The extraction interval is determined based on the growth comparison value corresponding to the target breeding monitoring data. Under preset abnormal conditions, the extraction interval is adjusted to be reduced based on the comparison difference in order to extract a number of initial extraction images. Whether to perform extraction optimization is determined based on the anomaly degree of the initial extracted image. When performing extraction optimization, the feature segment image is supplemented or supplementary analysis is performed to obtain several supplementary extracted images based on the anomaly deviation degree. In the supplementary analysis, the selection parameters are determined based on the associated parameter group, and the feature points are determined based on the feature monitoring values ​​determined by the numerical comparison of each selection parameter. The abnormal state is determined based on the average change of the feature points and the average feature length. Based on the abnormal state, the unstable point is supplemented or the abnormal segment is uniformly extracted. An alert is issued when the abnormal reference value corresponding to the extracted image exceeds the preset abnormal reference value; The mean feature length is the average of the feature lengths corresponding to the historical monitoring data with the label "abnormal". The growth comparison value is the maximum value among the sub-comparison values ​​corresponding to each monitoring parameter at the end of the current monitoring cycle; for a single monitoring parameter, the average value of the monitoring parameter corresponding to the reference time of each historical breeding data that is marked as qualified is denoted as a, and the sub-comparison value corresponding to the monitoring parameter at the target time is (a - the value of the monitoring parameter corresponding to the target time of the target breeding data) / a; Comparison difference = anomaly coefficient - preset anomaly coefficient; Anomaly coefficient = Number of comparison monitoring cycles in which warnings are sent to users / Total number of comparison monitoring cycles; The numerical comparison degree corresponding to a single selected parameter = |Average value of the selected parameter at the reference time point corresponding to each historical monitoring data with a qualified label - Value of the selected parameter at the reference time point corresponding to each historical monitoring data with an abnormal label| / Average value of the selected parameter at the reference time point corresponding to each historical monitoring data with a qualified label; The anomaly rate is the average of the comparison reference values ​​corresponding to each initial extracted image. The comparison reference value corresponding to a single initial extracted image is the maximum value among the sub-comparison values ​​corresponding to each monitoring parameter at the acquisition time corresponding to that initial extracted image. The anomaly deviation is the standard deviation of the comparison reference value corresponding to each initially extracted image; Anomaly reference value = |fluctuation reference value - preset fluctuation reference value|, where the fluctuation reference value is the standard deviation of the anomaly comparison value corresponding to each extracted image; the anomaly comparison value corresponding to a single extracted image is the average of the sub-comparison values ​​corresponding to each selected parameter at the acquisition time corresponding to that extracted image; The segments with a volatility anomaly degree greater than the preset volatility anomaly degree are categorized as characteristic segments. The instability point is the time point at which the frequency of the feature point exceeds the preset frequency of the feature point; The time points in the target breeding monitoring data corresponding to each starting feature point are detected and recorded as matching time points. The time point in the target breeding monitoring data corresponding to a single starting feature point is the time point in the target breeding monitoring data that is in the same order as the starting feature point. Each starting feature point corresponds to a matching time point. The smallest time segment that can contain all matching time points is recorded as an abnormal segment. The correlation parameter group is determined based on the correlation coefficient of the feature monitoring parameters. Correlation analysis is performed on each feature monitoring parameter. When performing correlation analysis on a single feature monitoring parameter, the reference feature monitoring parameter whose correlation coefficient with the target feature monitoring parameter is greater than the preset correlation coefficient and the target feature monitoring parameter are recorded as a correlation parameter group. Parameters are selected for each associated parameter group. When selecting parameters for a single associated parameter group, the feature monitoring parameter with the largest parameter comparison value in that associated parameter group is selected as the selection parameter. The feature monitoring parameter is a monitoring parameter whose parameter comparison value is greater than a preset parameter comparison value; The parameter comparison value corresponding to a single monitoring parameter = the average of the monitoring mean values ​​corresponding to that monitoring parameter in each historical breeding data set that is labeled as qualified - the average of the monitoring mean values ​​corresponding to that monitoring parameter in each historical breeding data set that is labeled as abnormal; The feature monitoring value corresponding to a single time point in historical breeding data with a single label being abnormal is the maximum value among the numerical comparison degrees corresponding to each selected parameter; each historical breeding data with a label being abnormal corresponds to several feature points, and the feature point corresponding to a single historical breeding data with a label being abnormal is the time point where the feature monitoring value is greater than the preset feature monitoring value. The mean change of feature points is the average of the sub-change values ​​corresponding to each historical monitoring data with an abnormal label. The sub-change value corresponding to a single historical monitoring data with an abnormal label is equal to |the average of the feature monitoring values ​​corresponding to each feature point in the historical monitoring data with an abnormal label - the average of the feature monitoring values ​​corresponding to each time point in the historical monitoring data with an abnormal label|. The extraction interval is the smallest integer greater than or equal to w1, where w1 = the average of the growth comparison values ​​corresponding to each reference history record / growth comparison value × reference interval, and the reference interval is the average of the extraction intervals determined based on the growth comparison values ​​in each reference history record. The reduction value of the extraction interval is the smallest integer greater than or equal to w2, where w2 = (comparison difference / preset anomaly coefficient) × extraction interval determined based on growth comparison value × ratio coefficient, and the ratio coefficient is 0.

5. If the abnormal deviation is greater than or equal to the preset abnormal deviation, then feature segment image supplementation is performed; In the feature segment image supplementation, based on each extraction point, the monitoring period corresponding to the target breeding monitoring data is divided into several segments. The segment with a fluctuation anomaly degree greater than the preset fluctuation anomaly degree is recorded as a feature segment, and the number of supplemented extraction images corresponding to each feature segment is determined based on the anomaly degree difference between the fluctuation anomaly degree and the preset fluctuation anomaly degree. If the abnormal deviation is less than the preset abnormal deviation, then supplementary analysis will be performed; The fluctuation anomaly degree corresponding to a single segment = the absolute value of the difference between the comparison reference values ​​corresponding to the two initial extracted images of the segment / the larger value among the comparison reference values ​​corresponding to the two initial extracted images of the segment; The anomaly difference for a single feature segment = the fluctuation anomaly corresponding to that feature segment - the preset fluctuation anomaly; The number of supplementary extracted images corresponding to a single feature segment is the smallest integer greater than or equal to g, where g = (the anomaly difference corresponding to the feature segment / the preset fluctuation anomaly) × the total number of images in the feature segment × the ratio coefficient, and the ratio coefficient is 0.5; If the abnormal state is that the average change of feature points is greater than or equal to the preset average change of feature points or the average feature length is greater than or equal to the preset average feature length, then unstable points will be supplemented and extracted. In the process of extracting additional instability points, the instability points corresponding to the target breeding monitoring data are determined based on the frequency of feature points, and images are extracted at each instability point. If the abnormal state is that the average change of feature points is less than the preset average change of feature points and the average feature length is less than the preset average feature length, then the abnormal segment is extracted uniformly. The extraction interval corresponding to the abnormal segment is the smallest integer greater than or equal to w3, where w3 = number of images in the abnormal segment / average number × average interval. The mean number is the average number of images in the historical records corresponding to the abnormal segments that are uniformly extracted from each abnormal segment and can meet the user's needs. The mean interval is the average of the extraction intervals corresponding to the historical records that are uniformly extracted from each abnormal segment and can meet the user's needs. The frequency of feature points corresponding to the target time point = the number of feature points in the reference time points corresponding to the historical breeding data with the label of anomalous / the number of reference time points corresponding to the historical breeding data with the label of anomalous.

2. The intelligent plant breeding monitoring method based on image processing according to claim 1, characterized in that, The extraction interval is determined based on the growth comparison value corresponding to the target breeding monitoring data; The extraction interval and the growth comparison value are negatively correlated.

3. The intelligent plant breeding monitoring method based on image processing according to claim 2, characterized in that, Under preset abnormal conditions, the extraction interval is adjusted to be reduced based on the comparison difference; The preset abnormal condition is that the abnormal coefficient is greater than the preset abnormal coefficient.

4. The intelligent plant breeding monitoring method based on image processing according to claim 1, characterized in that, If the anomaly score of the initially extracted image is greater than or equal to the preset anomaly score, then extraction optimization is performed.

5. The intelligent plant breeding monitoring method based on image processing according to claim 1, characterized in that, The extracted images include the initial extracted image and the supplementary extracted image.

Images