A plant specimen classification system and method integrating multi-scale directional texture features

Abstract

The present invention discloses a plant specimen classification system and method that integrates multi-scale directional texture features, relating to the field of image processing and analysis technology. This technical solution effectively improves the coverage rate and recognition accuracy of plant species by integrating the plant specimen classification system with multi-scale directional texture features; the acquisition and modeling module deploys multi-spectral imaging equipment to capture multi-angle images of plants, adapting to different growth stages, and improving the recognition ability of rare and regionally unique species; the multi-scale texture analysis module uses a deep learning algorithm to perform texture pattern recognition on images, effectively handling the classification of seedlings and mature stages, as well as the accurate classification of variants or hybrids; the real-time monitoring module and the classification optimization module obtain key texture parameters through image processing algorithms, realizing efficient classification of comprehensive characteristics of plant species, which is not limited to leaf morphology, but also includes characteristics such as flowers and fruits, significantly improving the recognition efficiency of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and analysis technology, and in particular to a plant specimen classification system and method integrating multi-scale directional texture features. Background Art

[0002] The technical background of plant specimen classification systems dates back to the mid-20th century and has gradually matured with the development of computers and image processing technologies. In the early days, plant classification relied primarily on the experience of botanists and physical specimens, a process that was both time-consuming and susceptible to subjective factors. With the introduction of digital technology, digital images of plant specimens can be widely stored and analyzed, greatly improving the accuracy and efficiency of classification. In recent years, the application of deep learning and machine vision technologies has enabled plant specimen classification systems to automatically identify and classify thousands of plant species. These systems are now capable of processing large-scale plant image databases, supporting global botanical research and biodiversity conservation efforts.

[0003] Although the plant specimen classification system is becoming increasingly mature, it still has the following technical shortcomings in practical application:

[0004] Incomplete species coverage: Existing systems perform poorly when dealing with rare or regionally endemic plant species because there is less training data for these species, making it difficult for the systems to accurately identify them.

[0005] Difficulty identifying seedlings and varieties: Plants vary greatly in appearance at different growth stages, especially between seedlings and mature plants, making it difficult for systems to accurately classify them. Furthermore, systems often lack sufficient ability to identify plant varieties or hybrids. Limited by specific features: Most systems rely primarily on leaf morphology for classification, and may not be able to effectively support species that rely on flowers, fruits, or other features for identification. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the present invention provides a plant specimen classification system and method that integrates multi-scale directional texture features, which solves the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a plant specimen classification system integrating multi-scale directional texture features, including an acquisition modeling module, a multi-scale texture analysis module, a real-time monitoring module, a human-computer interaction module and a classification optimization module;

[0008] The acquisition modeling module is used to deploy multispectral imaging equipment and macro camera equipment at the acquisition site; then, the plant morphological characteristics are collected as a multi-scale image dataset according to the growth stage of the plant specimen;

[0009] The multi-scale texture analysis module is used to extract directional texture features of plants from a multi-scale image dataset; and uses a deep learning algorithm to analyze the extracted texture features, build a texture feature analysis model, and identify rare and regionally endemic plant species;

[0010] The real-time monitoring module is used to monitor the plant classification process in real time and predict plant species; calculate the plant texture heterogeneity index Hxzs during the classification process; and preset a first difference A1 and a second difference A2, evaluate the plant texture heterogeneity index Hxzs, and generate an early warning instruction;

[0011] The human-computer interaction module is used to display the plant classification process in real time through the human-computer interaction interface, and after receiving the warning instruction, the user can view the plant species and texture characteristics through the human-computer interaction interface;

[0012] The classification optimization module is used to optimize the classification results of plant specimens; collect classification data of plant specimens and construct a classification data set; at the same time, extract the classification data set for deep learning analysis to obtain the plant classification efficiency index Zyzs; finally, evaluate the plant classification efficiency index Zyzs and generate an optimization strategy.

[0013] Preferably, the acquisition modeling module acquires high-definition and multi-angle images of plant specimens in real time by deploying multispectral imaging equipment and macro camera equipment at the acquisition site; first, according to the growth stage of the plant specimen, including the seedling stage, the mature stage and the flowering stage, the corresponding imaging technology and parameter settings are selected, and then the focal length and spectral settings of the camera equipment are automatically adjusted to correspond to the size and surface characteristics of different plants, thereby collecting a multi-scale image dataset containing detailed texture and color information.

[0014] Preferably, the multi-scale texture analysis module includes a texture feature extraction unit and a texture analysis unit;

[0015] The texture feature extraction unit is used to perform preliminary processing on the plant specimen images in the multi-scale image dataset, including applying edge detection and texture analysis algorithms to identify and extract the texture patterns of the plants; identifying the main texture features of the plants, including leaf veins, petal textures, and fruit skin structures, by using image processing technology, and converting the main texture features of the plants into main texture digital data.

[0016] Preferably, the texture analysis unit utilizes the converted main texture digital data and applies deep learning algorithms including convolutional neural networks to perform complex pattern recognition and establish a texture feature analysis model; then, rare and regionally endemic plant species are identified and classified based on the trained model; wherein the deep learning model adjusts the recognition accuracy and efficiency of plant species by learning a large number of labeled plant images.

[0017] Preferably, the real-time monitoring module includes a texture heterogeneity calculation unit and a heterogeneity evaluation unit;

[0018] The texture heterogeneity calculation unit is used to calculate the plant texture heterogeneity index Hxzs; by using a specific image processing algorithm and other texture analysis tools, it identifies and measures the texture detail changes on the plant surface, converts these changes into numerical indicators, and obtains the plant image contrast Con, plant image homogeneity Hom, and plant image angular second moment value Ene. Finally, the texture heterogeneity index Hxzs is calculated using the following formula:

[0019]

[0020] Preferably, the heterogeneity evaluation unit is used to preset a first difference A1 and a second difference A2, and evaluate the texture heterogeneity index Hxzs to determine whether there are potential problems or abnormalities in the texture state of the plant; and the first difference A1 is greater than the second difference A2. The specific evaluation content is as follows:

[0021] When the texture heterogeneity index Hxzs ≤ the second difference value A2, it means that the plant texture state is normal and no further action is required;

[0022] When the second difference A2 is less than the texture heterogeneity index Hxzs and less than the first difference A1, a first warning instruction is generated to prompt the user that the plant texture state is abnormal but within an acceptable range, and a detailed inspection and monitoring is performed at this time;

[0023] When the texture heterogeneity index Hxzs>the first difference A1, a second warning instruction is generated to prompt the user that there is an abnormality in the plant texture state and immediate measures need to be taken, including further diagnosis and intervention.

[0024] Preferably, the human-computer interaction module is responsible for displaying the plant classification process in real time through an integrated human-computer interaction interface, and after receiving the warning instruction from the heterogeneity assessment unit, provides a user-friendly interface for users to view plant species and their texture characteristics; first, the plant images and classification information obtained from the multi-scale texture analysis module are displayed in real time, and when the warning instruction is issued, the warning instruction is highlighted on the user-friendly interface to remind the user to pay attention to potential problems; in addition, the user further explores the specific texture features that affect the classification decision through the graphical user interface, including the specific values of the plant image contrast Con, the plant image homogeneity Hom, and the plant image angular second moment value Ene.

[0025] Preferably, the classification optimization module includes a data acquisition and calculation unit and an optimization analysis unit;

[0026] The data acquisition and calculation unit is used to collect data from the multi-scale texture analysis module and construct a classification data set; the classification data set includes various morphological and texture feature-related data of the plant. Then, by extracting the classification data set, the plant classification efficiency index Zyzs is calculated using the following formula:

[0027]

[0028] Where Lea represents the leaf edge sharpness in the classification dataset, Col represents the color distribution uniformity in the classification dataset, Ven represents the vein texture density in the classification dataset, and Fru represents the fruit shape description matching degree in the classification dataset.

[0029] Preferably, the optimization analysis unit is used to preset a plant classification performance threshold Qy, and compare and evaluate it with the plant classification efficiency index Zyzs, and finally generate a specific optimization strategy;

[0030] The specific contents are as follows:

[0031] When the plant classification efficiency index Zyzs ≥ the plant classification performance threshold Qy, it indicates that the performance of the current classification system is normal or better than expected, and no additional adjustment is required;

[0032] When the plant classification efficiency index Zyzs is less than the plant classification performance threshold Qy, it means that the performance of the current classification system is abnormal and does not meet the expected standards. At this time, optimization measures need to be taken and optimization strategies need to be generated;

[0033] The optimization strategy specifically includes:

[0034] Adjust the parameters of the classification model, including modifying the network structure, optimizing the training algorithm, and retraining the model to use more and higher-quality data;

[0035] Adjust data preprocessing steps, including image enhancement techniques and feature extraction methods;

[0036] Rebalance the training and test datasets, including increasing the number of samples from rare and hard-to-identify categories.

[0037] A plant specimen classification method integrating multi-scale directional texture features comprises the following steps:

[0038] Step 1: Deploy multispectral imaging equipment and macro camera equipment at the collection site; then collect plant morphological characteristics into multi-scale image datasets based on the growth stage of plant specimens;

[0039] Step 2: Extract directional texture features of plants from multi-scale image datasets; use deep learning algorithms to analyze the extracted texture features, build a texture feature analysis model, and identify rare and regionally endemic plant species;

[0040] Step 3: monitor the plant classification process in real time and predict the plant species; calculate the plant texture heterogeneity index Hxzs during the classification process; and preset a first difference value A1 and a second difference value A2, evaluate the plant texture heterogeneity index Hxzs, and generate an early warning instruction;

[0041] Step 4: The plant classification process is displayed in real time through the human-computer interaction interface. After receiving the warning instruction, the user can view the plant species and texture characteristics through the human-computer interaction interface;

[0042] Step 5: Optimize the classification results of plant specimens; collect the classification data of plant specimens and construct a classification dataset; at the same time, extract the classification dataset for deep learning analysis to obtain the plant classification efficiency index Zyzs; finally, evaluate the plant classification efficiency index Zyzs and generate an optimization strategy.

[0043] The present invention provides a plant specimen classification system and method that integrates multi-scale directional texture features. It has the following beneficial effects:

[0044] (1) This plant specimen classification system and method that integrates multi-scale directional texture features effectively solves the limitations of existing classification technologies in processing specific scenarios. First, to address the problem of incomplete species coverage, the system deploys multispectral imaging equipment and macro camera equipment at the collection site through the acquisition modeling module to obtain high-definition and multi-angle images of plant specimens in real time; this enables the system to select corresponding imaging technologies and parameter settings according to the growth stage of the plant specimen, including seedling stage, maturity stage and flowering stage, automatically adjust the focal length and spectral settings of the camera equipment, and collect multi-scale image datasets containing detailed texture and color information; this strategy greatly improves the ability to identify rare and regionally endemic plant species.

[0045] (2) A plant specimen classification system and method that integrates multi-scale directional texture features. To address the challenge of seedling and variant identification, the system's multi-scale texture analysis module includes a texture feature extraction unit and a texture analysis unit; the texture feature extraction unit is responsible for performing preliminary processing of plant specimen images in the multi-scale image dataset, including applying edge detection and texture analysis algorithms to identify and extract plant texture patterns; in addition, the texture analysis unit uses the converted primary texture digital data and deep learning algorithms including convolutional neural networks to perform complex pattern recognition and establish a texture feature analysis model; these steps ensure that the system can effectively identify the appearance differences of plants at different growth stages, especially the seedling stage and the mature stage, as well as accurately classify variants or hybrids. (3) A plant specimen classification system and method that integrates multi-scale directional texture features. To address the problem that the system is limited by specific features, a real-time monitoring module and a classification optimization module work together to improve the system's comprehensive recognition ability; the real-time monitoring module includes a texture heterogeneity calculation unit and a heterogeneity evaluation unit. These units use specific image processing algorithms and other texture analysis tools to identify and measure the texture detail changes on the plant surface, and convert these changes into numerical indicators to obtain plant image contrast Con, plant image homogeneity Hom, and plant image angular second-order moment value Ene; the classification optimization module collects data through the data acquisition calculation unit and uses a formula to calculate the plant classification efficiency index Zyzs; this comprehensive data analysis enables the system to not only rely on traditional leaf morphological features, but also integrate other morphological features such as flowers and fruits, thereby providing more comprehensive and accurate plant classification support. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a schematic diagram of the framework structure of a plant specimen classification system that integrates multi-scale directional texture features according to the present invention;

[0047] Figure 2 The figure is a schematic flow chart of the steps of a plant specimen classification method integrating multi-scale directional texture features according to the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] Example 1

[0050] See also Figure 1,A plant specimen classification system integrating multi-scale directional texture features, including acquisition modeling module, multi-scale texture analysis module, real-time monitoring module, human-computer interaction module and classification optimization module;

[0051] The acquisition modeling module is used to deploy multispectral imaging equipment and macro camera equipment at the acquisition site; then, the plant morphological characteristics are collected as a multi-scale image dataset according to the growth stage of the plant specimen;

[0052] The multi-scale texture analysis module is used to extract directional texture features of plants from a multi-scale image dataset; and uses a deep learning algorithm to analyze the extracted texture features, build a texture feature analysis model, and identify rare and regionally endemic plant species;

[0053] The real-time monitoring module is used to monitor the plant classification process in real time and predict plant species; calculate the plant texture heterogeneity index Hxzs during the classification process; and preset a first difference A1 and a second difference A2, evaluate the plant texture heterogeneity index Hxzs, and generate an early warning instruction;

[0054] The human-computer interaction module is used to display the plant classification process in real time through the human-computer interaction interface, and after receiving the warning instruction, the user can view the plant species and texture characteristics through the human-computer interaction interface;

[0055] The classification optimization module is used to optimize the classification results of plant specimens; collect classification data of plant specimens and construct a classification data set; at the same time, extract the classification data set for deep learning analysis to obtain the plant classification efficiency index Zyzs; finally, evaluate the plant classification efficiency index Zyzs and generate an optimization strategy.

[0056] In this embodiment, each module achieves specific beneficial effects; the acquisition modeling module can accurately collect plant morphological characteristics according to the growth stage of plant specimens by deploying multispectral imaging equipment and macro camera equipment at the acquisition site, forming a multi-scale image data set, thereby improving data coverage and classification accuracy; the multi-scale texture analysis module extracts directional texture features from these data sets, and uses deep learning algorithms such as convolutional neural networks to analyze these features and construct a texture feature analysis model, so that the system can effectively identify rare and regional plant species, thereby enhancing the generalization ability and application scope of the model; the real-time monitoring module monitors the plant classification process in real time and predicts plant species, and calculates the plant texture heterogeneity index. The system generates a warning instruction based on the preset first difference A1 and second difference A2, which effectively warns of potential classification errors and improves the response speed and safety of the system. The human-computer interaction module displays the plant classification process in real time through the human-computer interaction interface. After receiving the warning instruction, the user can directly view the plant species and their texture characteristics through the interface, which improves the user's operation convenience and the interactivity of the system. The classification optimization module collects the classification data of plant specimens, constructs the classification data set and performs deep learning analysis to obtain the plant classification efficiency index Zyzs, evaluates the index and generates an optimization strategy to continuously improve the classification performance, ensure that the system continues to adapt to new classification challenges, and improve the overall efficiency and accuracy.

[0057] Example 2

[0058] The acquisition modeling module acquires high-definition, multi-angle images of plant specimens in real time by deploying multispectral imaging equipment and macro cameras at the acquisition site. It first selects corresponding imaging techniques and parameter settings based on the growth stages of the plant specimens, including seedling, maturity, and flowering. It then automatically adjusts the focal length and spectral settings of the cameras to correspond to the size and surface characteristics of different plants, thereby collecting a multi-scale image dataset containing detailed texture and color information.

[0059] The multi-scale texture analysis module includes a texture feature extraction unit and a texture analysis unit;

[0060] The texture feature extraction unit is used to perform preliminary processing on the plant specimen images in the multi-scale image dataset, including applying edge detection and texture analysis algorithms to identify and extract the texture patterns of the plants; identifying the main texture features of the plants, including leaf veins, petal textures, and fruit skin structures, by using image processing technology, and converting the main texture features of the plants into main texture digital data.

[0061] The texture analysis unit uses the converted primary texture digital data and employs deep learning algorithms, including convolutional neural networks, to perform complex pattern recognition and establish a texture feature analysis model. Rare and regionally endemic plant species are then identified and classified based on the trained model. The deep learning model adjusts the accuracy and efficiency of plant species recognition by learning from a large number of labeled plant images.

[0062] In this embodiment, the acquisition and modeling module uses multispectral imaging equipment and macro cameras to acquire high-definition, multi-angle images of plant specimens in real time. It selects appropriate imaging techniques and parameter settings based on the growth stages of the plant specimens, such as seedling, maturity, and flowering, and automatically adjusts the focal length and spectral settings to accommodate the size and surface characteristics of different plants, thereby efficiently collecting a multi-scale image dataset containing rich texture and color information. This process ensures the comprehensiveness and diversity of the data, providing high-quality input for subsequent texture analysis.

[0063] The multi-scale texture analysis module includes a texture feature extraction unit and a texture analysis unit. The texture feature extraction unit performs preliminary processing on plant specimens in the image dataset through edge detection and texture analysis algorithms, effectively identifying key texture features such as leaf veins, petal texture and fruit skin structure, and converting these features into digital data; the texture analysis unit uses these main texture digital data and deep learning algorithms such as convolutional neural networks to perform complex pattern recognition, establish a texture feature analysis model, and then accurately identify and classify rare and regionally unique plant species; the work of this module not only improves the system's recognition ability, but also enhances its ability to respond to plant diversity, making the system more powerful and flexible in practical applications; overall, by efficiently integrating and processing multi-scale image data and optimizing the recognition process through advanced image processing and deep learning technologies, the accuracy and operational efficiency of plant classification have been significantly improved, demonstrating great potential in biodiversity research and practical applications.

[0064] Example 3

[0065] The real-time monitoring module includes a texture heterogeneity calculation unit and a heterogeneity evaluation unit;

[0066] The texture heterogeneity calculation unit is used to calculate the plant texture heterogeneity index Hxzs; by using a specific image processing algorithm and other texture analysis tools, it identifies and measures the texture detail changes on the plant surface, converts these changes into numerical indicators, and obtains the plant image contrast Con, plant image homogeneity Hom, and plant image angular second moment value Ene. Finally, the texture heterogeneity index Hxzs is calculated using the following formula:

[0067]

[0068] The heterogeneity evaluation unit is used to preset a first difference A1 and a second difference A2, and evaluate the texture heterogeneity index Hxzs to determine whether there are potential problems or abnormalities in the texture state of the plant; and the first difference A1 is greater than the second difference A2. The specific evaluation content is as follows:

[0069] When the texture heterogeneity index Hxzs ≤ the second difference value A2, it means that the plant texture state is normal and no further action is required;

[0070] When the second difference A2 is less than the texture heterogeneity index Hxzs and less than the first difference A1, a first warning instruction is generated to prompt the user that the plant texture state is abnormal but within an acceptable range, and a detailed inspection and monitoring is performed at this time;

[0071] When the texture heterogeneity index Hxzs>the first difference A1, a second warning instruction is generated to prompt the user that there is an abnormality in the plant texture state and immediate measures need to be taken, including further diagnosis and intervention.

[0072] The human-computer interaction module is responsible for displaying the plant classification process in real time through an integrated human-computer interaction interface. After receiving a warning instruction from the heterogeneity assessment unit, it provides a user-friendly interface for users to view plant species and their texture characteristics. It first displays plant images and classification information obtained from the multi-scale texture analysis module in real time. When a warning instruction is issued, the warning instruction is highlighted on the user-friendly interface to alert users to potential problems. In addition, the user can further explore the specific texture features that affect the classification decision through a graphical user interface, including the specific values of plant image contrast (Con), plant image homogeneity (Hom), and plant image angular second-order moment (Ene).

[0073] The classification optimization module includes a data acquisition and calculation unit and an optimization analysis unit;

[0074] The data acquisition and calculation unit is used to collect data from the multi-scale texture analysis module and construct a classification data set; the classification data set includes various morphological and texture feature-related data of the plant. Then, by extracting the classification data set, the plant classification efficiency index Zyzs is calculated using the following formula:

[0075]

[0076] Where Lea represents the leaf edge sharpness in the classification dataset, Col represents the color distribution uniformity in the classification dataset, Ven represents the vein texture density in the classification dataset, and Fru represents the fruit shape description matching degree in the classification dataset.

[0077] The optimization analysis unit is used to preset the plant classification performance threshold Qy, and compare and evaluate it with the plant classification efficiency index Zyzs, and finally generate a specific optimization strategy;

[0078] The specific contents are as follows:

[0079] When the plant classification efficiency index Zyzs ≥ the plant classification performance threshold Qy, it indicates that the performance of the current classification system is normal or better than expected, and no additional adjustment is required;

[0080] When the plant classification efficiency index Zyzs is less than the plant classification performance threshold Qy, it means that the performance of the current classification system is abnormal and does not meet the expected standards. At this time, optimization measures need to be taken and optimization strategies need to be generated;

[0081] The optimization strategy specifically includes:

[0082] Adjust the parameters of the classification model, including modifying the network structure, optimizing the training algorithm, and retraining the model to use more and higher-quality data;

[0083] Adjust data preprocessing steps, including image enhancement techniques and feature extraction methods;

[0084] Rebalance the training and test datasets, including increasing the number of samples from rare and hard-to-identify categories.

[0085] In this embodiment, the accuracy and operational efficiency of plant identification are greatly improved through the coordinated work of carefully designed modules. The real-time monitoring module includes a texture heterogeneity calculation unit and a heterogeneity evaluation unit. The texture heterogeneity calculation unit calculates the plant texture heterogeneity index Hxzs through a specific image processing algorithm and other texture analysis tools, effectively identifies and measures the texture detail changes on the plant surface, and converts these changes into numerical indicators to obtain the plant image contrast Con, plant image homogeneity Hom, and plant image angular second moment value Ene. The calculation of these parameters helps the system accurately evaluate the texture state of the plant. The heterogeneity evaluation unit evaluates Hxzs based on the preset first difference A1 and second difference A2, effectively monitoring and warning potential problems or abnormalities in the texture state. The human-computer interaction module displays the classification process and warning information in real time through an integrated interface. Enhance the user interaction experience, allowing users to directly observe and analyze the key texture features that affect classification decisions; the classification optimization module collects and processes data from the multi-scale texture analysis module through the data acquisition and calculation unit to construct a classification data set, which includes leaf edge sharpness Lea, color distribution uniformity Col, vein texture density Ven and fruit shape description matching degree Fru. These data are calculated through a formula to form the plant classification efficiency index Zyzs. The optimization analysis unit compares and evaluates the preset plant classification performance threshold Qy and Zyzs to ensure that the system performance meets the standards or proposes optimization strategies, including network structure adjustment, training algorithm optimization and data set balance, to deal with rare and difficult-to-identify species. This integrated strategy not only optimizes classification accuracy, but also enhances the adaptability and foresight of the system, ensuring efficient and accurate classification performance in various environments.

[0086] Example 4

[0087] See also Figure 2 A plant specimen classification method integrating multi-scale directional texture features includes the following steps:

[0088] Step 1: Deploy multispectral imaging equipment and macro camera equipment at the collection site; then collect plant morphological characteristics into multi-scale image datasets based on the growth stage of plant specimens;

[0089] Step 2: Extract directional texture features of plants from multi-scale image datasets; use deep learning algorithms to analyze the extracted texture features, build a texture feature analysis model, and identify rare and regionally endemic plant species;

[0090] Step 3: monitor the plant classification process in real time and predict the plant species; calculate the plant texture heterogeneity index Hxzs during the classification process; and preset a first difference value A1 and a second difference value A2, evaluate the plant texture heterogeneity index Hxzs, and generate an early warning instruction;

[0091] Step 4: The plant classification process is displayed in real time through the human-computer interaction interface. After receiving the warning instruction, the user can view the plant species and texture characteristics through the human-computer interaction interface;

[0092] Step 5: Optimize the plant specimen classification results; collect classification data of the plant specimens and construct a classification dataset; extract the classification dataset for deep learning analysis to obtain the plant classification efficiency index Zyzs; finally, evaluate the plant classification efficiency index Zyzs and generate an optimization strategy. Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A plant specimen classification system integrating multi-scale directional texture features, characterized by: It includes acquisition modeling module, multi-scale texture analysis module, real-time monitoring module, human-computer interaction module and classification optimization module; The acquisition modeling module is used to deploy multispectral imaging equipment and macro camera equipment at the acquisition site; then, the plant morphological characteristics are collected as a multi-scale image dataset according to the growth stage of the plant specimen; The multi-scale texture analysis module is used to extract directional texture features of plants from a multi-scale image dataset; Deep learning algorithms are used to analyze the extracted texture features and build a texture feature analysis model to identify rare and regionally endemic plant species. The real-time monitoring module is used to monitor the plant classification process in real time and predict plant species; during the classification process, the texture heterogeneity index Hxzs is calculated using the following formula: Where Con represents the contrast of plant images, Hom represents the homogeneity of plant images, and Ene represents the second-order angular moment of plant images. Then, the first difference A1 and the second difference A2 are preset to evaluate the plant texture heterogeneity index Hxzs and generate an early warning instruction. The human-computer interaction module is used to display the plant classification process in real time through the human-computer interaction interface, and after receiving the warning instruction, the user can view the plant species and texture characteristics through the human-computer interaction interface; The classification optimization module is used to optimize the classification results of plant specimens; collect classification data of plant specimens and construct a classification data set; extract the classification data set for deep learning analysis, and obtain the plant classification efficiency index Zyzs by calculation. The specific calculation formula is as follows: Where Lea represents the sharpness of the leaf edge, Col represents the uniformity of color distribution, Ven represents the density of vein texture, and Fru represents the matching degree of fruit shape description. Finally, the plant classification efficiency index Zyzs is evaluated and an optimization strategy is generated.

2. The plant specimen classification system integrating multi-scale directional texture features according to claim 1, characterized in that: The acquisition modeling module acquires high-definition, multi-angle images of plant specimens in real time by deploying multispectral imaging equipment and macro cameras at the acquisition site. It first selects corresponding imaging techniques and parameter settings based on the growth stages of the plant specimens, including seedling, maturity, and flowering. It then automatically adjusts the focal length and spectral settings of the cameras to correspond to the size and surface characteristics of different plants, thereby collecting a multi-scale image dataset containing detailed texture and color information.

3. The plant specimen classification system integrating multi-scale directional texture features according to claim 2, characterized in that: The multi-scale texture analysis module includes a texture feature extraction unit and a texture analysis unit; The texture feature extraction unit is used to perform preliminary processing on the plant specimen images in the multi-scale image dataset, including applying edge detection and texture analysis algorithms to identify and extract the texture patterns of the plants; identifying the main texture features of the plants, including leaf veins, petal textures, and fruit skin structures, by using image processing technology, and converting the main texture features of the plants into main texture digital data.

4. The plant specimen classification system integrating multi-scale directional texture features according to claim 3, characterized in that: The texture analysis unit uses the converted primary texture digital data and employs deep learning algorithms, including convolutional neural networks, to perform complex pattern recognition and establish a texture feature analysis model. Rare and regionally endemic plant species are then identified and classified based on the trained model. The deep learning model adjusts the accuracy and efficiency of plant species recognition by learning from a large number of labeled plant images.

5. The plant specimen classification system integrating multi-scale directional texture features according to claim 4, characterized in that: The real-time monitoring module includes a texture heterogeneity calculation unit and a heterogeneity evaluation unit; The texture heterogeneity calculation unit is used to calculate the plant texture heterogeneity index Hxzs; by identifying and measuring the texture detail changes on the plant surface, and converting these changes into numerical indicators, and obtaining the plant image contrast Con, plant image homogeneity Hom and plant image angular second moment value Ene.

6. The plant specimen classification system integrating multi-scale directional texture features according to claim 5, characterized in that: The heterogeneity evaluation unit is used to preset a first difference A1 and a second difference A2, and evaluate the texture heterogeneity index Hxzs to determine whether there are potential problems or abnormalities in the texture state of the plant; and the first difference A1 is greater than the second difference A2. The specific evaluation content is as follows: When the texture heterogeneity index Hxzs ≤ the second difference value A2, it means that the plant texture state is normal and no further action is required; When the second difference A2 is less than the texture heterogeneity index Hxzs and less than the first difference A1, a first warning instruction is generated to prompt the user that the plant texture state is abnormal but within an acceptable range, and a detailed inspection and monitoring is performed at this time; When the texture heterogeneity index Hxzs>the first difference A1, a second warning instruction is generated to prompt the user that there is an abnormality in the plant texture state and immediate measures need to be taken, including further diagnosis and intervention.

7. The plant specimen classification system integrating multi-scale directional texture features according to claim 6, characterized in that: The human-computer interaction module is responsible for displaying the plant classification process in real time through an integrated human-computer interaction interface. After receiving a warning instruction from the heterogeneity assessment unit, it provides a user-friendly interface for users to view plant species and their texture characteristics. It first displays plant images and classification information obtained from the multi-scale texture analysis module in real time. When a warning instruction is issued, the warning instruction is highlighted on the user-friendly interface to alert users to potential problems. In addition, the user can further explore the specific texture features that affect the classification decision through a graphical user interface, including the specific values of plant image contrast (Con), plant image homogeneity (Hom), and plant image angular second-order moment (Ene).

8. The plant specimen classification system integrating multi-scale directional texture features according to claim 7, characterized in that: The classification optimization module includes a data acquisition and calculation unit and an optimization analysis unit; The data acquisition and calculation unit is used to collect data from the multi-scale texture analysis module and construct a classification data set; The classification dataset contains various morphological and texture feature-related data of plants. Then, the classification dataset is extracted, including leaf edge sharpness Lea, color distribution uniformity Col, vein texture density Ven, and fruit shape description matching degree Fru; finally, the plant classification efficiency index Zyzs is obtained by calculation.

9. The plant specimen classification system integrating multi-scale directional texture features according to claim 8, characterized in that: The optimization analysis unit is used to preset the plant classification performance threshold Qy, and compare and evaluate it with the plant classification efficiency index Zyzs, and finally generate a specific optimization strategy; The specific contents are as follows: When the plant classification efficiency index Zyzs ≥ the plant classification performance threshold Qy, it indicates that the performance of the current classification system is normal or better than expected, and no additional adjustment is required; When the plant classification efficiency index Zyzs is less than the plant classification performance threshold Qy, it means that the performance of the current classification system is abnormal and does not meet the expected standards. At this time, optimization measures need to be taken and optimization strategies need to be generated; The optimization strategy specifically includes: Adjust the parameters of the classification model, including modifying the network structure, optimizing the training algorithm, and retraining the model to use more and higher-quality data; Adjust data preprocessing steps, including image enhancement techniques and feature extraction methods; Rebalance the training and test datasets, including increasing the number of samples from rare and hard-to-identify categories.

10. A method for classifying plant specimens by integrating multi-scale directional texture features, according to a plant specimen classification system by integrating multi-scale directional texture features according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Deploy multispectral imaging equipment and macro camera equipment at the collection site; then collect plant morphological characteristics into multi-scale image datasets based on the growth stage of plant specimens; Step 2: Extract directional texture features of plants from multi-scale image datasets; use deep learning algorithms to analyze the extracted texture features, build a texture feature analysis model, and identify rare and regionally endemic plant species; Step 3: Monitor the plant classification process in real time and predict the plant species; during the classification process, calculate the texture heterogeneity index Hxzs using the following formula: Where Con represents the contrast of plant images, Hom represents the homogeneity of plant images, and Ene represents the second-order angular moment of plant images. Then, the first difference A1 and the second difference A2 are preset to evaluate the plant texture heterogeneity index Hxzs and generate an early warning instruction. Step 4: The plant classification process is displayed in real time through the human-computer interaction interface. After receiving the warning instruction, the user can view the plant species and texture characteristics through the human-computer interaction interface; Step 5: Optimize the classification results of plant specimens; collect classification data of plant specimens and construct a classification data set; extract the classification data set for deep learning analysis, and calculate the plant classification efficiency index Zyzs. The specific calculation formula is as follows: Where Lea represents the sharpness of the leaf edge, Col represents the uniformity of color distribution, Ven represents the density of vein texture, and Fru represents the matching degree of fruit shape description. Finally, the plant classification efficiency index Zyzs is evaluated and an optimization strategy is generated.

Images