A crop growth assessment method based on depth images

Abstract

The application discloses a crop growth evaluation method based on a depth image and belongs to the technical field of crops. The crop growth evaluation method based on the depth image comprises the following steps: S1, growth classification model training: widely obtaining RGB images of field corn in different growth periods in the vertical direction, and ensuring that samples under various growth environments and conditions are covered; accurately and comprehensively labeling corn leaf positions and regions on the RGB images by professional personnel, and the labeling should include detailed information such as the outline, size and position of the leaf; using a large amount of data after labeling, training a target recognition model by using an advanced machine learning algorithm, and enabling the target recognition model to accurately and efficiently recognize the corn leaf; and the application is used to solve the problem that the adaptability of the existing image acquisition mode is also relatively limited, the obtained samples are not rich and diversified enough, and the generalization ability of a subsequent evaluation model is affected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crops, in particular to a crop growth evaluation method based on depth images. BACKGROUND

[0002] With the development of agricultural modernization, accurate evaluation of crop growth is of great significance for optimizing agricultural production management and improving crop yield and quality. Traditional crop growth evaluation methods mainly rely on manual observation and experience-based judgment, which is not only inefficient and subjective, but also difficult to obtain comprehensive and accurate crop growth information. In recent years, with the continuous progress of image technology and computer algorithms, image-based crop growth evaluation methods have gradually emerged. However, existing methods still have many shortcomings.

[0003] In terms of image acquisition, only single direction or single type images are usually acquired, which cannot fully reflect the three-dimensional growth conditions of crops in different growth periods. For complex growth environments and conditions, the adaptability of existing image acquisition methods is also limited, resulting in insufficient and diverse samples, affecting the generalization ability of subsequent evaluation models. In the process of depth image acquisition and processing, the precision of camera calibration is insufficient, making it difficult to effectively handle problems such as occlusion, reflection and low-texture areas, resulting in poor quality of depth images and affecting subsequent crop growth analysis. Therefore, a crop growth evaluation method based on depth images is proposed. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the background art, and to provide a crop growth evaluation method based on depth images. In order to more accurately, efficiently and comprehensively evaluate crop growth, it is necessary to develop a crop growth evaluation method based on depth images to overcome the shortcomings of existing technology and provide stronger support for agricultural production.

[0005] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0006] A crop growth evaluation method based on depth images, comprising the following steps:

[0007] S1 Growth classification model training:

[0008] -Extensively acquire RGB images of field corn in different growth periods in the vertical direction, ensuring that samples under various growth environments and conditions are covered;

[0009] -Professional personnel accurately and comprehensively label the position and area of corn leaves on the RGB image, including detailed information such as the outline, size and position of the leaves;

[0010] - Utilize the large amount of labeled data to train the target recognition model using advanced machine learning algorithms, enabling it to accurately and efficiently identify corn leaves. The model should have the ability to recognize corn leaves under different lighting conditions, angles, and complex backgrounds.

[0011] S2 Determine the model parameter library for different growth stages:

[0012] - Depth image acquisition:

[0013] - Perform monocular camera calibration, including but not limited to determining the camera's internal and external parameters, to accurately obtain the geometric information of the image;

[0014] - Perform binocular camera calibration by accurately determining the internal and external parameters of the binocular camera and the relative position relationship between the two cameras, providing an accurate basis for subsequent stereo matching;

[0015] - Use efficient stereo matching algorithms to match the images obtained by the binocular camera and obtain depth information, thereby generating a depth image;

[0016] - Stereo matching.

[0017] S3 Corn leaf angle calculation:

[0018] - Use advanced local feature recognition technology to accurately identify the local features of corn leaves, including but not limited to texture, shape, and edge features;

[0019] - Determine accurate matching parameters to ensure accuracy and reliability when comparing RGB images and depth images;

[0020] - Strictly compare RGB images and depth images to obtain more comprehensive and accurate leaf information;

[0021] - Run the target recognition model trained in S1 to obtain all leaf regions of the field corn, and loop and crop these regions on the depth image;

[0022] - For each cropped single region, calculate its maximum and minimum depth values, combine the detailed parameters of the depth camera, and accurately obtain the highest position, lowest position, and distance between the two positions of a single leaf. Then, according to the triangular relationship of the vertical triangle, accurately calculate the leaf angle.

[0023] Preferably, during the process of obtaining vertical direction RGB images of field corn in different growth stages, samples of field corn of different varieties, different planting densities, and different geographical regions are included to enhance the generalization ability of the model.

[0024] Preferably, when labeling the position and area of corn leaves on the RGB image, a unified and strict labeling standard is followed, and professional training and quality control are conducted on the labeling personnel to ensure the accuracy and consistency of the labeling.

[0025] Preferably, the machine learning algorithm used to train the target recognition model has the ability to adaptively adjust and optimize model parameters to cope with different data distributions and complex scenes.

[0026] Preferably, when calibrating monocular cameras and binocular cameras, high-precision calibration tools and standard reference objects are used to improve the accuracy and reliability of the calibration.

[0027] Preferably, in the process of stereo matching, advanced stereo matching algorithms that can handle occlusion, reflection and low-texture areas are used to ensure the quality of the obtained depth image.

[0028] Preferably, in local feature recognition, the feature recognition technology used can adapt to the morphological changes of leaves and the subtle differences in the growth process.

[0029] Preferably, when determining the matching parameters, a large number of experiments and data analysis are used to optimize the parameters to improve the accuracy of the RGB image and depth image comparison.

[0030] Preferably, in the process of calculating the leaf angle, the measurement error is considered and an error correction method is used to improve the accuracy of the calculation result.

[0031] Preferably, the entire evaluation method has real-time and scalability, and can adapt to large-scale crop monitoring and evaluation needs of new crop types.

[0032] Compared with the prior art, the present application provides a crop growth evaluation method based on depth images, which has the following beneficial effects:

[0033] 1、The present application can more accurately identify corn leaves by widely obtaining vertical direction RGB images of field corn in different growth stages, various growth environments and conditions, and performing accurate and comprehensive labeling, combining advanced machine learning algorithm to train target recognition model, reduce misjudgment, and provide reliable data basis for subsequent growth evaluation;

[0034] 2、The present application enhances the model generalization ability: covers corn samples of different varieties, planting densities and geographical regions, so that the model can adapt to various actual planting conditions, not limited to specific conditions, so as to effectively evaluate the growth in different scenes;

[0035] 3. This invention is adaptable to complex scenarios: The trained target recognition model has the ability to adapt to different lighting, angles and complex backgrounds, and can accurately identify corn leaves in complex and ever-changing field environments without being disturbed by external factors;

[0036] 4. Improve depth image quality: Employ high-precision camera calibration tools and advanced stereo matching algorithms to effectively handle occlusion, reflection, and low-texture areas, acquiring high-quality depth images. This provides clear image information for accurately assessing crop growth. Through advanced local feature recognition technology and precise matching parameters, combined with depth camera parameters, the leaf angle is accurately calculated, reflecting the crop's growth status in greater detail. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structure of a crop growth assessment method based on depth images according to the present invention.

[0038] Figure 2 This is a schematic diagram of a crop growth assessment method based on depth images according to the present invention.

[0039] Figure 3 This is a schematic diagram of a crop growth assessment method based on depth images according to the present invention;

[0040] Figure 4 This is a screenshot of the interface of the teacher-side software for the crop growth assessment method based on depth images according to the present invention. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0042] Reference Figures 1-4 A method for assessing crop growth based on depth images: Experimental environment and material preparation:

[0043] Multiple field maize planting areas with different geographical locations, soil conditions, and climate characteristics were selected as experimental sites. High-resolution monocular and binocular cameras, as well as corresponding calibration tools and standard references, were prepared. At the same time, a professional annotation team was formed and equipped with high-performance computers for data processing and model training.

[0044] Step 1: Training the Growth Classification Model

[0045] A large number of field corn vertical direction RGB images were collected from multiple experimental fields during different growth stages, covering various common corn varieties and different planting densities. The location, contour, size, and area of the corn leaves on the RGB images were accurately labeled by trained professionals following uniform labeling standards.

[0046] Using these labeled data, a deep learning-based convolutional neural network algorithm was used to train the target recognition model. After multiple iterations and optimization, the model can accurately identify corn leaves under different lighting conditions (such as strong light, weak light, and shadow), different shooting angles (shooting from below, shooting from above, and shooting from the side), and complex backgrounds (such as weeds and other crops).

[0047] Step two: Depth image acquisition

[0048] First, monocular camera calibration was performed using a high-precision calibration board to determine the camera's internal parameters (such as focal length and principal point coordinates) and external parameters (camera position and attitude), thus accurately obtaining the geometric information of the image.

[0049] Then, binocular camera calibration was performed by measuring the relative position and attitude relationship between the binocular cameras, laying the foundation for subsequent stereo matching.

[0050] During image acquisition, a stereo matching algorithm that can adapt to occlusion, reflection, and low-texture areas was used to match the images obtained by the binocular cameras, successfully generating high-quality depth images.

[0051] Step three: Corn leaf angle calculation

[0052] Local feature recognition techniques based on edge detection and texture analysis were used to accurately identify the texture, shape, and edge features of corn leaves.

[0053] Through extensive experiments and data analysis, the optimal matching parameters were determined to ensure the accuracy of the RGB image and depth image comparison.

[0054] Strictly comparing the RGB image and depth image, comprehensive and accurate leaf information was obtained. The trained target recognition model was run to obtain all the leaf regions of the field corn, and the depth image was looped and cropped.

[0055] For each cropped region, the maximum and minimum values of its depth were calculated, combined with the detailed parameters of the depth camera, such as baseline distance and focal length, to accurately calculate the highest position, lowest position, and distance between the two of a single leaf. Then, according to the triangular relationship of the vertical triangle and the corresponding error correction method, the leaf angle was accurately calculated.

[0056] Experimental results and analysis

[0057] After evaluating the growth of corn in multiple experimental fields, the method of the present application can quickly and accurately obtain relevant information and leaf angle of corn leaves, and the evaluation efficiency is improved by 20% compared with the traditional manual evaluation method, and the accuracy of the evaluation result is improved by 35%.

[0058] In corn fields of different varieties, planting densities and geographical regions, the model shows good generalization ability and can accurately evaluate the growth of crops.

[0059] By monitoring and analyzing the growth of crops in real time, farmers can adjust planting management strategies such as fertilization, irrigation, pest control, etc., to improve the yield and quality of corn.

[0060] Reference Figures 1-4 In the process of obtaining vertical direction RGB images of corn in different growth stages, corn samples of different varieties, different planting densities and different geographical regions are covered to enhance the generalization ability of the model. By covering samples of multiple varieties, planting densities and geographical regions, the trained model can adapt to various actual corn planting conditions, not just specific conditions, so that accurate growth evaluation can be performed in different agricultural environments.

[0061] Reference Figures 1-4 When labeling the position and area of corn leaves on the RGB image, a unified and strict labeling standard is followed, and professional training and quality control are conducted on the labeling personnel to ensure the accuracy and consistency of the labeling. Through unified and strict labeling standards and professional training, high-quality labeled data can be ensured, and errors and biases can be reduced. Unified labeling standards and quality control can ensure that labeling results produced by different times and different personnel are consistent, which helps the repeatability of scientific research and experiments. In the field of agriculture, there is a great demand for automated analysis of RGB images, and accurate labeling can support the large-scale application of algorithms, improve the efficiency and accuracy of agricultural production.

[0062] Reference Figures 1-4 The machine learning algorithm used to train the target recognition model has the ability to adaptively adjust and optimize model parameters to cope with different data distributions and complex scenarios. Adaptive adjustment of model parameters enables the algorithm to better adapt to different data distributions, thereby improving the generalization ability of the model on unknown data. By optimizing model parameters to cope with complex scenarios, the algorithm can more robustly handle noise and outliers, reducing the risk of overfitting.

[0063] Reference Figures 1-4In the process of monocular camera calibration and binocular camera calibration, high-precision calibration tools and standard references are used to improve the accuracy and reliability of the calibration. High-precision calibration tools can provide more accurate physical dimensions and geometric shapes, which helps to improve the accuracy of measurements during the calibration process. Through accurate camera calibration, image distortion and perspective correction can be handled more accurately, thereby improving the accuracy of image analysis.

[0064] Referring to Figures 1-4 In the process of stereo matching, advanced stereo matching algorithms that can handle occlusion, reflection and low-texture areas are used to ensure the quality of the obtained depth image. In the presence of occlusion and reflection, traditional algorithms often have difficulty in accurate matching, while advanced algorithms enhance the processing capability for these complex situations by fusing multi-scale local features and depth features.

[0065] Referring to Figures 1-4 In local feature recognition, the feature recognition technology used can adapt to the morphological changes of leaves and the subtle differences in the growth process, and can capture the subtle changes in leaf morphology, so that the recognition system can more accurately distinguish between different types or varieties of plant leaves. By adapting to the morphological changes of leaves and the subtle differences in the growth process, the feature recognition technology can better handle noise and variations in the image, thereby maintaining a high recognition rate under different environmental conditions.

[0066] Referring to Figures 1-4 In determining the matching parameters, a large number of experiments and data analysis are used to optimize the parameters to improve the accuracy of the RGB image and depth image comparison. By optimizing the matching parameters through a large number of experiments and data analysis, the feature point matching between the RGB image and the depth image can be more accurate, thereby improving the overall image registration quality. Proper parameter setting can reduce the waste of computing resources, improve the running efficiency of the algorithm, and make the stereo matching process more rapid and efficient.

[0067] Referring to Figures 1-4 In the process of calculating the leaf angle, the measurement error is considered and an error correction method is used to improve the accuracy of the calculation results. Through the error correction method, the measurement error can be significantly reduced, and the accuracy of the leaf angle calculation can be improved. The error correction method can improve the reliability of the measurement data, making the obtained leaf angle data more reliable and providing more accurate basic data for subsequent analysis and research.

[0068] Referring to Figures 1-4The entire evaluation method has real-time and scalability, and can adapt to large-scale crop monitoring and evaluation requirements of new crop types. The scalability makes the evaluation method not only suitable for current crop types, but also easily extended to new crop types. Real-time data provides more accurate crop growth information, helping farmers and agricultural enterprises make more scientific planting and management decisions.

[0069] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes to the technical solutions and inventive concepts of the present application within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

[0070] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, different embodiments or examples described in the present specification and the features of different embodiments or examples can be combined and combined by those skilled in the art without contradiction.

[0071] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for crop growth assessment based on depth images, characterized in that, Comprising the following steps: S1 Longness classification model training: - Obtain RGB images of field corn in different growth stages in the vertical direction, ensuring that samples under various growth environments and conditions are covered; - Label the location and area of corn leaves on the RGB image by professional personnel, which should include the outline, size, and location of the leaves; - Use a large amount of labeled data to train a target recognition model using a convolutional neural network algorithm based on deep learning, enabling the model to recognize corn leaves under different lighting, angles, and complex backgrounds; S2 Determine the model parameter library for different growth stages: - Depth image acquisition: - Perform monocular camera calibration, including but not limited to determining the camera's internal and external parameters to accurately obtain the geometric information of the image; - Perform binocular camera calibration by determining the internal and external parameters of the binocular camera and the relative position relationship between the two cameras; - Use stereo matching algorithms to match the images obtained by the binocular camera to obtain depth information and generate depth images; - Stereo matching; S3 Calculation of corn leaf angle: - Use local feature recognition technology based on edge detection and texture analysis to identify the local features of corn leaves, including but not limited to texture, shape, and edges; - Determine matching parameters to ensure accuracy and reliability when comparing RGB images and depth images; - Strictly compare RGB images and depth images to obtain leaf information; - Run the target recognition model trained in S1 to obtain all leaf regions of field corn and loop through these regions on the depth map; - For each cropped single region, calculate the maximum and minimum values of its depth, combine the detailed parameters of the depth camera, including baseline distance and focal length, to obtain the highest position, lowest position, and distance between the two for a single leaf, then calculate the leaf angle based on the triangular relationship of the vertical triangle.

2. The crop growth assessment method based on depth images according to claim 1, characterized in that: In the process of obtaining RGB images of field corn in different growth stages in the vertical direction, samples of field corn of different varieties, different planting densities, and different geographical regions are covered.

3. The crop growth assessment method based on depth images according to claim 1, characterized in that: When labeling the location and area of corn leaves on the RGB image, follow the unified labeling standard.

4. The crop growth assessment method based on depth images according to claim 1, characterized in that: When performing monocular camera calibration and binocular camera calibration, use high-precision calibration tools and standard references.

5. The method for crop growth assessment based on depth images according to claim 1, characterized in that: In the stereo matching process, advanced stereo matching algorithms that can handle occlusion, glare, and low-texture areas are used.

6. The crop growth assessment method based on depth images according to claim 1, characterized in that: When determining matching parameters, optimize the parameters through extensive experiments and data analysis.

7. The method of claim 1, wherein: In the process of calculating the leaf angle, consider the existing measurement errors and use error correction methods to improve the accuracy of the calculation results.

Images