Method and system for monitoring health status of garden plants based on image analysis

Abstract

The application discloses a garden plant health state monitoring method based on image analysis and belongs to the technical field of image analysis and plant health monitoring, and comprises the following steps: collecting reflection images and polarized light images of the same plant area under multiple wave bands; calculating the penetration depth of each wave band in the plant medium based on a radiation transmission model and a plant optical parameter library; enhancing the reflection images by adopting a multi-scale Retinex algorithm, removing the specular reflection interference by utilizing the polarized light images, and obtaining corrected images; extracting abnormal texture areas of the corrected images, combining the penetration depth information, and constructing a multispectral feature fusion matrix; based on the fusion matrix, shallow normal and deep abnormal areas and areas with abnormal layers are compared and analyzed, internal abnormal classification discrimination rules are executed, the tomographic distribution of the plant internal health state is determined, and a tomographic imaging diagram distinguishing areas by different colors is generated, so that nondestructive, non-contact and accurate detection of the internal hollow rot of the tree trunk is realized, and the detection result is intuitive and visual.

Description

Technical Field

[0001] This invention relates to a method for monitoring the health status of garden plants, and more particularly to a method and system for monitoring the health status of garden plants based on image analysis. Background Technology

[0002] As an important component of the urban ecosystem, garden plants not only beautify the environment and purify the air, but also carry the historical and cultural memory of the city. Newly transplanted precious trees and ancient trees, in particular, have extremely high conservation value. The internal health of a tree trunk directly affects its survival ability and wind resistance. However, internal diseases such as xylem rot, cavity formation, and insect tunnels of wood-boring pests are often highly concealed and difficult to detect. By the time obvious external symptoms appear on the trunk, the internal structure has usually already suffered irreversible and severe damage, leading to a significant increase in tree mortality. Therefore, how to achieve early and accurate diagnosis of the internal health of tree trunks without damage has always been a pressing technical challenge in the field of garden maintenance.

[0003] Currently, methods for detecting the internal health of tree trunks mainly rely on manual experience and physical detection. The traditional method of tapping and listening to the sound relies on maintenance personnel tapping the trunk and subjectively judging the presence of cavities based on the clarity or dullness of the echo. This method is highly dependent on personal experience, lacks quantitative standards, and struggles to pinpoint the specific range and depth of areas prone to rot. While drilling can directly obtain internal samples or measure wood density changes using a resistance meter, its invasive nature causes secondary damage to the tree, and the wounds can become entry points for pests and diseases, making it unsuitable for the regular monitoring of valuable trees.

[0004] In recent years, with the development of non-destructive testing technology, techniques such as stress wave imaging and electrical impedance tomography have been attempted to be applied to the detection of internal defects in trees. However, these methods require the deployment of multiple sensor probes around the trunk, resulting in complex equipment installation, low detection efficiency, and difficulty in achieving rapid, non-contact screening over long distances. While visual image-based detection methods have achieved non-contact monitoring, existing technologies are mainly limited to visible light image analysis of the tree trunk bark, unable to penetrate the xylem to obtain internal structural information, and making it difficult to establish a quantitative correlation model between bark appearance features and internal decay status. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method and system for monitoring the health status of garden plants based on image analysis.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for monitoring the health status of garden plants based on image analysis, comprising the following steps:

[0007] S1. Acquire reflection images of the same area of ​​the same plant under multiple preset different wavelengths, and simultaneously acquire polarized light images of the same area of ​​the same plant.

[0008] S2. Based on the radiative transfer model and combined with the optical attenuation coefficients of plants in different bands pre-stored in the plant optical parameter library, the penetration depth of the reflected image in the plant medium is calculated respectively, and the penetration depth information corresponding to multiple different bands is obtained.

[0009] S3. The multi-scale Retinex algorithm is used to enhance the reflection image, and the specular reflection component is removed from the enhanced reflection image using polarized light image to obtain a reflection image II with non-skin reflection interference removed.

[0010] S4. Extract the abnormal texture regions of the reflection image II, and construct a multispectral feature fusion matrix based on the abnormal texture regions and the penetration depth information;

[0011] S5. Based on the multispectral feature fusion matrix, by comparing and analyzing the areas where the shallow images are normal and the deep images are abnormal, as well as the areas where all images are abnormal, internal anomaly classification and discrimination rules are executed to determine the chromatographic distribution of the plant's internal health status.

[0012] S6. Based on the tomographic distribution, generate a tomographic image of the plant's internal health status, and distinguish healthy plant areas, areas with initial rot, and severely rotten areas with different colors.

[0013] In a preferred embodiment of the present invention, the reflectance image includes: a red band reflectance image, a near-infrared band reflectance image, and a short-wave infrared band reflectance image of the same area of ​​the same plant, which are simultaneously acquired by a multispectral camera.

[0014] Polarized light images of the same region of the same plant in orthogonal polarization directions are acquired synchronously by a polarization camera and a multispectral camera to characterize the non-Near reflectance components of the plant epidermis.

[0015] In a preferred embodiment of the present invention, the process of obtaining the penetration depth information includes:

[0016] S201. Obtain the first attenuation coefficient in the visible red band, the second attenuation coefficient in the near-infrared band, and the third attenuation coefficient in the short-wave infrared band from the plant optical parameter library that match the plant medium.

[0017] S202. Based on the exponential decay law of the radiative transfer model, calculate the first penetration depth value corresponding to the red band reflection image using the first attenuation coefficient, calculate the second penetration depth value corresponding to the near-infrared band reflection image using the second attenuation coefficient, and calculate the third penetration depth value corresponding to the short-wave infrared band reflection image using the third attenuation coefficient.

[0018] The first penetration depth value is less than the second penetration depth value, and the second penetration depth value is less than the third penetration depth value.

[0019] In a preferred embodiment of the present invention, the process of enhancing the reflected image includes:

[0020] S301. Decompose reflection images under multiple different spectral bands into illumination component images and reflection component images of multiple different scales;

[0021] S302. Perform contrast stretching on the reflection component image at each scale, and then perform weighted fusion on the processed reflection component images at multiple scales to obtain the enhanced reflection image for the corresponding band.

[0022] The scale parameters of the multi-scale Retinex algorithm are adaptively adjusted based on the penetration depth information of the corresponding band, and the penetration depth is positively correlated with the scale parameters used.

[0023] In a preferred embodiment of the present invention, the process of acquiring the reflection image II includes:

[0024] S311. Perform polarization degree analysis on the polarized light image to generate a polarization degree distribution map characterizing the specular reflection intensity of the plant epidermis;

[0025] S312. Based on the polarization degree distribution map, subtract the specular reflection component from the enhanced reflection images of each band to obtain the reflection image II of the corresponding band.

[0026] In a preferred embodiment of the present invention, the extraction of the abnormal texture region includes:

[0027] S401. Perform gray-level co-occurrence matrix calculation on the reflection image II of each band to obtain the texture second-order statistics of the reflection image II of each band. The texture second-order statistics include angular second moment, contrast, correlation and entropy.

[0028] S402. Compare the second-order statistics of the texture with the reference threshold of healthy plant texture in the corresponding band, and mark the pixel area that exceeds the reference threshold as a candidate abnormal texture area.

[0029] S403. Perform connected component analysis on the candidate abnormal texture regions, and select connected components with an area greater than a preset area threshold as the abnormal texture regions.

[0030] In a preferred embodiment of the present invention, the process of constructing the multispectral feature fusion matrix includes:

[0031] S411. Mark the abnormal texture region of the red band reflection image II as a first spatial location set, and associate it with the first penetration depth value;

[0032] S412. Mark the abnormal texture region of the near-infrared band reflection image II as a second spatial location set, and associate it with the second penetration depth value;

[0033] S413. Mark the abnormal texture region of the shortwave infrared band reflection image II as a third spatial location set, and associate it with the third penetration depth value;

[0034] S414. Overlay the first spatial location set, the second spatial location set, and the third spatial location set in a unified spatial coordinate system to generate the multispectral feature fusion matrix.

[0035] In a preferred embodiment of the present invention, the process of executing the internal anomaly classification and discrimination rules includes:

[0036] S501. When no abnormal texture area is detected in the red band reflection image II and the near-infrared band reflection image II, but an abnormal texture area exists in the short-wave infrared band reflection image II, it is determined to be a deep-layer cavitation area.

[0037] S502. When no abnormal texture area is detected in the red band reflection image II, but abnormal texture areas are detected in the near-infrared band reflection image II and the short-wave infrared band reflection image II, it is determined to be a decayed area of ​​the middle to deep layer of the material.

[0038] S503. When abnormal texture areas are detected in the red band reflection image II, near-infrared band reflection image II and short-wave infrared band reflection image II, they are determined to be severe cavitation areas or epidermal lesion areas that run through the entire mass.

[0039] In a preferred embodiment of the present invention, an image analysis-based garden plant health status monitoring system is provided, which is used to implement the garden plant health status monitoring method.

[0040] In a preferred embodiment of the present invention, the garden plant health status monitoring system includes: an image acquisition module, a penetration depth calculation module, an image enhancement and correction module, a feature extraction and fusion module, a tomographic distribution discrimination module, and an imaging output module;

[0041] The image acquisition module is used to execute S1, and its output is connected to the input of the penetration depth calculation module and the image enhancement and correction module, respectively.

[0042] The penetration depth calculation module is used to execute S2. Its input end receives the reflection image output by the image acquisition module, and its output end is connected to the input end of the feature extraction and fusion module.

[0043] The image enhancement and correction module is used to execute S3. Its input end receives the reflected image and polarized light image output by the image acquisition module, and its output end is connected to the input end of the feature extraction and fusion module.

[0044] The feature extraction and fusion module is used to execute S4. Its input end receives the penetration depth information output by the penetration depth calculation module and the reflection image II output by the image enhancement and correction module, respectively. Its output end is connected to the input end of the tomographic distribution discrimination module.

[0045] The tomographic distribution discrimination module is used to execute S5. Its input end receives the multispectral feature fusion matrix output by the feature extraction and fusion module, and its output end is connected to the input end of the imaging output module.

[0046] The imaging output module is used to execute S6, and its input end receives the tomographic distribution information output by the tomographic distribution discrimination module and generates a tomographic image.

[0047] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0048] (1) This invention achieves non-invasive and non-contact precise detection of the internal health status of tree trunks, overcoming the shortcomings of traditional methods that cause secondary damage to trees or rely on subjective human judgment. In view of the problems mentioned in the background art, such as the strong subjectivity of the tapping and listening method and the large destructiveness of the drilling and detection method, this invention establishes an optical tomography method that can obtain internal structural information without contacting the tree by collecting multispectral reflectance images and polarized light images of the tree trunk surface and utilizing the differences in the penetration depth of different spectral bands in the wood medium. Through this layered imaging mechanism, this invention can realize the location and assessment of the hollow and rotten areas inside the tree trunk without damaging the tree body, and is particularly suitable for the regular health monitoring of newly transplanted precious trees and ancient and famous trees.

[0049] (2) This invention achieves precise localization and early warning of hollow decay areas inside tree trunks by constructing a multispectral feature fusion matrix and internal anomaly classification and discrimination rules, significantly improving the accuracy and reliability of detection. Addressing the problem that existing visual image technologies cannot penetrate the xylem to obtain internal structural information, this invention first uses a multi-scale Retinex algorithm to enhance images of different wavelengths, and combines polarized light images to remove epidermal specular reflection interference, extracting pure internal scattering information; then, it calculates the second-order statistics of texture using a gray-level co-occurrence matrix to identify abnormal texture areas in each wavelength image; finally, it constructs a multispectral feature fusion matrix based on penetration depth information, and by comparing and analyzing areas where shallow images are normal but deep images are abnormal, as well as areas where all images are abnormal, it finely classifies the internal health status of tree trunks into different levels such as deep hollow decay, middle-layer decay, and severe penetrating hollow decay. This layered discrimination method based on the principle of optical tomography can not only detect early internal lesions that are invisible to the naked eye, but also accurately determine the depth and severity of the lesions, providing a scientific basis for decision-making by garden maintenance personnel.

[0050] (3) Compared with the stress wave imaging and electrical impedance tomography methods mentioned in the background art, which require multiple sensor probes to be arranged around the tree trunk, resulting in complex equipment installation and low detection efficiency, the present invention only requires one multispectral camera and one polarization camera to complete image acquisition. The detection process is fast and convenient, and it is suitable for rapid screening of large-scale garden plants. At the same time, the present invention finally outputs color tomographic images that distinguish healthy areas, areas with initial decay, and areas with severe hollow decay with different colors. The results are intuitive and easy to understand, and even non-professionals can quickly understand the health status of trees. In addition, the method of the present invention can be integrated into drones or handheld detection devices to realize long-distance monitoring of tall trees, providing a new type of efficient, economical, and non-destructive technical means for urban landscaping maintenance. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a preferred embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0055] This invention relates to a method and system for monitoring the health status of garden plants based on image analysis, particularly suitable for the non-destructive detection of internal hollow rot in tree trunks. The method combines multispectral imaging and polarization imaging, utilizing the differences in penetration depth of different spectral bands in the plant medium, and combining image enhancement and texture analysis techniques to achieve optical tomographic imaging of the plant's internal health status.

[0056] The method is executed by a computer system or an embedded image processing device, which includes a processor, a memory, and an image acquisition interface. The memory stores computer program instructions, and when the instructions are executed by the processor, the method described in this invention is implemented.

[0057] like Figure 1 As shown, the image analysis-based method for monitoring the health status of garden plants includes the following steps:

[0058] S1. Acquire reflection images of the same area of ​​the same plant under multiple preset different wavelengths, and simultaneously acquire polarized light images of the same area of ​​the same plant.

[0059] Furthermore, the reflection images include: red band reflection images, near-infrared band reflection images, and short-wave infrared band reflection images of the same area of ​​the same plant, which are simultaneously acquired by a multispectral camera.

[0060] Polarized light images of the same region of the same plant in orthogonal polarization directions are acquired synchronously by a polarization camera and a multispectral camera to characterize the non-Near reflectance components of the plant epidermis.

[0061] In practice, the first step is to acquire images of the target plant's trunk (such as valuable trees in urban gardens) and the area to be measured. This step is accomplished through the combined use of multispectral imaging and polarization imaging equipment.

[0062] As a preferred embodiment, an imaging system integrating a multispectral camera and a polarization camera is employed. The multispectral camera is equipped with three independent optical channels, corresponding to the visible red band with a wavelength of 660 nm, the near-infrared band with a wavelength of 850 nm, and the short-wave infrared band with a wavelength of 1450 nm, respectively.

[0063] The selection of these bands is based on research on the optical properties of plant xylem: the 660nm band has a shallow penetration depth and mainly reflects information about the epidermis and phloem; the 850nm band has a medium penetration depth and can reflect information about the outer xylem; and the 1450nm band has a deep penetration depth and can reflect information about the deep xylem.

[0064] Under the control of the same trigger signal, the multispectral camera simultaneously acquires reflection images of these three bands, denoted as follows: , and ;in, Represents the pixel coordinates of the image.

[0065] Simultaneously, a polarization camera and a multispectral camera synchronously acquire polarized light images of the same tree trunk region. The polarization camera is equipped with an orthogonal polarization analyzer, which acquires images of two orthogonal polarization directions (0° and 90°) by rotating a polarizer or using a beam splitter, and calculates the degree of polarization. Polarized light images are primarily used to characterize the non-Snell's reflection component of plant epidermis, because specular reflection from the epidermis has polarization characteristics, while internal scattered light is usually depolarized. By acquiring polarized light images, specular reflection interference from the epidermis can be effectively eliminated in subsequent steps, allowing for the extraction of true internal scattering information.

[0066] S2. Based on the radiative transfer model and combined with the optical attenuation coefficients of plants in different bands pre-stored in the plant optical parameter library, the penetration depth of the reflected image in the plant medium is calculated respectively, and the penetration depth information corresponding to multiple different bands is obtained.

[0067] Furthermore, the process of obtaining the penetration depth information includes:

[0068] S201. Obtain the first attenuation coefficient in the visible red band, the second attenuation coefficient in the near-infrared band, and the third attenuation coefficient in the short-wave infrared band from the plant optical parameter library that match the plant medium.

[0069] S202. Based on the exponential decay law of the radiative transfer model, the first penetration depth value corresponding to the red band reflection image is calculated using the first attenuation coefficient, the second penetration depth value corresponding to the near-infrared band reflection image is calculated using the second attenuation coefficient, and the third penetration depth value corresponding to the short-wave infrared band reflection image is calculated using the third attenuation coefficient.

[0070] The first penetration depth value is less than the second penetration depth value, and the second penetration depth value is less than the third penetration depth value.

[0071] In practice, a radiative transfer model is used to calculate the penetration depth of different spectral bands in the woody medium of plants. The radiative transfer model describes the propagation of light in a radiative medium. For wood, a strongly scattering medium, the attenuation of light intensity follows an exponential decay law, expressed as:

[0072] .

[0073] in, For the incident light intensity, For penetration depth The light intensity at that location, This is the extinction coefficient (i.e., optical attenuation coefficient). Penetration depth is typically defined as the point at which the light intensity decreases to the surface light intensity. The depth corresponding to the time, i.e. .

[0074] In the specific calculations, the optical attenuation coefficient matching the plant species under test is first obtained from the plant optical parameter library. The plant optical parameter library is a pre-built database that stores the optical attenuation coefficients of different garden plant species (such as ginkgo, locust, and pine) in multiple wavelength bands. For the three wavelength bands in this embodiment, the following are obtained respectively:

[0075] First attenuation coefficient in the visible red band (660nm) .

[0076] Second attenuation coefficient in the near-infrared band (850nm) .

[0077] The third attenuation coefficient in the shortwave infrared band (1450nm) .

[0078] Based on the exponential decay law of the radiative transfer model, the penetration depth of each band is calculated:

[0079] .

[0080] .

[0081] .

[0082] Because wood exhibits different absorption and scattering characteristics for different wavelengths of light, it typically has a higher attenuation coefficient and shallower penetration depth in the red band; a moderate attenuation coefficient and moderate penetration depth in the near-infrared band; and a lower attenuation coefficient and greater penetration depth in the short-wave infrared band. Therefore, there are... These three penetration depth values ​​will be used for subsequent multispectral feature fusion as a basis for determining the depth location of abnormal texture regions.

[0083] S3. The multi-scale Retinex algorithm is used to enhance the reflection image, and the specular reflection component is removed from the enhanced reflection image using polarized light image to obtain a reflection image II with non-skin reflection interference removed.

[0084] Furthermore, the process of enhancing the reflected image includes:

[0085] S301. Decompose reflection images under multiple different spectral bands into illumination component images and reflection component images of multiple different scales;

[0086] S302. Perform contrast stretching on the reflection component image at each scale, and then perform weighted fusion on the processed reflection component images at multiple scales to obtain the enhanced reflection image for the corresponding band.

[0087] The scale parameters of the multi-scale Retinex algorithm are adaptively adjusted based on the penetration depth information of the corresponding band, and the penetration depth is positively correlated with the scale parameters used.

[0088] The process of acquiring the reflected image II includes:

[0089] S311. Perform polarization degree analysis on the polarized light image to generate a polarization degree distribution map characterizing the specular reflection intensity of the plant epidermis;

[0090] S312. Based on the polarization degree distribution map, subtract the specular reflection component from the enhanced reflection images of each band to obtain the reflection image II of the corresponding band.

[0091] In practice, this step includes two sub-processes: image enhancement and specular culling, which aim to improve image quality and highlight the detailed features inside the wood.

[0092] First, the multi-scale Retinex algorithm is used to enhance the reflection images of the three bands. Retinex theory, based on the color constancy of the human visual system, decomposes the image into illumination and reflection components. For the input image, the basic form of the Retinex algorithm is:

[0093] .

[0094] in, For the reflection component image, The function is a Gaussian wrapping function, and * denotes the convolution operation. The input image is shown. Multi-scale Retinex (MSR) is a weighted fusion of Retinex results from multiple scales, and its expression is:

[0095] .

[0096] in, For the number of scales, For the first The weights of each scale satisfy... Gaussian wrap function Standard deviation This determines the size of the scale.

[0097] In this invention, the scale parameters of the multi-scale Retinex algorithm are adaptively adjusted based on the penetration depth information obtained in step S2. Specifically, for bands with larger penetration depths (such as shortwave infrared), larger scale parameters are used to enhance the texture information of deep structures; for bands with smaller penetration depths (such as the red band), smaller scale parameters are used to preserve shallow details. The adaptive adjustment formula for the scale parameters is:

[0098] .

[0099] in, As the benchmark, For adjustment coefficients, This represents the penetration depth for the corresponding spectral band. In this way, the enhancement process can be optimized for the characteristics of different spectral bands, resulting in a more refined output image. , , It offers better detail.

[0100] After image enhancement, specular reflection components are removed using polarized light images. Specular reflection is formed by the direct reflection of light onto a medium surface, and its intensity can be estimated through polarization analysis. (Polarization degree image) This reflects the degree of polarization of each pixel; the polarization degree is usually higher in the specular reflection region of the skin. For the enhanced image of each band, the specular reflection component... It can be estimated as follows:

[0101] .

[0102] in, This is a scaling factor, which can be set empirically or obtained through calibration. The reflected image II after removing non-dermal reflection interference (denoted as...) )for:

[0103] .

[0104] After this processing, corrected images in three bands can be obtained: , , These images more accurately reflect the scattering information resulting from the interaction of light with the internal structure of wood.

[0105] S4. Extract the abnormal texture region of the reflection image II, and construct a multispectral feature fusion matrix based on the abnormal texture region and the penetration depth information.

[0106] Furthermore, the extraction of the abnormal texture region includes:

[0107] S401. Perform gray-level co-occurrence matrix calculation on the reflection image II of each band to obtain the texture second-order statistics of the reflection image II of each band. The texture second-order statistics include angular second moment, contrast, correlation and entropy.

[0108] S402. Compare the second-order statistics of the texture with the reference threshold of healthy plant texture in the corresponding band, and mark the pixel area that exceeds the reference threshold as a candidate abnormal texture area.

[0109] S403. Perform connected component analysis on the candidate abnormal texture regions, and select connected components with an area greater than a preset area threshold as the abnormal texture regions.

[0110] The construction process of the multispectral feature fusion matrix includes:

[0111] S411. Mark the abnormal texture region of the red band reflection image II as a first spatial location set, and associate it with the first penetration depth value;

[0112] S412. Mark the abnormal texture region of the near-infrared band reflection image II as a second spatial location set, and associate it with the second penetration depth value;

[0113] S413. Mark the abnormal texture region of the shortwave infrared band reflection image II as a third spatial location set, and associate it with the third penetration depth value;

[0114] S414. Overlay the first spatial location set, the second spatial location set, and the third spatial location set in a unified spatial coordinate system to generate the multispectral feature fusion matrix.

[0115] In practice, texture features are described using a gray-level co-occurrence matrix (GLCM). For each band of the corrected image... First, quantize its grayscale level to Level (usually 256 levels), then calculate its gray-level co-occurrence matrix. ,in This represents the grayscale value of two pixels. Indicates pixel spacing. Indicates direction (usually four directions: 0°, 45°, 90°, and 135°, then averaged).

[0116] Based on the gray-level co-occurrence matrix, the following four second-order texture statistics are calculated:

[0117] Second moment of angle (energy): .

[0118] Contrast: .

[0119] Correlation: .

[0120] entropy: .

[0121] in This is the mean of the rows and columns; The standard deviation is denoted as .

[0122] For each pixel position The texture statistics mentioned above are calculated using a neighborhood window centered on the texture (e.g., 15×15 pixels) to obtain the texture feature map. , , , Each texture feature map is compared with a pre-stored healthy plant texture benchmark threshold. The healthy plant texture benchmark threshold is obtained by collecting images of the same parts of a large number of healthy trees and statistically analyzing the texture features.

[0123] For the Each texture feature, whose baseline threshold range is denoted as . If any texture feature exceeds the baseline threshold range, the pixel is marked as a candidate anomalous texture pixel, and a binary mask is generated. :

[0124] .

[0125] Perform connected component analysis on candidate abnormal texture regions and calculate the area of ​​each connected component. Set area threshold Filter out areas larger than Connected components are used as the final abnormal texture regions to generate abnormal texture region masks. .

[0126] After obtaining the abnormal texture region, a multispectral feature fusion matrix is ​​constructed by combining the penetration depth information from step S2. Specifically, three spatial location sets are defined:

[0127] .

[0128] .

[0129] .

[0130] Each location set is associated with a corresponding penetration depth value: Related , Related , Related These three location sets are superimposed in a unified spatial coordinate system to generate a multispectral feature fusion matrix. Its element value is a triple. This indicates whether the pixel is in an abnormal texture region in any band:

[0131] .

[0132] S5. Based on the multispectral feature fusion matrix, by comparing and analyzing the areas where the shallow images are normal and the deep images are abnormal, as well as the areas where all images are abnormal, internal anomaly classification and discrimination rules are executed to determine the chromatographic distribution of the plant's internal health status.

[0133] Furthermore, the process of executing internal anomaly classification and discrimination rules includes:

[0134] S501. When no abnormal texture area is detected in the red band reflection image II and the near-infrared band reflection image II, but an abnormal texture area exists in the short-wave infrared band reflection image II, it is determined to be a deep-layer cavitation area.

[0135] S502. When no abnormal texture area is detected in the red band reflection image II, but abnormal texture areas are detected in the near-infrared band reflection image II and the short-wave infrared band reflection image II, it is determined to be a decayed area of ​​the middle to deep layer of the material.

[0136] S503. When abnormal texture areas are detected in the red band reflection image II, near-infrared band reflection image II and short-wave infrared band reflection image II, they are determined to be severe cavitation areas or epidermal lesion areas that run through the entire mass.

[0137] In practice, this step utilizes a multispectral feature fusion matrix. Based on the penetration depth information, each pixel location is classified and judged to determine its corresponding internal health status. The classification and judgment rules are designed based on the differences in penetration depth across different bands, as follows:

[0138] Rule 1: Determination of deep xylem hollow decay areas.

[0139] When the condition is met , , This indicates that the pixel has no abnormal texture in the red and near-infrared band images, but only exhibits abnormal texture in the short-wave infrared band image. Based on the penetration depth relationship... This means that the shallow layer (from the bark to the outer xylem) has normal texture, while the deep layer (inner xylem) has abnormal texture. Therefore, this area is determined to be a deep xylem cavity decay area, corresponding to a health status label. .

[0140] Rule 2: Determination of decayed areas in the middle to deep xylem.

[0141] When the condition is met , , This indicates that the pixel shows no abnormal texture in the red band image, but exhibits abnormal texture in the near-infrared and short-wave infrared band images. This means the shallow (epidermal) texture is normal, while the texture in the middle to deep layers (near-infrared penetration depth to short-wave infrared penetration depth) is abnormal. Therefore, this area is determined to be a middle to deep xylem decay area, corresponding to the health status label. .

[0142] Rule 3: Determination based on severely hollow areas or epidermal lesions that extend throughout the entire xylem.

[0143] When the condition is met , , When the value is zero, it indicates that the pixel exhibits abnormal texture in all three spectral bands of the image. This means that texture abnormalities exist from the epidermis to the deep xylem. This situation may correspond to two scenarios: one is a severely decayed area that runs through the entire xylem, and the other is a lesion in the epidermis itself (such as bark damage, ulcers, etc.). To further distinguish between these two situations, the specific manifestations of texture features can be used for auxiliary judgment, but as a basic classification, a uniform "healthy state" label is assigned. .

[0144] For other combinations (such as) This indicates an abnormality only in the epidermis; (etc.), which can be further classified according to actual needs.

[0145] By applying the above rules, a health status label is assigned to each pixel location in the fusion matrix, thus forming a tomographic distribution map. This distribution map contains two-dimensional spatial location information and corresponding health status category information.

[0146] S6. Based on the tomographic distribution, generate a tomographic image of the plant's internal health status, and distinguish healthy plant areas, areas with initial rot, and severely rotten areas with different colors.

[0147] In practice, this step will involve the chromatographic distribution map. The image is visualized as a color tomographic image. Based on the health status labels obtained in step S5, a color mapping relationship is established:

[0148] Healthy plant areas (areas without abnormal textures) ): Mapped to green

[0149] The initial decay zone (including the deep hollow decay of Rule 1 and the middle decay of Rule 2) ): is mapped to yellow.

[0150] Severely cavitated areas (penetrating cavitation or epidermal lesions as defined in Rule 3), ): is mapped to red.

[0151] Color mapping can be represented as:

[0152] .

[0153] in This is a color lookup function. It generates a three-channel color image. This is a tomographic image showing the internal health status of the tree trunk.

[0154] This image visually illustrates the distribution of rotten or decayed areas inside a tree trunk: green areas represent healthy wood, yellow areas indicate varying degrees of decay or cavities, and red areas represent severe rot or bark disease. By observing this image, landscape maintenance personnel can quickly pinpoint health problems inside the tree trunk, providing a basis for decision-making regarding subsequent precise maintenance.

[0155] This invention also provides an image analysis-based system for monitoring the health status of garden plants, used to implement the above-mentioned method. The system includes: an image acquisition module, a penetration depth calculation module, an image enhancement and correction module, a feature extraction and fusion module, a tomographic distribution discrimination module, and an imaging output module.

[0156] The image acquisition module includes a multispectral camera and a polarization camera. The multispectral camera is equipped with three independent imaging channels corresponding to wavelengths of 660nm, 850nm, and 1450nm, respectively, while the polarization camera is equipped with an orthogonal polarization analyzer. The output of the image acquisition module is connected to the input of the penetration depth calculation module and the image enhancement and correction module, respectively, for transmitting the acquired multi-band reflection images and polarized light images.

[0157] The penetration depth calculation module is connected to a plant optical parameter library, which stores the optical attenuation coefficients of different garden plant species in the visible red band, near-infrared band, and short-wave infrared band, as well as the texture reference thresholds for healthy plants. The penetration depth calculation module receives the reflected image output by the image acquisition module, calculates the penetration depth of each band based on the radiative transfer model and the optical attenuation coefficient, and outputs the penetration depth information to the feature extraction and fusion module.

[0158] The image enhancement and correction module receives the reflected image and polarized light image output by the image acquisition module, performs multi-scale Retinex enhancement processing and specular reflection component removal processing to obtain the reflected image II, and outputs the result to the feature extraction and fusion module.

[0159] The feature extraction and fusion module receives the penetration depth information output by the penetration depth calculation module and the reflection image II output by the image enhancement and correction module, respectively. It performs abnormal texture region extraction and multispectral feature fusion matrix construction, and outputs the fusion matrix to the tomographic distribution discrimination module.

[0160] The tomographic distribution discrimination module receives the multispectral feature fusion matrix output by the feature extraction and fusion module, executes the internal anomaly classification discrimination rules, determines the tomographic distribution of the plant's internal health status, and outputs the tomographic distribution information to the imaging output module.

[0161] The imaging output module receives the tomographic distribution information output by the tomographic distribution discrimination module, generates a color tomographic image, and outputs it through a display device.

[0162] The above modules can be integrated into an embedded device or distributed between a cloud server and a field acquisition terminal. Those skilled in the art can deploy them flexibly according to the actual application scenario.

[0163] Based on the preferred embodiments of the present invention described above, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for monitoring the health status of garden plants based on image analysis, characterized in that, Includes the following steps: S1. Acquire reflection images of the same area of ​​the same plant under multiple preset different wavelengths, and simultaneously acquire polarized light images of the same area of ​​the same plant. S2. Based on the radiative transfer model and combined with the optical attenuation coefficients of plants in different bands pre-stored in the plant optical parameter library, the penetration depth of the reflected image in the plant medium is calculated respectively, and the penetration depth information corresponding to multiple different bands is obtained. S3. The multi-scale Retinex algorithm is used to enhance the reflection image, and the specular reflection component is removed from the enhanced reflection image using polarized light image to obtain a reflection image II with non-skin reflection interference removed. S4. Extract the abnormal texture regions of the reflection image II, and construct a multispectral feature fusion matrix based on the abnormal texture regions and the penetration depth information; S5. Based on the multispectral feature fusion matrix, by comparing and analyzing the areas where the shallow images are normal and the deep images are abnormal, as well as the areas where all images are abnormal, internal anomaly classification and discrimination rules are executed to determine the chromatographic distribution of the plant's internal health status. S6. Based on the tomographic distribution, generate a tomographic image of the plant's internal health status, and distinguish healthy plant areas, areas with initial rot, and severely rotten areas with different colors.

2. The method for monitoring the health status of garden plants based on image analysis according to claim 1, characterized in that: The reflection images include: red band reflection images, near-infrared band reflection images, and short-wave infrared band reflection images of the same area of ​​the same plant, which are simultaneously captured by a multispectral camera. Polarized light images of the same region of the same plant in orthogonal polarization directions are acquired synchronously by a polarization camera and a multispectral camera to characterize the non-Near reflectance components of the plant epidermis.

3. The method for monitoring the health status of garden plants based on image analysis according to claim 1, characterized in that: The process of obtaining the penetration depth information includes: S201. Obtain the first attenuation coefficient in the visible red band, the second attenuation coefficient in the near-infrared band, and the third attenuation coefficient in the short-wave infrared band from the plant optical parameter library that match the plant medium. S202. Based on the exponential decay law of the radiative transfer model, the first penetration depth value corresponding to the red band reflection image is calculated using the first attenuation coefficient, the second penetration depth value corresponding to the near-infrared band reflection image is calculated using the second attenuation coefficient, and the third penetration depth value corresponding to the short-wave infrared band reflection image is calculated using the third attenuation coefficient. The first penetration depth value is less than the second penetration depth value, and the second penetration depth value is less than the third penetration depth value.

4. The method for monitoring the health status of garden plants based on image analysis according to claim 1, characterized in that: The process of enhancing the reflected image includes: S301. Decompose reflection images under multiple different spectral bands into illumination component images and reflection component images of multiple different scales; S302. Perform contrast stretching on the reflection component image at each scale, and then perform weighted fusion on the processed reflection component images at multiple scales to obtain the enhanced reflection image for the corresponding band. The scale parameters of the multi-scale Retinex algorithm are adaptively adjusted based on the penetration depth information of the corresponding band, and the penetration depth is positively correlated with the scale parameters used.

5. The method for monitoring the health status of garden plants based on image analysis according to claim 1, characterized in that: The process of acquiring the reflected image II includes: S311. Perform polarization degree analysis on the polarized light image to generate a polarization degree distribution map characterizing the specular reflection intensity of the plant epidermis; S312. Based on the polarization degree distribution map, subtract the specular reflection component from the enhanced reflection images of each band to obtain the reflection image II of the corresponding band.

6. The method for monitoring the health status of garden plants based on image analysis according to claim 1, characterized in that: The extraction of the abnormal texture region includes: S401. Perform gray-level co-occurrence matrix calculation on the reflection image II of each band to obtain the texture second-order statistics of the reflection image II of each band. The texture second-order statistics include angular second moment, contrast, correlation and entropy. S402. Compare the second-order statistics of the texture with the reference threshold of healthy plant texture in the corresponding band, and mark the pixel area that exceeds the reference threshold as a candidate abnormal texture area. S403. Perform connected component analysis on the candidate abnormal texture regions, and select connected components with an area greater than a preset area threshold as the abnormal texture regions.

7. The method for monitoring the health status of garden plants based on image analysis according to claim 3, characterized in that: The construction process of the multispectral feature fusion matrix includes: S411. Mark the abnormal texture region of the red band reflection image II as a first spatial location set, and associate it with the first penetration depth value; S412. Mark the abnormal texture region of the near-infrared band reflection image II as a second spatial location set, and associate it with the second penetration depth value; S413. Mark the abnormal texture region of the shortwave infrared band reflection image II as a third spatial location set, and associate it with the third penetration depth value; S414. Overlay the first spatial location set, the second spatial location set, and the third spatial location set in a unified spatial coordinate system to generate the multispectral feature fusion matrix.

8. The method for monitoring the health status of garden plants based on image analysis according to claim 1, characterized in that: The process of executing internal anomaly classification rules includes: S501. When no abnormal texture area is detected in the red band reflection image II and the near-infrared band reflection image II, but an abnormal texture area exists in the short-wave infrared band reflection image II, it is determined to be a deep-layer cavitation area. S502. When no abnormal texture area is detected in the red band reflection image II, but abnormal texture areas are detected in the near-infrared band reflection image II and the short-wave infrared band reflection image II, it is determined to be a decayed area of ​​the middle to deep layer of the material. S503. When abnormal texture areas are detected in the red band reflection image II, near-infrared band reflection image II and short-wave infrared band reflection image II, they are determined to be severe cavitation areas or epidermal lesion areas that run through the entire mass.

9. A garden plant health status monitoring system based on image analysis, characterized in that, The garden plant health status monitoring system is used to implement the garden plant health status monitoring method according to any one of claims 1 to 8.

10. The image analysis-based garden plant health status monitoring system according to claim 9, characterized in that: The garden plant health status monitoring system includes: an image acquisition module, a penetration depth calculation module, an image enhancement and correction module, a feature extraction and fusion module, a tomographic distribution discrimination module, and an imaging output module; The image acquisition module is used to execute S1, and its output is connected to the input of the penetration depth calculation module and the image enhancement and correction module, respectively. The penetration depth calculation module is used to execute S2. Its input end receives the reflection image output by the image acquisition module, and its output end is connected to the input end of the feature extraction and fusion module. The image enhancement and correction module is used to execute S3. Its input end receives the reflected image and polarized light image output by the image acquisition module, and its output end is connected to the input end of the feature extraction and fusion module. The feature extraction and fusion module is used to execute S4. Its input end receives the penetration depth information output by the penetration depth calculation module and the reflection image II output by the image enhancement and correction module, respectively. Its output end is connected to the input end of the tomographic distribution discrimination module. The tomographic distribution discrimination module is used to execute S5. Its input end receives the multispectral feature fusion matrix output by the feature extraction and fusion module, and its output end is connected to the input end of the imaging output module. The imaging output module is used to execute S6, and its input end receives the tomographic distribution information output by the tomographic distribution discrimination module and generates a tomographic image.

Images