Method for predicting nitrogen oxide infection change of plant based on AI image

Abstract

The invention discloses a method for predicting nitrogen oxide infection change of a plant based on an AI image in the technical field of plant research, and the method comprises the steps: building a dynamic mapping relation of morphological indexes-infection concentration and time; the image morphology of the plant under time lapse and / or other concentrations is directly predicted by inputting the current image of the plant, the plant tissue does not need to be damaged, the whole-stage dynamic monitoring of the same plant from the health state to the pollution stress can be realized through regular image acquisition, the interference of traditional chemical sampling on the plant growth is avoided, and the monitoring accuracy is improved. The method is particularly suitable for rare plant or long-term ecological research, the abstract RGB value and the nitrogen oxide concentration are converted into visual results such as a color gradient diagram and a dynamic feature point evolution curve, non-professionals can rapidly judge the pollution level through color depth and patch distribution, and the environment early warning response time is shortened.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of plant research, and particularly relates to a method for predicting changes of plants infected by nitrogen oxides based on AI images. BACKGROUND

[0002] As one of the main atmospheric pollutants, nitrogen oxides have various effects on plant leaves. Nitrogen oxides can enter the plant body through the stomata, trigger oxidative stress, cause degradation of leaf pigments, damage to cell structure, and disorder of physiological functions. Studying the changes of plants infected by nitrogen oxides can help establish biological monitoring and early warning, and is also conducive to cultivating plants with strong resistance to nitrogen oxides.

[0003] The existing research methods for changes of plants infected by nitrogen oxides are relatively simple. Generally, destructive measurements are made on plant leaves by hand, and changes of plants before and after infection are compared by analyzing static images by hand. This method is low in efficiency and lacks visualization, and it is difficult to intuitively, accurately and timely predict the effects of nitrogen oxides on plants. SUMMARY

[0004] The present application provides a method for predicting changes of plants infected by nitrogen oxides based on AI images, which solves the problem of low efficiency and lack of visualization of the existing research prediction methods.

[0005] The present application achieves the above-mentioned purpose through the following technical solutions: A method for predicting changes of plants infected by nitrogen oxides based on AI images, comprising: processing a reference image of a target plant object based on a pre-trained image recognition model to extract a to-be-measured image including plant color features and morphological parameters, wherein the color features include leaf RGB range, dominant color, and color uniformity; and the reference image includes different regions of plant leaves and a plurality of feature point positions thereof pre-labeled; inputting the to-be-measured image into a dynamic correlation model for describing the relationship between plant features and nitrogen oxide concentration to calculate concentration prediction values of the morphological parameters and the color features of the feature point positions infected by nitrogen oxides, and generate a color gradient visualization prediction image; comparing the color features and morphological parameters before and after infection based on the prediction image to determine the degree of infection and dynamic change trend of the target plant object by nitrogen oxides; The dynamic correlation model is constructed in the following manner: under different nitrogen oxide concentrations and different infection time environments, the to-be-measured images of sample plants are collected, the color features of the morphological parameters and the preset feature point positions in each image are extracted as target features; the nitrogen oxide concentration and the infection time are input, and the target features are output, and the dynamic correlation model is trained by a machine learning algorithm.

[0006] As a preferred scheme of the present application, the pre-trained image recognition model processes the reference image of the target plant object to extract the to-be-tested image including the plant color feature and the morphological parameter, comprising: An image of the target plant object under unknown nitrogen oxide concentration and infection time is obtained and used as the reference image; The corresponding feature point selection mode is matched based on the plant color type; A plurality of feature points are preset on the reference image based on the matched feature point selection mode, and the morphological parameter and the RGB color value of each feature point are extracted; The color feature of the feature point is determined according to the RGB color value of each feature point, and the to-be-tested image is obtained based on the color feature and the morphological parameter.

[0007] As a preferred scheme of the present application, the corresponding feature point selection mode matched based on the plant color type comprises: For a solid color plant, the feature points at least include the center of the leaf or the intersection point of the geometric diagonal; For a two-color or mottled plant, the feature points include the center of different color regions and the midpoint of the color transition zone.

[0008] As a preferred scheme of the present application, the RGB color value of each feature point is extracted, comprising: The boundary of the plant tissue unit in the reference image is extracted, and a plurality of feature points uniformly distributed on the plant tissue unit are generated; The feature points are geometrically positioned to generate a coordinate file and obtain the RGB color value of the feature points.

[0009] As a preferred scheme of the present application, the pre-trained image recognition model construction mode comprises: A sample plant image set containing different nitrogen oxide concentration and infection time labels is collected; The images in the image set are labeled, and the labeling content includes morphological parameters, color features and environment labels; the morphological parameters include leaf area, perimeter, edge jaggedness, and lesion area ratio; the environment labels include nitrogen oxide concentration, infection time, shooting time, and plant growth stage; The labeled image set is trained by using a deep learning algorithm to obtain the image recognition model; Among them, the RGB color value and the morphological parameter of the sample plant with zero nitrogen oxide concentration in the environment label are counted to determine the initial morphological parameter range and the initial RGB health range.

[0010] As a preferred scheme of the present application, the comparison of the color features and the morphological parameters before and after the infection determines the degree of infection and the dynamic change trend of the target plant object by nitrogen oxides, comprising: Comparing the real-time RGB values and morphological parameters of the feature points of the target plant object with the initial morphological parameter range and the initial RGB health range; According to the comparison result, the degree of infection of the target plant object by nitrogen oxides is matched, the concentration of nitrogen oxides and the infection time are determined, and the dynamic change trend is determined based on the concentration of nitrogen oxides and the infection time.

[0011] As a preferred scheme of the present application, the method further comprises: Comparing the real-time extracted RGB color values and morphological parameters with the initial RGB health range and the initial morphological parameter range; When the real-time RGB color values or morphological parameters exceed the health threshold set based on the initial RGB health range and the initial morphological parameter range, an abnormal area is marked on the prediction image and a warning is triggered.

[0012] As a preferred scheme of the present application, the visual prediction image comprises: A color gradient image showing the degree of influence of the concentration of nitrogen oxides on different regions of the leaf in the form of color gradient; A dynamic feature point evolution map superimposed with the morphological parameter change trend by comparing the RGB value change curve of the feature points under different concentrations and infection times; When the warning is triggered, an abnormal warning map is marked with a set color and shape in the abnormal area.

[0013] As a preferred scheme of the present application, when constructing the dynamic correlation model, different modeling strategies are adopted for different color types of plants: If the target plant is a solid color plant, a single model covering the entire area is constructed; If the target plant is a two-color or mottled plant, independent sub-models are constructed for different color regions.

[0014] As a preferred scheme of the present application, for two-color or mottled plants, when generating the color gradient visual prediction image, the model output is used to show the influence of the pollutant on the color division, including the visual effects of transition band blurring, color division disorder or forming irregular color spots.

[0015] The present application has the following advantages: (1) The prediction method can realize dynamic monitoring of the same plant from a healthy state to pollution stress through periodic image acquisition, avoid the interference of traditional chemical sampling on plant growth, and is especially suitable for rare plants or long-term ecological research. The method visualizes the dynamic feature point evolution curve, and non-professionals can quickly judge the pollution level through the color depth and patch distribution, thereby shortening the environmental early warning response time. (2) Through standardized image acquisition, deep learning feature extraction, dynamic model construction, and visual image generation, the method realizes accurate analysis of the RGB range and morphological parameters of plants, and solves the problems of destructive detection, inefficient manual analysis, and insufficient visualization in the prior art. (3) Independent analysis processes are designed for single-color and double-color plants, and the method is compatible with multiple species and covers more than 80% of common plant types. The method does not require chemical reagents or radioactive detection throughout the process, and only uses optical imaging and algorithm analysis to realize pollution prediction, without any interference to the surrounding environment such as soil and water, which meets the current needs of ecological protection and sustainable development. (4) By pre-setting feature points on uninfected plant leaves, the analysis area of different samples and different periods is ensured to be completely consistent, avoiding data deviation caused by different point selection, and improving the comparability of cross-experimental data. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The figure is a schematic diagram of the method of the present application; Figure 2 The figure is an example of a single-color and double-color plant annotation point generated by the present application; Figure 3 The figure is a single-color plant image under different nitrogen oxide concentrations in Example 1 of the present application; Figure 4 The figure is a double-color plant image under different nitrogen oxide concentrations in Example 2 of the present application; DETAILED DESCRIPTION

[0017] It is necessary to point out here that the following detailed description is only used to further illustrate the present application, and cannot be understood as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application according to the above application content.

[0018] Example 1 As shown in the figure, the method of the present application for predicting the change of a plant infected by nitrogen oxides based on AI image selects a healthy single-color safflower plant with uniform color and no natural spots or discoloration area. The main steps are as follows: Figures 1-4 S1: Construct an image recognition model; S2: Collect images of the plant under different nitrogen oxide concentrations; S11: Place the plants in a closed climate chamber with nitrogen oxide concentrations of 5 ppm, 10 ppm, 15 ppm, and 20 ppm, respectively, and the unified infection time is 7 days; S12: Collect high-definition images of plant tissue units regularly, including flowers or leaves. Standardize image collection: Equipment parameters: use a high-resolution camera with fixed parameters, resolution ≥3000x2000 pixels, fixed on a tripod, shooting distance 1.5 meters, lens perpendicular to the leaf plane; Light source: use indoor LED flat panel lights with color temperature 5500K and light intensity 1000Lux, ensuring no environmental light interference; Collection range: a single image covers the entire leaf, with a 2cm margin for later segmentation, ensuring that images at different times and different concentrations have high consistency, low data error rate, and provide a reliable benchmark for subsequent model construction. The 0ppm control group benchmark image is obtained based on the following method: collect at 10am every day for 30 days, and obtain a set of sample plant images; S13: Label the images in the image set, including color features, morphological parameters, and environmental labels. Color features include leaf RGB range, dominant color, and color uniformity. Morphological parameters include leaf area, perimeter, edge jaggedness, and lesion area ratio. Environmental labels include nitrogen oxide concentration, infection time, shooting time, and plant growth stage. RGB is the abbreviation of Red, Green, and Blue, and the core principle is additive color mixing, which produces various colors by adding different intensities of red, green, and blue primary color light. Please refer to Figure 2 The method for labeling the RGB range is as follows: extract the boundary of the plant tissue unit in the benchmark image, and generate a number of evenly distributed feature points on the plant tissue unit. Geometrically position the feature points to generate a coordinate file and obtain the RGB color values of the feature points. Among them, the RGB color values and morphological parameters of the sample plants with zero nitrogen oxide concentration in the environmental label are counted to determine the initial RGB healthy range and the initial morphological parameter range. Use a deep learning algorithm to train the labeled image set to obtain the image recognition model.

[0019] S2: Extract the plant color features and morphological parameters of the image to be tested; S21: Obtain the image of the target plant object under unknown nitrogen oxide concentration and infection time, and use it as the benchmark image; S22: Select the corresponding feature point selection method based on the color type of the plant. In this embodiment, the plant is a solid color plant, and the selection method is as follows: the feature points at least include the center of the leaf or the intersection of the geometric diagonal, the boundary of the plant tissue unit in the reference image is extracted, a plurality of feature points uniformly distributed on the plant tissue unit are generated, the feature points are geometrically positioned, a coordinate file is generated, and the RGB color value of the feature points is obtained; S23: Based on the matched feature point selection method, a plurality of feature points are preset on the reference image, and the shape parameters and the RGB color values of the feature points are extracted; S24: According to the RGB color value of each feature point, the color feature of the feature point is determined, and the to-be-measured image is obtained based on the color feature and the shape parameter.

[0020] S3: Generate a color gradient visualization prediction image; S31: Based on different nitrogen oxide concentrations and different infection times, the to-be-measured images of sample plants are collected, the shape parameters and the color features of the preset feature point positions in each image are extracted as target features, the nitrogen oxide concentration and the infection time are taken as input, the target features are taken as output, and the dynamic correlation model is trained by a machine learning algorithm; S32: The to-be-measured image is input into the dynamic correlation model for describing the relationship between the plant features and the nitrogen oxide concentration, and the color features and the shape parameters of the feature point positions are calculated to obtain the concentration prediction value of the nitrogen oxide infection; S33: The real-time extracted RGB color value and shape parameter are compared with the initial RGB health range and initial shape parameter range; when the real-time RGB color value exceeds the health threshold set based on the initial RGB health range and initial shape parameter range, the abnormal area is marked on the prediction image and a warning is triggered; S34: The color gradient image showing the influence degree of different areas of the leaf under the nitrogen oxide concentration is displayed in the form of color gradient; the RGB value change curve of the feature points under different concentrations and infection times is compared, and the dynamic feature point evolution graph of the shape parameter change trend is superimposed; when the warning is triggered, the abnormal area is marked with a set color and shape in the abnormal warning graph.

[0021] S4: Determine the nitrogen oxide infection degree and dynamic change trend of the target plant object; S41: Compare the real-time RGB value and shape parameter of the corresponding feature point of the target plant object with the initial RGB health range and initial shape parameter range; S42: Match the nitrogen oxide infection degree of the target plant object according to the comparison result, determine the nitrogen oxide concentration and infection time, and determine the dynamic change trend based on the nitrogen oxide concentration and infection time.

[0022] In addition, the model of the embodiment is automatically updated by importing new data in real time, can dynamically adapt to the response differences of plants in different growth stages (such as seedling stage and flowering stage) to pollution, and can also set personalized RGB value warning thresholds according to plant species, such as triggering an alarm when the R value of a pure color leaf plant is lower than 200, and the system automatically marks the abnormal area and generates a warning report to gain a golden time window for pollution emergency disposal.

[0023] The prediction method can realize dynamic monitoring of the same plant from a healthy state to pollution stress through periodic image acquisition without damaging the plant tissue, avoid the interference of traditional chemical sampling on plant growth, and is especially suitable for rare plants or long-term ecological research. The method converts the abstract RGB value and nitrogen oxide concentration into visual results such as color gradient map and dynamic feature point evolution curve, so that non-professionals can quickly judge the pollution level through the color depth and patch distribution, and shorten the environmental warning response time.

[0024] Not only the change of the RGB value range of the plant color is analyzed, but also the main color tone and color uniformity are tracked synchronously to realize the correlation analysis of “color anomaly-structure damage” and comprehensively reveal the combined stress effect of nitrogen oxides on plants.

[0025] In the embodiment, the infection results of pure red flower plants are as follows: Nitrogen oxide concentration / ppm R value range G value range B value range Dominant color tone, color uniformity 0 （211,252） （7，34） （43，130） Uniform color 5 （204,252） （12，223） （61，247） Slightly dispersed color, slight damage 10 （145，248） （31，213） （74，248） Uniformity is destroyed, local discoloration 15 （186，243） （50，196） （125，249） Stress is intensified, and brown phenomenon appears 20 （125，249） （58，193） （89，216） Overall ash, pigment metabolism disorder Under the concentration of 0 ppm nitrogen oxides, the RGB range is R ∈ (211, 252), G ∈ (7, 34), and B ∈ (43, 130), the color characteristics are high saturation red and uniformity, the plant state is that the plant color is uniform and consistent under the condition of no pollution, which is a healthy pure color tissue (such as petals, fruits or specific leaves) not affected by stress, the morphological parameters are that the average leaf area is 148.2 cm 2 , the average leaf perimeter is 49.6 cm, the average edge serration degree is 0.12, there is no obvious serration, and the average proportion of lesion area is 0.02%; Please refer to Figure 3 Under the concentration of 5 ppm nitrogen oxides, the plant is infected for 7 days, and after statistical analysis, the RGB range is R ∈ (204, 252), G ∈ (12, 223), and B ∈ (61, 247); the color change is that the red color is still dominant, but part of the measured values show that the green and blue colors have slightly risen, and the color tone has shifted to light red or magenta; the plant state is that the color begins to appear slightly dispersed under the influence of initial pollution, but no obvious damage has been formed; the morphological parameters are that the average leaf area is 143.5 cm 2Compared with 0 ppm, the area decreased slightly, indicating that the plant was slightly stressed, the average leaf perimeter was 47.8 cm, which was slightly reduced compared with 0 ppm, indicating that the leaf edge appeared slight damage, the average edge serration was 0.21, indicating that the leaf edge appeared slight serration, and the average lesion area ratio was 0.06%, indicating that the plant was slightly stressed and a small amount of lesions appeared; Under the concentration of 10 ppm of nitrogen oxides, the RGB range was R∈(145, 248), G∈(31, 213), and B∈(74, 248) after 7 days of infection; color change: red fluctuation intensified, some samples approached white or light purple due to high blue-green value, while other samples tended to deep purple red, showing uneven color mottling, green and blue significantly increased, especially the blue-green tone in high concentration samples was prominent; plant status: under pollution stress, color uniformity was destroyed, local tissue discoloration may occur, indicating that physiological function began to disorder; morphological parameters: the average leaf area was 132.7 cm 2 , the area further decreased, indicating that the plant growth was inhibited to a certain extent, the average leaf perimeter was 44.9 cm, the perimeter further decreased, indicating that the leaf edge damage intensified, the average edge serration was 0.33, indicating that the leaf edge damage intensified, and the serration change was obvious, the average lesion area ratio was 0.14%, indicating that the plant was significantly stressed and the number of lesions increased; Under the concentration of 15 ppm of nitrogen oxides, the RGB range was R∈(186, 243), G∈(50, 196), and B∈(125, 249) after 7 days of infection; color change: red overall decreased, green and blue became the main tone, and the color tended to blue-green, cyan or gray-green. Plant status: pollution stress intensified, the color changed from "red dominant" to "blue-green dominant", and began to appear discoloration or browning signs; morphological parameters: the average leaf area was 118.3 cm 2 , the area significantly decreased, indicating that the plant was under severe stress, the average leaf perimeter was 40.2 cm, the perimeter significantly decreased, indicating that the leaf edge damage was serious, the average edge serration was 0.45, indicating that the leaf edge damage was serious, and the serration change was significant, the average lesion area ratio was 0.26%, indicating that the plant was under severe stress and the number of lesions significantly increased; RGB range is R∈(154, 235), G∈(58, 193), B∈(89, 216); color change: red fluctuation range expands, low concentration samples present dark brown red or gray brown, high concentration samples deviate to light pink or light purple due to G / B rising, overall tone is dim, saturation decreases, green and blue still have higher values, but compared with 15ppm, G / B values of some samples decrease, which may be caused by severe damage to tissues leading to pigment degradation; plant state: under high concentration pollution, plant color is significantly disordered, large area of browning, wilting or cell necrosis occurs, the discreteness of RGB range reflects that the tissues have suffered irreversible damage, and the physiological function is close to collapse; morphological parameters: the average leaf area is 102.4cm 2 , the area is greatly reduced, indicating that plant growth is severely inhibited, some leaves appear wilting phenomenon, the average leaf perimeter is 35.4cm, the perimeter is greatly reduced, indicating that the leaf edge is severely damaged, some leaves appear wilting phenomenon, the average edge serration is 0.58, indicating that the leaf edge is severely damaged, the serration changes significantly, and the average proportion of lesion area is 0.38%, indicating that the plant is under severe stress, the number of lesions increases significantly, and some leaves have large area of lesions.

[0026] Embodiment 2 In this embodiment, a typical bicolor flower plant is selected for infection, and the leaf is divided into a red dominant color region (about 60% of the proportion), a white mottling region (about 40% of the proportion), and a red-white color transition zone in a natural state. The main steps are as follows: The main steps are the same as in embodiment 1, but different from embodiment 1 are as follows: In the selection method of feature points, the feature points of this embodiment at least include the centers located in different color regions and the midpoint in the color transition zone; In step S31 of constructing a dynamic correlation model, this embodiment constructs independent sub-models for different color regions respectively; In step S34 of generating a color gradient visualization prediction image, based on embodiment 1, the output of the dynamic correlation model of this embodiment is also used to show the effect of the pollutant on the color boundary, including the visual effects of transition zone blurring, color boundary disorder or irregular color spot formation.

[0027] For details, please refer to Figure 2 In this embodiment, the feature points are designed as follows: in the red region, multiple points are marked and distributed in the center of the red patch; in the white region, multiple points are marked and distributed in the center of the white mottling; and in the transition zone, multiple points are marked and located at the midpoint of the red-white boundary.

[0028] In this embodiment, the infection results of the bicolor red flower plant are as follows: Nitrogen oxide concentration / ppm R value range G value range B value range Dominant color tone, color uniformity 0 （144，248） （10，247） （6，252） Uniform color 5 （168，226） （72，123） （83，136） Color shift, mixed color 10 （162，253） （45，232） （18，239） Color dispersion, significant pollution sensitivity 15 （176，247） （81，249） （2，252） Color transition disappears, irregular color spots 20 （143，253） （105，224） （29，207） Overall ash, pigment metabolism disorder At 0 ppm nitrogen oxide concentration, the RGB value range: red region, R ∈ (144, 248), G ∈ (10, 26), B ∈ (6, 52); white region, R ∈ (194, 246), G ∈ (192, 247), B ∈ (212, 252); boundary, R ∈ (210, 232), G ∈ (110, 172), B ∈ (144, 173), color characteristics: the red region presents high saturation of red (R value is much higher than G and B), the white region is close to white (R, G, B values are all high and close), and the boundary is a transition color (R value is high, and G and B values are moderate), plant state: no obvious color change caused by nitrogen oxide pollution overall, and the color distribution conforms to the natural characteristics of double-colored leaf plants; morphological parameters: the average leaf area of the red region is 89.5 cm 2 ; the average leaf area of the white region is 58.3 cm 2 , the average leaf perimeter of the red region is 39.8 cm; the average leaf perimeter of the white region is 29.5 cm, the average edge serration of the red region is 0.11; the average edge serration of the white region is 0.10, without obvious serration, the average lesion area ratio of the red region is 0.02%, and the average lesion area ratio of the white region is 0.03%; Please refer to Figure 4 At 5 ppm nitrogen oxide concentration, after 7 days of infection, the RGB range is R ∈ (168, 226), G ∈ (72, 123), and B ∈ (83, 136), and the color characteristics are: in the red region, the R value is still high, but the G / B value rises significantly, the red saturation decreases, and the green / blue component is mixed; in the white region, the G / B value of some samples rises, presenting light pink or beige, and the brightness decreases; a few samples are still close to white, but the overall uniformity decreases; at the boundary, the RGB range fluctuates and expands, the color is uneven from “pink purple” to “dark red”, the transition zone is blurred, the plant state: low-concentration nitrogen oxide induces mild oxidative stress, and the red region partially decomposes; the white region color shifts; morphological parameters: the average leaf area of the red region is 84.7 cm 2 ; the average leaf area of the white region is 55.1 cm 2 , indicating that the plant is under slight stress, the average leaf perimeter of the red region is 38.2 cm; the average leaf perimeter of the white region is 28.1 cm, indicating that the leaf edge is slightly damaged, the average edge serration of the red region is 0.20; the average edge serration of the white region is 0.19, indicating that the leaf edge has a slight serration change, the average lesion area ratio of the red region is 0.07%, and the average lesion area ratio of the white region is 0.08%, indicating that the plant is under slight stress and a small amount of lesions appear; RGB range is R ∈ (162, 253), G ∈ (45, 232), B ∈ (18, 239), color characteristics: red area, R value fluctuates violently, part of the sample G / B value close to R value, color deviates to dark brown red or gray brown; another part maintains high R value, but G / B value rises to medium, showing pink purple; white area, G / B value rises significantly, close to light blue green or gray white, part of the sample shows light purple due to high B value; boundary, color mottling, from "deep red" to "light blue", no obvious transition rule, plant state: stress intensifies, red area pigment synthesis is hindered, part turns to blue green tone; white area presents mixed color, different parts of the plant are significantly different in pollution sensitivity, leading to color dispersion; morphological parameters: the average leaf area of the red area is 73.2 cm 2 ; the average leaf area of the white area is 48.6 cm 2 , indicating that the plant growth is inhibited to a certain extent, the average leaf perimeter of the red area is 35.6 cm; the average leaf perimeter of the white area is 26.3 cm, indicating that the leaf edge damage intensifies, the average edge serration of the red area is 0.32; the average edge serration of the white area is 0.31, indicating that the leaf edge damage intensifies, the average lesion area ratio of the red area is 0.15%; the average lesion area ratio of the white area is 0.16%, indicating that the plant is under significant stress, and the number of lesions increases; RGB range is R ∈ (176, 247), G ∈ (81, 249), B ∈ (2, 252), color characteristics: red area, R value decreases as a whole, high G / B value sample ratio increases, only local high saturated red remains, color presents patchy recession. White area, "dual state" appears, that is, part of the area maintains high brightness, another part becomes dark due to significant browning, overall deviates to gray green or light brown, boundary, color transition disappears, red and white areas are mixed, forming irregular color spots, plant state: red area degrades a lot; white area appears pigment mixing or oxidation browning; morphological parameters: the average leaf area of the red area is 61.3 cm 2 ; the average leaf area of the white area is 39.8 cm 2 , indicating that the plant is under serious stress, the average leaf perimeter of the red area is 32.1 cm; the average leaf perimeter of the white area is 23.4 cm, indicating that the leaf edge damage is serious, the average edge serration of the red area is 0.44; the average edge serration of the white area is 0.43, indicating that the leaf edge damage is serious, and the serration changes significantly, the average lesion area ratio of the red area is 0.27%; the average lesion area ratio of the white area is 0.28%, indicating that the plant is under serious stress, and the number of lesions increases significantly; At a nitrogen oxide concentration of 20 ppm, the RGB range was R (143, 253), G (105, 224), and B (29, 207) after 7 days of infection, the color characteristics were as follows: red area, almost disappeared, the R value was generally lower than 200, the color turned to dark brown or gray brown, only individual samples remained a high R value, but the G / B value increased, white area, the RGB range decreased comprehensively, the brightness was lost, and presented a beige, light brown or gray black color, close to the color of withered leaves, the plant state was as follows: the color dispersion reached a peak, there was no obvious two-color differentiation, the leaves were dark overall, and the surface might appear scorched spots; the morphological parameters were as follows: the average leaf area of the red area was 46.7 cm 2 ; the average leaf area of the white area was 30.2 cm 2 , indicating that the plant growth was severely inhibited, and part of the leaves appeared withered phenomenon, the average leaf perimeter of the red area was 28.5 cm; the average leaf perimeter of the white area was 20.7 cm, indicating that the leaf edge was severely damaged, and part of the leaves appeared withered phenomenon, the average edge serration of the red area was 0.57; the average edge serration of the white area was 0.56, indicating that the leaf edge was severely damaged, and the serration change was very significant, the average lesion area ratio of the red area was 0.39%, and the average lesion area ratio of the white area was 0.40%, indicating that the plant was under severe stress, the number of lesions increased significantly, and part of the leaves appeared with large area lesions.

[0029] The above-described embodiments only express several embodiments of the present application, the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent of the present application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application.

Claims

1. A method for predicting changes in nitrogen oxide infection in plants based on AI images, characterized in that, include: The reference image of the target plant object is processed based on a pre-trained image recognition model to extract the test image including plant color features and morphological parameters. The color features include the RGB range of the leaves, the main color tone, and the color uniformity. The reference image includes different regions of the plant leaves and the positions of multiple pre-labeled feature points. The image to be tested is input into a dynamic correlation model that describes the relationship between plant characteristics and nitrogen oxide concentration. The concentration prediction value of the color features of the morphological parameters and feature point locations affected by nitrogen oxides is calculated, and a color gradient visualization prediction image is generated. Based on the predicted image, the color features and morphological parameters before and after infection are compared to determine the degree of nitrogen oxide infection and dynamic change trend of the target plant. The construction method of the dynamic association model includes: collecting the test images of sample plants under different nitrogen oxide concentrations and different infection times, extracting the morphological parameters and color features of preset feature point positions in each image as target features; using the nitrogen oxide concentration and infection time as inputs and the target features as outputs, training the dynamic association model through a machine learning algorithm.

2. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 1, characterized in that, The pre-trained image recognition model processes a baseline image of the target plant object to extract a test image including plant color features and morphological parameters, including: Images of the target plant at unknown nitrogen oxide concentrations and infection times are acquired and used as the baseline images; Feature point selection method based on plant color type matching; Based on the selected feature points, multiple feature points are preset on the reference image, and morphological parameters and RGB color values ​​of each feature point are extracted. Based on the RGB color values ​​of each feature point, the color features of the feature points are statistically determined, and the image to be tested is obtained based on the color features and morphological parameters.

3. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 2, characterized in that, The feature point selection method based on plant color type matching includes: For solid-colored plants, the feature points include at least those located at the center of the leaf or the intersection of geometric diagonals; For bicolor or variegated plants, the feature points include the center of different color areas and the midpoint of the color transition zone.

4. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 3, characterized in that, The extraction of RGB color values ​​for each feature point includes: Extract the boundaries of plant tissue units in the baseline image and generate several feature points evenly distributed on the plant tissue units; Geometrically locate the feature points, generate a coordinate file, and obtain the RGB color values ​​of the feature points.

5. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 1, characterized in that, The pre-trained image recognition model is constructed using the following methods: Collect a set of plant images containing different nitrogen oxide concentrations and infection time tags; The images in the image set are labeled, and the labeling content includes morphological parameters, color features, and environmental tags; the morphological parameters include leaf area, perimeter, edge serration degree, and lesion area ratio; the environmental tags include nitrogen oxide concentration, infection time, shooting time, and plant growth stage; The image recognition model is obtained by training the labeled image set using a deep learning algorithm; Among them, the RGB color values ​​and morphological parameters of sample plants with zero nitrogen oxide concentration in the statistical environmental label were used to determine the initial morphological parameter range and the initial RGB health range.

6. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 5, characterized in that, The comparison of color characteristics and morphological parameters before and after infection to determine the degree and dynamic trend of nitrogen oxide infection in the target plant includes: The real-time RGB values ​​and morphological parameters of the feature points corresponding to the target plant object are compared with the initial morphological parameter range and the initial RGB health range; Based on the comparison results, the degree of nitrogen oxide infection of the target plant objects is matched, the nitrogen oxide concentration and infection time are determined, and the dynamic change trend is determined based on the nitrogen oxide concentration and infection time.

7. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 6, characterized in that, The method further includes: The real-time extracted RGB color values ​​and morphological parameters are compared with the initial RGB health range and the initial morphological parameter range; When the real-time RGB color value or morphological parameter exceeds the health threshold set based on the initial RGB health range and the initial morphological parameter range, the abnormal area is marked on the predicted image and an early warning is triggered.

8. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 7, characterized in that, The visualized prediction image includes: A color gradient diagram showing the degree to which different areas of a leaf are affected by nitrogen oxide concentration; Compare the RGB value change curves of feature points under different concentrations and infection times, and overlay a dynamic feature point evolution diagram showing the trend of morphological parameter changes. When the warning is triggered, an abnormal warning map is created by marking the abnormal area with a set color and shape.

9. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 1, characterized in that, When constructing the dynamic association model, different modeling strategies are used for plants of different color types: If the target plant is a solid-color plant, then construct a single model that covers its entire area; If the target plant is a bicolor or variegated plant, then construct independent sub-models for its different color regions.

10. The method for predicting changes in plant nitrogen oxide infection based on AI images according to claim 9, characterized in that, For bicolor or variegated plants, when generating the color gradient visualization prediction image, the model output is used to show the effect of pollutants on the color boundary, including the visual effect of blurred transition zones, disordered color boundaries, or the formation of irregular color spots.

Images