A method and system for identifying seedling growth

Abstract

The present invention relates to the technical field of image processing, and in particular to a method and system for identifying seedling growth. The method comprises: firstly collecting images of plants during their growth process and extracting plant regions, determining frequency importance according to the frequency of occurrence of grayscale values ​​in each plant region, then calculating the grayscale complexity of the grayscale value, multiplying the frequency importance and the grayscale complexity to normalize the importance, thereby distinguishing important grayscale values ​​from unimportant grayscale values, and then constructing feature vectors of each plant region based on the important grayscale values. Finally, the feature vectors are input into a pre-trained decision tree classification model, thereby obtaining the seedling growth category of each plant region and realizing seedling growth recognition. The method improves the accuracy and stability of decision tree classification by finely extracting image features and effectively quantifying the image features, and can accurately identify the seedling growth of each plant region.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and system for identifying seedling growth. Background Art

[0002] In the process of sustainable development of modern agriculture, precision agriculture has become a core strategy to promote the improvement of crop production efficiency and quality, and accurate identification and monitoring of seedling growth is a key component of precision agriculture. Traditional seedling identification methods mostly rely on manual direct observation of plant growth conditions to determine. This manual judgment mode is extremely inefficient, and due to differences in experience and judgment standards among different personnel, it is highly subjective and difficult to meet the requirements of large-scale planting and refined management of modern agriculture.

[0003] With the gradual application of image processing technology in the agricultural field, image processing methods are now introduced into decision tree algorithms to classify plant images. For example, the existing Chinese patent document with the announcement number CN112215186B discloses a method for classifying swamp wetland vegetation, which specifically uses drones to take images of swamp wetlands, and uses decision trees and Gini indexes to determine the attributes of the acquired images, thereby performing initial classification, and using the maximum flow / minimum cut algorithm to segment the images after the initial classification, and combining the RF classifier to determine the category, thereby realizing a vegetation classification method based on drone images.

[0004] However, the decision tree in the above technical solution classifies vegetation at a relatively macroscopic level. When judging the growth of plant seedlings, it is necessary to perform refined feature extraction on plant images at different growth stages. Plant images contain rich and diverse information features. These information present complex changing states in plant images. The decision tree classification method of the above technical solution is difficult to accurately capture and quantify these feature changes, and is unable to determine the precise correspondence and quantitative boundaries between these feature changes and seedling conditions. As a result, the constructed classification rules lack accuracy and pertinence, and the decision tree is difficult to accurately identify seedling growth. Summary of the invention

[0005] In order to solve the problem that the traditional decision tree classification method is difficult to accurately capture and quantify the feature changes in the plant image when identifying the seedling growth, resulting in the inability to accurately identify the seedling growth, the present invention proposes a seedling growth identification method, comprising:

[0006] Collect multiple images during the growth of the plant, extract the plant area in each image, and determine the frequency importance of the gray value according to the frequency of occurrence of the gray value in each plant area, wherein the frequency importance is used to reflect the criticality of the gray value in each plant area;

[0007] Get all the pixel points corresponding to the grayscale value in each plant area and calculate the grayscale complexity of the grayscale value:

[0008] , Gray value The grayscale complexity, , Grayscale values In the The corresponding The total number of pixels contained in the local range of pixels and the standard deviation of the grayscale values ​​of the pixels contained in the local range, is the first The gray value of each pixel;

[0009] The importance of the gray value is obtained by multiplying the frequency importance and the gray complexity and normalizing them, and the important gray value and the unimportant gray value are determined according to the importance, and the characteristic vector of each plant region is obtained based on the important gray value;

[0010] The feature vectors of each plant region are input into a pre-trained decision tree classification model to obtain the seedling growth category of each plant region to realize the identification of the seedling growth.

[0011] The above technical solution can cover the appearance of plants at different growth stages and under different environmental conditions by collecting multiple images, thereby increasing the diversity and comprehensiveness of the data. Extracting the plant area focuses on the research object, eliminating the interference of other irrelevant background information in the image, so that the subsequent analysis can be more accurately targeted at the characteristics of the plant itself. Furthermore, the distribution of grayscale values ​​is converted into a quantitative indicator with relative importance, namely, frequency importance, so that the originally complex and disordered grayscale information has a comparable and analyzable basis, and can preliminarily screen out the grayscale value range that may be important for seedling judgment, which helps to highlight key information in subsequent processing, reduce unnecessary information processing, and improve the pertinence and efficiency of feature extraction and analysis. Furthermore, the characteristics of grayscale values ​​are deeply analyzed, not only considering the frequency of occurrence of grayscale values, but also measuring the changes in grayscale values ​​in local areas by calculating their grayscale complexity, which can more carefully describe the distribution characteristics of grayscale values ​​in plant areas. The grayscale values ​​of the leaves of healthy plants are relatively concentrated and change more regularly, while the grayscale values ​​of the leaves with diseases or environmental stress have increased dispersion and higher complexity. Through this quantitative calculation, these differences can be captured, providing richer information for seedling judgment. Furthermore, the importance of the grayscale value can be considered together with the frequency importance and grayscale complexity, which can more comprehensively and balancedly evaluate the overall contribution of each grayscale value to the seedling judgment. Determining important and unimportant grayscale values ​​can further screen out the most valuable information for seedling identification and remove redundant data. Constructing feature vectors based on these important grayscale values ​​integrates and abstracts the key information in the image to form an effective data form that can be input into the classification model, so that the classification model can better learn and distinguish the image feature patterns of different seedling conditions. Furthermore, using the decision tree classification model to classify based on the input feature vector, it can quickly and accurately classify and judge the seedling conditions of different plant areas, and output the corresponding seedling growth category information, providing timely and reliable data support for agricultural production management.

[0012] Preferably, the frequency importance of the gray value is determined according to the following formula:

[0013]

[0014] In the formula, Gray value The frequency importance of is the serial number of the plant region, Gray value In the The frequency of occurrence in the plant area, Gray value In the The frequency of occurrence in the plant area, is the total number of plant areas, is the multiplication symbol, is the absolute value symbol.

[0015] The above technical solution can highlight the grayscale values ​​with obvious frequency changes in different areas according to the frequency differences of grayscale values ​​in various plant areas. These grayscale values ​​often have a more critical indicative role in judging the growth of seedlings, which helps to focus on key features and improve recognition accuracy.

[0016] Preferably, the method for obtaining the feature vector of each plant region based on the important gray value is:

[0017] Obtain all important grayscale values ​​of each plant region and calculate the grayscale symbiosis matrix of each plant region; remove textures corresponding to unimportant grayscale values ​​in the grayscale symbiosis matrix of each plant region and retain textures corresponding to important grayscale values; the frequency of occurrence of textures corresponding to all important grayscale values ​​in the grayscale symbiosis matrix constitutes the feature vector of the plant region.

[0018] The above technical solution removes textures corresponding to unimportant grayscale values, retains only textures corresponding to important grayscale values, and forms feature vectors based on the frequency of their appearance in the grayscale co-occurrence matrix, so that the feature vectors can better highlight the key texture features related to seedling growth. Compared with directly using all grayscale values ​​to construct feature vectors, this method can effectively reduce the interference of irrelevant information and improve the quality and representativeness of feature vectors.

[0019] Preferably, the method for determining the important grayscale value and the unimportant grayscale value according to the magnitude of the importance is:

[0020] All gray values ​​are sorted according to their importance K-MEANS clustering is performed to obtain two clusters, and the average importance of the grayscale values ​​in each cluster is calculated;

[0021] The cluster with the highest average importance is determined as the important cluster, and the minimum importance among the important clusters is used as the importance threshold;

[0022] Among all the gray values, the gray values ​​whose importance is lower than the importance threshold are regarded as unimportant gray values, and the gray values ​​whose importance is not lower than the importance threshold are regarded as important gray values.

[0023] The above technical solution realizes the screening and optimization of the grayscale features of plant images, can screen out key parts from numerous grayscale values, focus on important grayscale values ​​that are more critical to the growth of seedlings, reduce data redundancy and noise, better adapt to the complexity of plant images, and help to dig out more distinctive features, thereby improving the accuracy of seedling growth identification.

[0024] Preferably, the method for obtaining the pre-trained decision tree classification model is:

[0025] Collect historical images of plants under different seedling growth conditions during historical growth, and mark corresponding seedling growth condition categories for each historical image, wherein the seedling growth condition categories include healthy, medium and poor;

[0026] The historical images and their corresponding feature vectors are used as input data to train the classification model so that the classification model can learn the mapping relationship between the historical images and their corresponding seedling growth categories, and then the classification model can classify the seedling growth of the plants according to the input feature vectors.

[0027] The above technical solution constructs a classification model for the task of identifying seedling growth. The collected historical images cover different seedling conditions and can comprehensively reflect various possible states in the plant growth process. This enables the trained model to accurately distinguish and judge different degrees of seedling changes, providing strong technical support for precision agricultural management.

[0028] Preferably, only one plant region is extracted from each image.

[0029] Preferably, the local range of a pixel is centered on the pixel, and the size of the area around the center is The rectangular area is taken as the local range of the pixel point, where is the default value.

[0030] Preferably, the plant region in each image is extracted based on semantic segmentation, and the semantic segmentation is implemented based on a convolutional neural network CNN or a fully convolutional network FCN or a U-Net network structure.

[0031] The present invention also provides a seedling condition and growth identification system, which includes a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the steps of any one of the seedling condition and growth identification methods.

[0032] Preferably, the seedling growth recognition system further comprises an image acquisition unit, an image processing unit, a feature extraction unit and a classification unit, wherein:

[0033] The image acquisition unit is used to acquire multiple images during the growth process of the plant;

[0034] The image processing unit is used to pre-process the collected images and extract the plant area;

[0035] The feature extraction unit is used to extract the feature vector of each plant region according to the importance of the gray value;

[0036] The classification unit is used to input the feature vector into a pre-trained decision tree classification model and output the seedling growth category of each plant area.

[0037] The present invention has the following effects:

[0038] The present invention deeply mines the grayscale value features of plant images, determines their importance based on the grayscale value frequency, and accurately calculates the grayscale complexity. After the two are fused and normalized, important grayscale values ​​are accurately screened out to construct feature vectors. This process effectively quantifies the image information of the plant during its growth process, overcoming the problem that conventional technologies are difficult to accurately quantify when using decision tree classification due to changes in image features, resulting in inaccurate decision tree classification results. After the feature vector of the plant area of ​​the plant image constructed by the present invention is input into the decision tree classification model, it can accurately identify the healthy, medium, poor and other seedling growth categories, thereby improving the accuracy and stability of decision tree classification, realizing the automation and efficiency of seedling identification, and providing reliable technical support for large-scale planting and refined management of modern agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0040] Figure 1 It is a schematic flow chart of the method of the present invention;

[0041] Figure 2 is a schematic diagram of the method flow of step S31 to step S34 of the present invention;

[0042] Figure 3 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0044] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0045] Reference Figure 1 The present invention provides a method for identifying the growth of seedlings, comprising steps S1 to S5:

[0046] S1: Collect plant images and perform preprocessing to obtain plant regions.

[0047] The growth images of a single plant are collected once a day. This frequency can capture the continuous dynamics of plant growth, such as leaf extension, stem changes, color changes and other daily subtle changes, and achieve a balance between data volume and representativeness to avoid redundancy or omission of important growth information, and build a rich and orderly basic data sequence for subsequent precise seedling analysis. It is especially suitable for scenarios such as plant breeding experiments that require accurate grasp of individual growth characteristics and environmental response differences.

[0048] Since the original plant image contains a large number of background elements, such as soil, weeds and cultivation facilities, these backgrounds will interfere with the analysis based on plant characteristics and increase the complexity and difficulty of data processing. Therefore, this step uses semantic segmentation technology based on convolutional neural network CNN, full convolutional network FCN or U-Net network structure to process the collected images and segment the plant area. Each image will segment a plant area. Semantic segmentation can accurately identify and separate the plant area, so that subsequent analysis can focus on plant morphology, texture, color distribution and other characteristics, which not only improves the accuracy of analysis, but also reduces the amount of data processing and improves the overall efficiency of the seedling identification process. In addition, the "plant area" referred to in the subsequent text refers to the image corresponding to the plant area.

[0049] In summary, this step builds a high-quality data input link for the entire seedling growth recognition system through a reasonable image acquisition plan and effective preprocessing methods, which is an important prerequisite for subsequent accurate analysis and judgment of the seedling condition.

[0050] S2: Obtain the gray level co-occurrence matrix of the plant area to represent the texture characteristics of the plant area.

[0051] Gray-level co-occurrence matrix is ​​an effective tool in image texture analysis. It describes the texture information of an image by counting the frequency of gray-level combinations of pixel pairs in a specific direction and a specific distance in the image. In this step, we focus on the construction of the gray-level co-occurrence matrix in the horizontal direction (0 degree direction). The rows and columns of the gray-level co-occurrence matrix are composed of gray values, which means that each position in the matrix corresponds to a specific gray value combination.

[0052] When calculating the grayscale co-occurrence matrix of each plant region, the value at each position in the matrix records the frequency of the grayscale value combination composed of the grayscale values ​​of the row and column at that position in the horizontal direction of the plant region. For example, the element value of the Lth row and Jth column in the grayscale co-occurrence matrix is ​​K, which indicates that the frequency of the grayscale value combination (L, J) in the horizontal direction of the plant region is K, and this specific grayscale value combination reflects a texture feature of the plant region.

[0053] In an example, assuming that the value at the 5th row and 6th column of the gray-level co-occurrence matrix is ​​10, it means that the frequency of occurrence of the gray-level value combination (5, 6) in the plant area is 10, that is, the texture feature appears 10 times in the horizontal direction of the plant area.

[0054] Through comprehensive calculation and analysis of the grayscale symbiosis matrix of the entire plant area, the rich and diverse texture features of the plant area in the horizontal direction can be accurately captured. These texture features are closely related to factors such as the growth condition, tissue structure and external environmental influence of the plant, thus providing a key and reliable basis for subsequent in-depth judgment of the seedling growth condition. It helps to accurately identify the subtle changes in texture characteristics of plants in different growth stages and different health conditions, and further improve the accuracy and scientificity of seedling identification.

[0055] Next, the grayscale co-occurrence matrix of the plant image is deeply analyzed and processed through steps S3 to S5 to extract feature vectors that can effectively characterize the plant growth process and are of key significance for judging the seedling growth. The core idea is to screen out important texture-related information based on the changing rules of texture features during plant growth, and construct feature vectors to distinguish plant images at different growth stages, thereby achieving accurate identification and monitoring of seedling conditions.

[0056] S3: Evaluate the importance of each grayscale value, and filter out important grayscale values ​​according to the importance.

[0057] In this step, the importance of each gray value has the following meanings:

[0058] Able to highlight key features: During the growth of plants, the texture features corresponding to different grayscale values ​​have different importance in reflecting the growth status of plants. By determining the importance of grayscale values, it is possible to highlight key texture features that are closely related to plant growth changes, such as texture features that mark plant maturity, while ignoring some grayscale values ​​and their corresponding textures that are of little significance to growth analysis, thereby more accurately grasping the key information of plant growth.

[0059] Can reduce the amount of calculation: When processing a large number of plant images, by identifying and focusing on important grayscale values, it is possible to fuse or eliminate unimportant grayscale values ​​while retaining the key details of the plants in the image, effectively reducing the amount of data and computational complexity, and improving the efficiency of subsequent analysis and processing.

[0060] Able to better characterize growth conditions: Important grayscale values ​​often correspond to locations that undergo significant changes during plant growth, such as branches and nodes. Changes in these locations can better reflect the growth of the seedlings. Therefore, the importance of grayscale values ​​helps to more accurately obtain and characterize the dynamic changes in plant growth, providing a more reliable basis for seedling identification and growth stage distinction.

[0061] Therefore, the importance of the grayscale value is determined through steps S31 to S33.

[0062] S31: Determine the frequency importance of the gray value.

[0063] Frequency importance is an indicator that measures the importance of a gray value based on its frequency of occurrence in all plant areas.

[0064] For a grayscale value, if it appears frequently in multiple images of plant growth, it means that the grayscale value is relatively common in the entire growth process and may correspond to a relatively stable and frequently appearing part in the plant image. The grayscale value is related to some common features or key features in the plant growth process. For example, as the frequency of certain specific textures increases as the plant grows, the frequency of occurrence of the corresponding grayscale value will also increase accordingly.

[0065] For a gray value, if its frequency of appearance in certain images of plant growth increases or decreases significantly, it means that the plant position corresponding to the gray value has undergone a significant change during the growth process, resulting in a change in its frequency of appearance. This change will be reflected in the frequency importance, making it more valuable for analysis.

[0066] Specifically, the frequency of each gray value appearing in each plant area is counted, and the frequency importance of each gray value is calculated according to the frequency of each gray value appearing in each plant area:

[0067]

[0068] In the formula, Gray value The frequency importance of is the serial number of the plant region, Gray value In the The frequency of occurrence in the plant area, Gray value In the The frequency of occurrence in the plant area, is the total number of plant areas, is the multiplication symbol, is the absolute value symbol.

[0069] In this formula, Reflects the gray value In the Plant area and The difference in frequency of occurrence in the plant area, the denominator The gray value In the The frequency of occurrence in the plant area is obtained by dividing the two Gray value In the The frequency and gray value of each plant region In the The degree of change in the frequency of occurrence in a plant area, if Larger, that means the gray value The frequency of occurrence in these two adjacent plant areas has changed significantly, which implies that the plant-related characteristics corresponding to the grayscale value have changed significantly in these two adjacent plant areas. The grayscale value is more likely to be an important grayscale value that will change during the growth of the plant.

[0070] In this formula, the purpose of using the multiplication symbol and power operation is to comprehensively consider the frequency changes of each gray value in all adjacent plant areas. Through multiplication and power operation, the frequency importance of each gray value is finally obtained. This frequency importance can reflect the criticality of the gray value in the entire plant growth process as a whole. The larger the frequency importance, the more critical the gray value is in the plant growth process, and vice versa.

[0071] In summary, through such a quantitative calculation method based on the degree of change of grayscale value frequency, the dynamic change of the frequency of grayscale values ​​in images of different growth stages is converted into a measurable and comparable frequency importance index, so that in the subsequent screening out of many grayscale values, those grayscale values ​​that really play a key role in judging the growth changes of plants and the growth of seedlings can be based on this index, laying the foundation for more accurate subsequent analysis and construction of effective feature vectors.

[0072] S32: Calculate the grayscale complexity of the grayscale value.

[0073] Since frequency importance is not comprehensive enough to judge the importance of grayscale values ​​only from the frequency of their appearance in plant images, from the perspective of frequency, the greater the frequency change of a grayscale value in different regions during plant growth, the more likely it is an important grayscale value that will change during growth. However, during plant growth, leaves and other parts gradually become larger, which will cause the frequency of occurrence of certain grayscale values ​​to change. But this change may only be due to the natural growth of leaf area, rather than representing the change of the position of important branches and other locations with key and essential characteristic changes during plant growth. If the importance of grayscale values ​​is judged only by frequency importance, these grayscale values ​​whose frequency changes due to non-critical factors such as leaf growth may be misjudged as important grayscale values, but in fact they may not be so important for reflecting the key characteristics and stage changes of plant growth.

[0074] Therefore, in order to more accurately obtain the growth status of seedlings, it is necessary to distinguish which grayscale value changes are truly corresponding to changes in important positions such as key branches in plant growth, and which are only caused by interference from factors such as leaf growth. Therefore, the importance of grayscale values ​​cannot be determined solely by frequency importance, and it is necessary to further combine other factors for comprehensive judgment in order to more accurately screen out important grayscale values ​​that are truly valuable for plant growth analysis.

[0075] Specifically, for each pixel point in each plant area, the local range of the pixel point is set. The specific method is: take the pixel point as the center and take the rectangular area of ​​size N×N around the center as the local range of the pixel point, where: It is the default value, here N is set to 7 (experience value).

[0076] Next, all the pixels corresponding to each grayscale value in each plant area are obtained. One grayscale value often corresponds to many pixels, and these pixels have the same grayscale value. , get the gray value All the corresponding pixels in each plant area and calculate the grayscale complexity of the grayscale value:

[0077]

[0078] In the formula, Gray value The grayscale complexity, , Grayscale values In the The corresponding The total number of pixels contained in the local range of pixels and the standard deviation of the grayscale values ​​of the pixels contained in the local range, is the first The gray value of a pixel.

[0079] In this formula, This part is used to measure the gray value In the The corresponding The grayscale value and grayscale value of the pixel in the local range of the pixel The difference between Indicates the average difference. The larger the average difference, the The higher the difference in grayscale value between a pixel and the pixels in its local range, the more significant the grayscale change of the local environment of the pixel is relative to its own grayscale value. This may suggest that the local area is undergoing some special growth changes or structural differences, such as the local turning point of the plant texture, the junction of different tissues, etc., which makes its grayscale value deviate greatly from the surrounding area.

[0080] In this formula, The larger the value, the greater the discreteness of the grayscale values ​​of the pixels in the local area, which means that the grayscale distribution in the local area is more uneven and there are more diverse grayscale value combinations. This unevenness and diversity are often related to the complex physiological structure and growth dynamics of the plant. For example, at the growth of branches, cell differentiation, overlapping and interlacing of tissues will cause the grayscale values ​​to show greater discreteness.

[0081] In this formula, It captures the discrete degree and changing trend of the grayscale value in the local area, thereby comprehensively reflecting the structural complexity and growth activity of the local area in the plant image, providing detailed and key local feature information for further judging the growth of the plant seedlings, and helping to more accurately identify those areas that have undergone important changes in the growth process and their corresponding grayscale features, playing an important role in mining key information at the micro level in the entire seedling analysis system.

[0082] In this formula, The gray value In the All the corresponding pixels in the plant area are comprehensively considered to obtain the gray value In the The grayscale complexity in the plant area is finally calculated by The grayscale complexity in all plant regions is averaged to obtain the grayscale value This operation not only focuses on the neighborhood of a single pixel, but also realizes the feature integration from local to global through comprehensive consideration of all corresponding pixels of grayscale value in the whole plant area, and further average calculation of all plant areas. This multi-level calculation method enables the grayscale complexity of the final grayscale value to reflect its comprehensive performance in the whole plant image system, rather than being limited to a certain part or a certain plant area.

[0083] In summary, the higher grayscale complexity of a grayscale value intuitively reflects the rich, diverse and intricate relationship between it and the surrounding pixels in the local area of ​​the image. In the image structure of plants, branches and other parts often cause the grayscale values ​​of surrounding pixels to present more complex and dynamic change patterns because of the active cell division, multi-dimensional construction of tissues and continuous remodeling of morphology involved in the growth process. Therefore, the greater the grayscale complexity of a grayscale value, the stronger its correlation with key growth change areas such as branches and nodes, and it is very likely to accurately correspond to these locations that have undergone significant morphological and structural evolution during the growth process of the plant.

[0084] S33: Determine the importance of the grayscale value by comprehensively considering the importance of the grayscale value and the grayscale complexity.

[0085] Frequency importance reflects the changing characteristics of the frequency of grayscale values ​​in the plant image sequence. It can capture some overall trend information of grayscale values ​​as the plant grows, such as the frequency fluctuation of certain grayscale values ​​due to the gradual change of specific plant structures or physiological states. Grayscale complexity focuses on the complex changes of grayscale values ​​in pixels within a local range. It is extremely sensitive to the microstructure and texture changes in local areas of plant images. For example, in local areas with active growth changes such as branches, the grayscale complexity is often high.

[0086] Therefore, the frequency importance and grayscale complexity of the grayscale value are multiplied and normalized to obtain the importance of the grayscale value. The multiplication operation makes the frequency importance and grayscale complexity reinforce each other. When a grayscale value not only has obvious frequency changes in the overall image sequence, but also shows high grayscale complexity in the local area, its product will increase significantly, indicating that the grayscale value has a strong ability to represent the growth changes of the plant. The normalization process ensures that the importance of all grayscale values ​​can be compared and ranked on a unified scale, avoiding evaluation bias caused by magnitude differences.

[0087] By combining the two, integrating the macroscopic growth trend information with the microscopic structural texture change information, we can comprehensively and deeply explore the comprehensive influence of grayscale value in the plant growth process. This comprehensive consideration avoids the one-sidedness of single factor evaluation, can more accurately measure the importance of grayscale value in the plant growth process, and can accurately screen out those grayscale values ​​that are truly key indicative in the plant growth process.

[0088] S34: Filtering important grayscale values ​​and unimportant grayscale values ​​based on the importance of the grayscale values.

[0089] Specifically, all gray values ​​are sorted according to their importance. K-MEANS clustering is performed to obtain two clusters, and the average importance of the grayscale values ​​in each cluster is calculated;

[0090] The cluster with the highest average importance is determined as the important cluster, and the minimum importance among the important clusters is used as the importance threshold;

[0091] Among all the gray values, the gray values ​​whose importance is lower than the importance threshold are regarded as unimportant gray values, and the gray values ​​whose importance is not lower than the importance threshold are regarded as important gray values.

[0092] Through the above operations, the grayscale values ​​can be automatically and effectively divided into different groups based on the inherent distribution characteristics of the data. This data-driven classification method avoids the subjectivity and blindness of manually setting thresholds, making the classification results more in line with the real differences in grayscale values ​​in reflecting plant growth information.

[0093] This screening method significantly optimizes the utilization of data resources. By accurately separating important and unimportant gray values, it can effectively reduce subsequent computing and data storage requirements, avoiding the waste of resources and inefficient analysis caused by processing large amounts of irrelevant data.

[0094] When constructing feature vectors in the future, only important grayscale values ​​are retained, making the feature vectors more refined and rich in key information, and more accurately capturing the core feature changes in the plant growth process. This will directly improve the accuracy and reliability of seedling growth judgment based on these feature vectors, and help to timely and accurately discover the plant growth process.

[0095] S4: Obtain the feature vector of each plant region based on the important grayscale value.

[0096] First, all the important grayscale values ​​of each plant area are obtained. The growth process of plants contains rich and complex image information, and the important grayscale values ​​are precisely the parts that can effectively characterize key features such as their growth stage and health status. By focusing on these important grayscale values, a large amount of irrelevant grayscale interference can be eliminated, and attention can be focused on the most valuable information for judging the seedling condition.

[0097] Next, the textures corresponding to unimportant grayscale values ​​are removed from the grayscale co-occurrence matrix of each plant region, that is, the textures containing unimportant grayscale values ​​are retained, and the textures corresponding to important grayscale values ​​are retained. The textures in the grayscale co-occurrence matrix reflect the characteristic patterns of plant images under different grayscale combinations, while the textures corresponding to unimportant grayscale values ​​often have a weak indicative effect on plant growth changes and may even interfere with the judgment of key information. Through precise elimination operations, the texture information actually related to plant growth is further purified, making the subsequent constructed feature vectors more targeted and representative, and can accurately reflect the essential characteristic changes of plants during their growth process.

[0098] In an example, the unimportant grayscale values ​​of the first plant area are 60 and 100, and the following grayscale value combinations (textures) exist in the grayscale co-occurrence matrix: (10, 100), (20, 30), (30, 120), (120, 40), (110, 50), (60, 80) and (100, 130). At this time, since (10, 100), (60, 80) and (100, 130) contain unimportant grayscale values, (10, 100), (60, 80) and (100, 130) are regarded as textures corresponding to unimportant grayscale values ​​and are removed, and (20, 30), (30, 120), (120, 40) and (110, 50) are retained as textures corresponding to important grayscale values.

[0099] It is specially noted that if a gray value combination contains both unimportant gray values ​​and important gray values, it will also be eliminated. That is, as long as the gray value combination contains unimportant gray values, it means that the texture formed by the gray value combination in the plant area is less important and needs to be eliminated.

[0100] Finally, the frequency of textures corresponding to all important grayscale values ​​in the grayscale co-occurrence matrix constitutes the feature vector of the plant area. This approach quantifies and integrates the key texture information that has been screened and refined. As a digital expression, the feature vector can comprehensively and intuitively reflect the comprehensive characteristics of the plant area during the growth process. It is not just a simple data list, but also presents the texture characteristics represented by each important grayscale value during the plant growth process in a quantifiable way, providing a unified and standardized basis for subsequent data-driven seedling analysis, classification, and growth judgment.

[0101] In an example, the important grayscale values ​​of the first plant area are obtained, which are 10, 20, 30, 40 and 50, and the frequency of occurrence of these important grayscale values ​​in the plant area is counted (that is, the element value of the texture corresponding to the important grayscale value in the grayscale co-occurrence matrix). The results are as follows:

[0102] The frequency of occurrence of grayscale value 10: 0.15; the frequency of occurrence of grayscale value 20: 0.1; the frequency of occurrence of grayscale value 30: 0.25; the frequency of occurrence of grayscale value 40: 0.2; the frequency of occurrence of grayscale value 50: 0.2; then, the characteristic vector of the plant area is: [0.15, 0.1, 0.25, 0.2, 0.2].

[0103] In summary, the feature vector constructed by the frequency focuses on the texture frequency information corresponding to the important grayscale values, so that the plant regions at different growth stages and different health conditions show more obvious differences in the feature vector. For example, the texture occurrence frequency corresponding to the important grayscale values ​​of healthy plants may show a stable and orderly pattern, while the corresponding texture frequency of plants attacked by pests and diseases or poor growth will show obvious changes. This enhancement of feature discrimination helps to more accurately identify and distinguish different seedling categories using analysis methods such as decision tree classification models in the future, thereby improving the accuracy of seedling condition judgment.

[0104] S5: Use the decision tree classification model to classify the feature vectors of each plant area to obtain the seedling growth category of each plant area to achieve the identification of the seedling growth.

[0105] The decision tree classification model is a machine learning model that makes decisions based on a tree structure and is widely used in data analysis and classification tasks.

[0106] The method for obtaining the pre-trained decision tree classification model is:

[0107] Historical images of plants under different seedling growth conditions during their historical growth process were collected, and the seedling growth category of each historical image was manually marked. The seedling growth categories included healthy, medium, and poor.

[0108] According to the operations of step S1 to step S4, a feature vector of each historical image is obtained, and a training set is obtained by taking each historical image as a sample to train the decision tree model, wherein each sample of the training set includes a feature vector and a corresponding seedling growth category label;

[0109] Select the splitting criterion of the decision tree model: Common splitting criteria include information gain, Gini index and chi-square test. Here, Gini index is used as the splitting criterion. Gini index measures the probability that two samples randomly selected from a data set have inconsistent categories. The smaller the Gini index, the higher the purity of the data set. When splitting the decision tree, select the feature and splitting point with the smallest Gini index so that the sum of the Gini index of the child nodes after the split is minimized, that is, let the samples in each child node belong to the same category as much as possible, thereby improving the accuracy of classification.

[0110] During the training process, the decision tree model automatically learns the mapping relationship between feature vectors and category labels based on the pre-selected splitting criteria, and continuously adjusts the tree structure and node splitting conditions to achieve the best classification effect. When the best classification effect is achieved, the decision tree classification model training is completed. At this time, the decision tree classification model can classify the growth of plant seedlings according to the input feature vector.

[0111] Finally, the feature vector of each plant region is input into the trained decision tree classification model to obtain the classification result of the seedling growth of each plant region.

[0112] In one example, there are three plant regions, marked as region A, region B and region C. After the processing of the previous steps S1-S4, for region A, its eigenvector is [0.12, 0.08, 0.3, 0.25, 0.25]. This eigenvector is constructed based on the frequency of occurrence of textures corresponding to important gray values ​​in the image of the region in the gray level co-occurrence matrix. The eigenvector of region B is [0.18, 0.1, 0.22, 0.3, 0.2], and the eigenvector of region C is [0.05, 0.15, 0.35, 0.2, 0.25].

[0113] The eigenvector [0.12, 0.08, 0.3, 0.25, 0.25] of region A is input into the trained decision tree classification model. The model makes decisions based on the decision rules learned internally. For example, if the decision rule is: when the first eigenvalue is less than 0.15, the third eigenvalue is greater than 0.28, and the fifth eigenvalue is greater than 0.23, the seedling growth category is "medium". At this time, the seedling growth category of region A is obtained as "medium".

[0114] Similarly, the eigenvector [0.18, 0.1, 0.22, 0.3, 0.2] of region B is input into the model, and the model makes the decision based on the decision rules learned internally, such as: When the first eigenvalue is greater than 0.15, the third eigenvalue is less than 0.3, and the fourth eigenvalue is greater than 0.25, the seedling growth category is "healthy". At this time, region B is judged to be "healthy".

[0115] Similarly, the eigenvector [0.05, 0.15, 0.35, 0.2, 0.25] of region C is input into the model. The model makes the decision based on the decision rules learned internally. For example, when the first eigenvalue is less than 0.1, the third eigenvalue is greater than 0.3, and the fifth eigenvalue is less than 0.3, the seedling growth category is "poor". At this time, region C is judged to be "poor".

[0116] In this way, the classification results of seedling growth can be accurately identified so that growers or agricultural managers can take corresponding measures based on these results, such as additional fertilization and pest and disease control for poorly growing plants in area C, while maintaining routine management for healthy plants in area B and appropriately optimizing the management of plants in the growth state in area A, thereby improving the crop yield and quality of the entire planting area.

[0117] In addition, the present invention also provides a seedling condition and growth identification system, which includes a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the steps of the seedling condition and growth identification method.

[0118] Furthermore, the seedling growth recognition system, such as Figure 3 As shown in FIG. , it includes an image acquisition unit S100, an image processing unit S200, a feature extraction unit S300 and a classification unit S400. Each unit works together to form a complete seedling condition identification system. The following is a detailed introduction and interaction process:

[0119] The image acquisition unit S100 is used to collect multiple images of the plant during its growth process. As the information portal of the entire system, it is equipped with professional image acquisition equipment, such as high-resolution industrial-grade HD cameras. Such cameras have excellent optical imaging systems and can shoot plants at different time points and from multiple angles to ensure the acquisition of rich and comprehensive plant image data. At the same time, in order to adapt to different planting environments and lighting conditions, it will also be equipped with automatic fill light equipment, such as LED fill light arrays, to ensure that clear images can be taken even in dim light conditions. In addition, the image acquisition unit S100 also includes image transmission equipment, such as a high-speed wireless transmission module or a wired Ethernet network interface, which is used to stably and efficiently transmit the collected image data to the image processing unit S200.

[0120] The image processing unit S200 is used to pre-process the collected images and extract the plant area. After receiving the image from the image acquisition unit S100, the pre-processing process is started: first, the image is denoised. This process usually uses a special ISP (image signal processor) chip. The chip uses advanced filtering algorithms to effectively eliminate image noise caused by environmental factors (such as light changes, electromagnetic interference, etc.), making the image clearer and purer for subsequent accurate analysis. Then, a deep learning-based image segmentation model, such as the U-Net model, is used, which can accurately separate the plant from the background image, thereby focusing on the feature analysis of the plant itself and reducing the interference of background information on the identification and judgment of the seedling growth. The pre-processed plant area image is passed to the feature extraction unit S300 to prepare for the extraction of feature vectors.

[0121] The feature extraction unit S300 is used to extract the feature vector of each plant area according to the importance of the grayscale value, receive the plant area image transmitted by the image processing unit S200, and extract the feature vector of each plant area. Specifically, it first deeply analyzes the frequency of occurrence of each grayscale value in each plant area, calculates the frequency importance of the grayscale value through a specific formula, so as to quantify the overall importance change trend of each grayscale value during the plant growth process. Subsequently, the complexity of each grayscale value in the local range of the image is further explored, and the grayscale complexity of the grayscale value is calculated, so as to accurately capture the change characteristics of the grayscale value in the local area. The frequency importance and grayscale complexity of the grayscale value are combined and normalized to determine the importance of each grayscale value. Based on this, the feature extraction unit S300 removes the textures corresponding to unimportant grayscale values ​​in the grayscale co-occurrence matrix of each plant area, retains only the textures corresponding to important grayscale values, and integrates the frequencies of occurrence of the textures corresponding to these important grayscale values ​​in the grayscale co-occurrence matrix to construct the feature vector of the plant area. The constructed feature vector is transmitted to the classification unit S400 for final seedling classification and determination.

[0122] The classification unit S400 is used to input the feature vector into a pre-trained decision tree classification model and output the seedling growth category of each plant area. After receiving the feature vector provided by the feature extraction unit S300, it is input into the pre-trained decision tree classification model. This decision tree classification model is trained with a large amount of historical plant image data and has fully learned the intrinsic mapping relationship between the feature vector and the seedling growth category. With the help of the classification ability of the model, the classification unit S400 conducts in-depth analysis and judgment of the feature vector according to the complex decision rules within the model, and quickly and accurately outputs the seedling growth category of each plant area, such as healthy, medium or poor. These classification results provide a valuable decision-making basis for refined management in agricultural production. Planters or agricultural managers can formulate agricultural measures such as irrigation, fertilization, and pest control in a targeted manner according to the seedling conditions of different plant areas, so as to achieve efficient and accurate management of agricultural production and promote the development of the agricultural industry in an intelligent and scientific direction.

[0123] In summary, the seedling growth identification system realizes the automated and precise identification of plant seedling conditions through close and orderly interactive collaboration among various units.

[0124] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0125] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for identifying the growth of seedlings, characterized in that: include: Collect multiple images during the growth of the plant, extract the plant area in each image, and determine the frequency importance of the gray value according to the frequency of occurrence of the gray value in each plant area, wherein the frequency importance is used to reflect the criticality of the gray value in each plant area; The frequency importance of the gray value is determined according to the following formula: ； In the formula, Gray value The frequency importance of is the serial number of the plant region, Gray value In the The frequency of occurrence in the plant area, Gray value In the The frequency of occurrence in the plant area, is the total number of plant areas, is the multiplication symbol, is the absolute value symbol; Get all the pixel points corresponding to the grayscale value in each plant area, and calculate the grayscale complexity of the grayscale value: , Gray value The grayscale complexity, , Grayscale values In the The corresponding The total number of pixels contained in the local range of pixels and the standard deviation of the grayscale values ​​of the pixels contained in the local range, is the first The gray value of each pixel; The importance of the gray value is obtained by multiplying the frequency importance and the gray complexity and normalizing them, and the important gray value and the unimportant gray value are determined according to the importance, and the characteristic vector of each plant region is obtained based on the important gray value; The feature vector of each plant region is input into a pre-trained decision tree classification model to obtain the seedling growth category of each plant region to realize the identification of the seedling growth.

2. The method for identifying the growth potential of seedlings according to claim 1, characterized in that: The method for obtaining the feature vector of each plant region based on the important gray value is: Obtain all important gray values ​​of each plant region and calculate the gray level co-occurrence matrix of each plant region; In the gray-level co-occurrence matrix of each plant region, textures corresponding to unimportant gray values ​​are eliminated, and textures corresponding to important gray values ​​are retained; The frequency of textures corresponding to all important gray values ​​in the gray-level co-occurrence matrix constitutes the feature vector of the plant area.

3. The method for identifying the growth potential of seedlings according to claim 1, characterized in that: The method for determining the important grayscale value and the unimportant grayscale value according to the magnitude of the importance is: All gray values ​​are sorted according to their importance K-MEANS clustering is performed to obtain two clusters, and the average importance of the grayscale values ​​in each cluster is calculated; The cluster with the highest average importance is determined as the important cluster, and the minimum importance among the important clusters is used as the importance threshold; Among all the gray values, the gray values ​​whose importance is lower than the importance threshold are regarded as unimportant gray values, and the gray values ​​whose importance is not lower than the importance threshold are regarded as important gray values.

4. The method for identifying the growth potential of seedlings according to claim 1, characterized in that: The method for obtaining the pre-trained decision tree classification model is: Collect historical images of plants under different seedling growth conditions during the historical growth process, mark the seedling growth category of each historical image, and obtain a feature vector of each historical image, wherein the seedling growth categories include healthy, medium and poor, and use each historical image as a sample to obtain a training set to train a decision tree model, wherein each sample in the training set contains a feature vector and a corresponding seedling growth category label; During the training process, the decision tree model automatically learns the mapping relationship between feature vectors and category labels based on the pre-selected splitting criteria, and continuously adjusts the tree structure and node splitting conditions to achieve the best classification effect, thereby enabling the decision tree classification model to classify the growth of plant seedlings according to the input feature vectors.

5. The method for identifying the growth potential of seedlings according to claim 1, characterized in that: Only one plant region was extracted from each image.

6. The method for identifying the growth potential of seedlings according to claim 1, characterized in that: The local range of a pixel is a rectangular area of ​​size N×N around the pixel, with the pixel as the center. is the default value.

7. The method for identifying the growth potential of seedlings according to claim 1, characterized in that: The plant region in each image is extracted based on semantic segmentation, and the semantic segmentation is implemented based on a convolutional neural network CNN or a fully convolutional network FCN or a U-Net network structure.

8. A seedling growth identification system, characterized in that: The growth potential identification system includes a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps of the seedling growth potential identification method as described in any one of claims 1-7.

9. The seedling growth identification system according to claim 8, characterized in that: The seedling growth recognition system further includes an image acquisition unit, an image processing unit, a feature extraction unit and a classification unit, wherein: The image acquisition unit is used to acquire multiple images during the growth process of the plant; The image processing unit is used to pre-process the collected images and extract the plant area; The feature extraction unit is used to extract the feature vector of each plant region according to the importance of the gray value; The classification unit is used to input the feature vector into a pre-trained decision tree classification model and output the seedling growth category of each plant area.

Images