Smartphone-based method for chlorophyll content inversion of plant leaf

Abstract

A smartphone-based method for chlorophyll content inversion of a plant leaf. The method comprises: controlling a camera of a smartphone to photograph a plant leaf to be tested, so as to obtain a picture to be tested; extracting key leaf regions from said picture, and inputting the key leaf regions into an inversion model preset in the smartphone; matching a key leaf region of a target type with sub-models to obtain a target sub-model, so as to perform, on the basis of the target sub-model, chlorophyll content inversion on the key leaf region of the target type to obtain a content result; and fusing a plurality of content results to obtain a chlorophyll inversion result of said plant leaf, so as to complete the chlorophyll content inversion of the plant leaf. The method solves the problem of a smartphone-based technology for chlorophyll content inversion of a plant leaf having a certain degree of error during the actual inversion process.

Description

A Smartphone-Based Method for Inverting Chlorophyll Content in Plant Leaves Technical Field

[0001] This invention relates to the field of plant leaf chlorophyll content inversion technology, and in particular to a method for inverting plant leaf chlorophyll content based on a smartphone. Background Technology

[0002] To facilitate the inversion of chlorophyll content in plant leaves, the technology of inverting chlorophyll content in plant leaves, which is commonly used by people, is combined with the technology. This results in a smartphone-based technology for inverting chlorophyll content in plant leaves. This technology combines image processing and machine learning algorithms to achieve non-destructive testing and has become a low-cost and high-efficiency agricultural monitoring technology in recent years.

[0003] However, due to the limitations of data resources on smartphones, the accuracy of chlorophyll content inversion technology based on smartphones has decreased. It cannot accurately distinguish different parts of plant leaves, nor can it accurately invert the chlorophyll content of different parts of plant leaves, resulting in certain errors in the actual inversion process of chlorophyll content inversion technology based on smartphones.

[0004] Therefore, the present invention provides a new solution to this problem. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for retrieving chlorophyll content in plant leaves based on smartphones, which effectively solves the problem of certain errors in the actual retrieval process of smartphone-based plant leaf chlorophyll content retrieval technology.

[0006] The technical solution is: a method for retrieving chlorophyll content in plant leaves based on a smartphone, the method comprising:

[0007] The camera of the smartphone is controlled to capture an initial test image of the plant leaf under the target environment with the target shooting parameters, and the initial test image is preprocessed to obtain the test image;

[0008] Extract key leaf regions from the image to be tested, and input the key leaf regions into a pre-set inversion model on the smartphone; the inversion model is pre-trained; the inversion model has multiple sub-models; the key leaf regions have multiple types;

[0009] The key leaf regions of the target type are matched with the sub-model to obtain the target sub-model corresponding to the key leaf regions of the target type. The chlorophyll content of the key leaf regions of the target type is then inverted based on the target sub-model to obtain the chlorophyll content result. There are multiple chlorophyll content results.

[0010] The multiple content results are combined to obtain the chlorophyll inversion results of the leaves of the plant under test, so as to complete the inversion of chlorophyll content in plant leaves.

[0011] Furthermore, the fusion of the multiple content results to obtain the chlorophyll inversion results of the tested plant leaves includes:

[0012] Based on the historical performance of the multiple sub-models, corresponding weights are assigned, and the weights are normalized to obtain normalized weights;

[0013] The chlorophyll inversion result is obtained by weighting and fusing the content results based on normalized weights.

[0014] Further, after obtaining the chlorophyll inversion result by weighted fusion of the content results based on normalized weights, the process includes:

[0015] Determine multiple evaluation dimensions for the chlorophyll inversion results and call the evaluation methods corresponding to these multiple evaluation dimensions;

[0016] The chlorophyll inversion results are evaluated using evaluation methods corresponding to multiple evaluation dimensions to obtain evaluation results.

[0017] Further, the extraction of key leaf regions from the image to be tested includes:

[0018] The extraction method is set according to different parts of the leaf in the image to be tested;

[0019] The extraction method is called to extract features from the leaves of the image to be tested in order to obtain the key leaf regions of the corresponding parts.

[0020] Furthermore, before extracting the key leaf regions from the image to be tested, the image to be tested also needs to be segmented.

[0021] Furthermore, the preprocessing method includes at least one of the following: color space conversion, color correction, and geometric correction.

[0022] This invention provides a method for inverting chlorophyll content in plant leaves based on a smartphone. The method first controls the smartphone's camera to capture an initial test image of the plant leaf under a target environment using target shooting parameters, and preprocesses the initial test image to obtain a test image. Next, key leaf regions are extracted from the test image and input into a pre-set inversion model on the smartphone. The inversion model is pre-trained and has multiple sub-models. The key leaf regions have multiple types. Then, the key leaf regions of a target type are matched with the sub-models to obtain a target sub-model corresponding to the target type's key leaf regions. Chlorophyll content is then inverted based on the target sub-model to obtain multiple content results. Finally, the multiple content results are fused to obtain the chlorophyll inversion result of the plant leaf under test, thus completing the chlorophyll content inversion of plant leaves. This application is based on the extraction of multiple types of key leaf regions from the leaves of the plant under test for inversion, thereby accurately distinguishing different parts of the plant leaves and accurately inverting the chlorophyll content of the plant leaves in different parts. The multiple content results are fused to obtain the chlorophyll inversion result of the plant leaves under test, thereby greatly improving the accuracy of the plant leaf chlorophyll content inversion technology based on smartphones and increasing the possibility of the application of the plant leaf chlorophyll content inversion technology based on smartphones. Attached Figure Description

[0023] Figure 1 is a schematic flowchart of a method for retrieving chlorophyll content in plant leaves based on a smartphone, provided by the present invention.

[0024] Figure 2 is a schematic diagram of the process for obtaining chlorophyll inversion results provided by the present invention. Detailed Implementation

[0025] The foregoing and other technical contents, features, and effects of the present invention will be clearly presented in the following detailed description of the embodiments with reference to Figures 1-2. All structural contents mentioned in the following embodiments are with reference to the accompanying drawings.

[0026] Embodiments of the present invention will now be described with reference to the accompanying drawings.

[0027] This invention describes a method for retrieving chlorophyll content in plant leaves based on a smartphone, as shown in Figure 1. The method includes:

[0028] S101. Control the camera of the smartphone to capture an initial test image of the plant leaf under the target environment with the target shooting parameters, and preprocess the initial test image to obtain the test image.

[0029] S102. Extract the key leaf region from the image to be tested, and input the key leaf region into the inversion model preset on the smartphone; the inversion model is pre-trained; the inversion model has multiple sub-models; the key leaf region has multiple types;

[0030] S103. Match the key leaf region of the target type with the sub-model to obtain the target sub-model corresponding to the key leaf region of the target type, and perform chlorophyll content inversion on the key leaf region of the target type based on the target sub-model to obtain the content result; the content result has multiple values.

[0031] S104. The multiple content results are fused to obtain the chlorophyll inversion results of the plant leaves to be tested, so as to complete the inversion of chlorophyll content of plant leaves.

[0032] In step S101, when it is necessary to photograph the leaves of the plant to be tested, the smartphone is controlled to take a picture based on the target shooting parameters to obtain the initial test image. The target shooting parameters include fixed shooting distance, angle and focal length, exposure compensation and fixed fill light source. The exposure compensation is to turn off automatic exposure, manually set ISO to 100 and shutter speed to 1 / 60s to avoid brightness differences under different lighting conditions. The fixed fill light source needs to be photographed under a fixed fill light source (such as an LED fill light) to avoid ambient light interference. A fixed color temperature (5500-6500K) LED ring light or white LED board is used to avoid RGB channel deviation caused by fluctuations in the red / blue light ratio in natural light (such as enhanced blue light on cloudy days). It is also necessary to control the smartphone to use a dark box or standardized background board (such as black velvet) to reduce background reflection, and preprocess the initial test image to obtain the test image to facilitate accurate processing of the initial test image.

[0033] In the specific implementation of step S101, one embodiment is as follows: the preprocessing method includes at least one of the following: color space conversion, color correction, and geometric correction.

[0034] In this embodiment: the color space conversion includes RGB to HSV and CIELAB conversion. RGB to HSV involves extracting the Hue channel to determine if the leaves in the test image are yellowed (e.g., Hue > 100° may indicate nitrogen deficiency), and the S / V channel is used to segment the withered areas of the test leaves. CIELAB conversion calculates the mean of the a* channel (red-green axis), which has a correlation of 0.87 with the SPAD value determined by chemical methods, superior to directly using the G channel. Color correction involves adding a 24-color Macbeth ColorChecker to the test image and using polynomial regression (3rd order) to eliminate color deviations caused by different phone models. Geometric correction includes perspective transformation and morphological processing. Perspective transformation corrects the tilted leaves in the test image to a frontal view using a 4-point perspective transformation (OpenCV's getPerspectiveTransform) to avoid pixel distortion in edge areas. Morphological processing uses an opening operation (erosion followed by dilation) to remove noise caused by leaf hairs or water droplets, with a kernel size of 3×3 pixels. The color correction and geometric correction are used when the contrast between the leaves and the background in the test image is significant. Color space conversion (such as HSV channel filtering) or morphological operations (such as opening and closing operations) can be used to optimize the segmentation results. If the image contains multiple leaves, instance segmentation is needed to distinguish different leaves, and an independent mask should be generated for each leaf.

[0035] In step S102, key leaf regions in the test image are extracted based on a preset extraction algorithm. Before this, Canny edge detection and contour finding are used to select the contour with the largest area from the test image as the leaf region, and key leaf regions are selected from these. The key leaf regions have multiple types, including leaf regions that can represent different plant species, health / disease status, morphological differences, etc. The key leaf regions are input to the preset inversion model on the smartphone. The inversion model is pre-trained based on a training set that includes multiple leaf types and labels the exclusive features (such as color and texture) of each type of leaf. The inversion model has multiple sub-models, and the key leaf regions are inverted separately based on the multiple sub-models to obtain a more accurate inversion result. For the leaves in the test image, the leaf type (such as healthy / disease) is first determined by a classification model, and then the corresponding segmentation is called to extract leaf regions of different plant species or morphological differences.

[0036] In the specific implementation of step S102, one embodiment is as follows: before extracting the key leaf region in the image to be tested, it is necessary to segment the image to be tested.

[0037] In this embodiment, complex backgrounds (such as soil, sky, and other plants) may be misidentified as leaf areas, leading to distorted results in subsequent analyses (such as nutrient inversion and disease diagnosis). Deep learning models (such as U-Net and Mask R-CNN) or traditional algorithms (such as color thresholding and edge detection) can be used to separate leaves from the background. A 1080P image requires approximately 2 million pixels for direct processing, while only 100,000 to 500,000 pixels of leaf area need to be retained after segmentation. Different leaf types (such as conifers / broadleaves, healthy / diseaseed) have significant morphological differences, requiring targeted processing after segmentation and localization. This can be achieved through fast segmentation methods based on traditional image processing, semantic segmentation methods based on deep learning, or hybrid methods (traditional + deep learning). Fast segmentation methods based on traditional image processing include color space conversion + thresholding or edge detection + contour filtering. Hybrid methods achieve coarse segmentation by quickly locating the test image using color thresholding and also refine the segmentation by inputting the test image into a lightweight CNN (such as UNet-Tiny).

[0038] In a specific implementation of step S102, another embodiment exists: extracting the key leaf region from the image to be tested includes:

[0039] The extraction method is set according to different parts of the leaf in the image to be tested;

[0040] The extraction method is called to extract features from the leaves of the image to be tested in order to obtain the key leaf regions of the corresponding parts.

[0041] In this embodiment, different parts of the leaf in the test image exhibit different visual characteristics, such as leaf veins: high contrast, linear structure, usually darker than the leaf mesophyll; leaf mesophyll: large, uniform area with relatively consistent color / texture; leaf margin: the edge of the leaf, which may be serrated or wavy; lesions / insect holes: abnormal areas, with colors (yellow / brown / black) or textures different from healthy parts. Based on the high contrast and linear structure of the leaf veins, an edge detection (Canny) + morphological refinement extraction method was designed; for the leaf mesophyll, a color threshold segmentation + region growth method was used; for the leaf margin, edge detection + polygon fitting was used; and for lesions / insect holes, an anomaly detection method (such as GMM, AutoEncoder) was used for extraction, thereby specifically extracting the key leaf areas of the corresponding parts in the test image.

[0042] In step S103, by matching the key leaf region of the target type with the sub-model, a target sub-model corresponding to the key leaf region of the target type is obtained. The chlorophyll content of the key leaf region of the target type is then inverted based on the target sub-model to obtain the chlorophyll content result. That is, the target sub-model inverts the corresponding chlorophyll content result based on the features of the key leaf region, such as multispectral bands, texture features, and geometric features. Multiple chlorophyll content results are obtained. Different sub-models are sampled and processed for different key leaf regions to improve the accuracy of the inversion. The sub-model includes leaf morphology analysis, spectral information analysis, and temporal analysis. The leaf morphology analysis uses CNN (such as ResNet, EfficientNet) to extract texture, lesion, and color features, suitable for disease diagnosis or nutrient inversion. The spectral information analysis method is used to calculate the sub-model using 1D-CNN or spectral indices (such as NDVI) if the input is a multispectral image. For continuously monitored test images, LSTM or 3D-CNN is used to process temporal changes to obtain the corresponding chlorophyll content result.

[0043] In step S104, in order to obtain the final chlorophyll inversion result, it is necessary to fuse the multiple content results. The specific fusion method can be set according to the actual situation to complete the inversion of chlorophyll content in plant leaves, thereby improving the accuracy of chlorophyll inversion.

[0044] In a specific implementation of step S104, one embodiment is as follows: As shown in Figure 2, fusing the multiple content results to obtain the chlorophyll inversion result of the plant leaf to be tested includes:

[0045] Based on the historical performance of the multiple sub-models, corresponding weights are assigned, and the weights are normalized to obtain normalized weights;

[0046] The chlorophyll inversion result is obtained by weighting and fusing the content results based on normalized weights.

[0047] In this embodiment, evaluation indicators are selected based on the performance of the sub-models on historical data. Accuracy evaluation indicators include root mean square error (RMSE) and coefficient of determination (R²); stability evaluation indicators include standard deviation and PSI; and timeliness evaluation indicators include completion time, on-time rate, and response time. Models with better performance are assigned higher weights, and vice versa. Historical performance of the multiple sub-models on these evaluation indicators is collected, and the weight of each model is calculated based on historical performance. For example, weight = 1 / MSE, or a dynamic adjustment method such as exponential decay is used to normalize the weights so that the sum is 1, reducing computational load. This allows the inversion technology to be implemented on smartphones, applying normalized weights to the content results of each sub-model for weighted averaging to obtain the final chlorophyll inversion result. The dynamic weight allocation mechanism automatically assigns higher weights to models with better recent performance while maintaining the effective accumulation of historical information. This multi-dimensional evaluation system ensures the reliability of the results in terms of statistical accuracy, spatial rationality, and temporal stability.

[0048] In the specific implementation of step S104, another embodiment exists as follows: after obtaining the chlorophyll inversion result by weighted fusion of the content results based on normalized weights, the following is included:

[0049] Determine multiple evaluation dimensions for the chlorophyll inversion results and call the evaluation methods corresponding to these multiple evaluation dimensions;

[0050] The chlorophyll inversion results are evaluated using evaluation methods corresponding to multiple evaluation dimensions to obtain evaluation results.

[0051] In this embodiment, multiple evaluation dimensions are determined for the chlorophyll inversion results. These multiple evaluation dimensions can be evaluation dimensions mapped from evaluation indicators that show similar historical performance of the sub-models. For example, accuracy evaluation indicators include root mean square error (RMSE) and coefficient of determination (R²), resulting in an accuracy dimension. Stability evaluation indicators include standard deviation and PSI, resulting in a stability dimension. Timeliness evaluation indicators include completion time, on-time rate, and response time, resulting in a timeliness dimension. The calculation methods corresponding to the evaluation methods of the above indicators are publicly available and will not be described here. If the evaluation results meet a preset threshold, the chlorophyll inversion results are deemed usable; otherwise, the chlorophyll inversion results need to be regenerated. A comprehensive evaluation report can also be generated based on the evaluation results, indicating which aspects of the multiple sub-models perform well and which aspects need improvement. Visual charts (such as residual plots and spatial distribution maps) can also be generated to intuitively display the evaluation results.

Claims

1. A method for retrieving chlorophyll content of plant leaves based on a smartphone, characterized in that, The method includes: The camera of the smartphone is controlled to capture an initial test image of the plant leaf under the target environment with the target shooting parameters, and the initial test image is preprocessed to obtain the test image. Extract key leaf regions from the image to be tested, and input the key leaf regions into a pre-set inversion model on the smartphone; the inversion model is pre-trained; the inversion model has multiple sub-models; the key leaf regions have multiple types; The key leaf regions of the target type are matched with the sub-model to obtain the target sub-model corresponding to the key leaf regions of the target type. The chlorophyll content of the key leaf regions of the target type is then inverted based on the target sub-model to obtain the chlorophyll content result. There are multiple chlorophyll content results. The multiple content results are combined to obtain the chlorophyll inversion results of the leaves of the plant under test, so as to complete the inversion of chlorophyll content in plant leaves.

2. The smartphone-based method for retrieving leaf chlorophyll content of a plant leaf according to claim 1, wherein, The process of fusing the multiple content results to obtain the chlorophyll inversion results of the leaves of the plant under test includes: Based on the historical performance of the multiple sub-models, corresponding weights are assigned, and the weights are normalized to obtain normalized weights; The chlorophyll inversion result is obtained by weighting and fusing the content results based on normalized weights.

3. The smartphone-based method for retrieving leaf chlorophyll content of a plant leaf according to claim 2, wherein, After obtaining the chlorophyll inversion result by weighted fusion of the content results based on normalized weights, the process includes: Determine multiple evaluation dimensions for the chlorophyll inversion results and call the evaluation methods corresponding to these multiple evaluation dimensions; The chlorophyll inversion results are evaluated using evaluation methods corresponding to multiple evaluation dimensions to obtain evaluation results.

4. The smartphone-based method for retrieving leaf chlorophyll content of a plant leaf according to claim 1, wherein, The extraction of key leaf regions from the image to be tested includes: The extraction method is set according to different parts of the leaf in the image to be tested; The extraction method is called to extract features from the leaves of the image to be tested in order to obtain the key leaf regions of the corresponding parts.

5. The smartphone-based method for retrieving leaf chlorophyll content of a plant leaf according to claim 1, wherein, Before extracting the key leaf regions from the image to be tested, the image to be tested needs to be segmented.

6. The smartphone-based method for retrieving leaf chlorophyll content of a plant leaf according to claim 1, wherein, The preprocessing method includes at least one of the following: color space conversion, color correction, and geometric correction.

Images