A method and system for obtaining high-throughput plant height phenotype of soybean in a plant factory environment

Abstract

The application discloses a kind of plant factory environment under soybean high-flux plant height phenotype acquisition method and system, belong to depth image acquisition and identification technical field, method includes: obtaining the depth image and RGB image containing soybean plant and potted platform;Image registration is carried out, and based on registration parameter, depth image is synchronously spliced, and depth splicing chart is obtained;Depth splicing chart is converted into pseudo-color depth chart and actual gray chart in sequence, and depth gray chart is obtained;The reference gray value of ground area is extracted from depth gray chart, and the correction coefficient is calculated based on the actual physical height of potted platform distance ground and the gray value of calibration reference point;Based on the top gray value of plant and the minimum gray value of image, the actual plant height of soybean plant is obtained by gray difference proportion mapping.The application realizes the nondestructive, high-flux, high-precision automatic measurement of soybean seedling stage plant height under plant factory environment, and provides reliable technical support for soybean generation breeding and phenotype identification.

Description

Technical Field

[0001] This invention relates to the field of depth image acquisition and recognition technology, and in particular to a method and system for high-throughput acquisition of soybean plant height phenotype in a plant factory environment. Background Technology

[0002] Plant factory breeding, as one of the core technologies of modern agricultural breeding, can break the limitations of the natural growth cycle of crops by artificially controlling environmental factors such as light, temperature, humidity, and nutrients. This enables continuous multi-generation breeding of crops, significantly shortening the breeding cycle and accelerating the breeding process. It is of irreplaceable importance for rapidly breeding new soybean materials, screening superior lines, and overcoming traditional breeding bottlenecks. In the standardized and controllable growth environment of a plant factory, the rapid and accurate acquisition of key phenotypic data such as soybean seedling height is fundamental for conducting research on soybean growth patterns, trait genetic analysis, and screening of superior lines, thereby improving the accuracy of breeding material evaluation and breeding efficiency. Only by achieving rapid acquisition and accurate analysis of phenotypic data can breeding materials with superior traits be screened in a timely manner, providing efficient and reliable technical support for soybean breeding. Soybean seedling height, as an important growth phenotypic indicator, directly affects the screening effect and breeding progress of new materials and superior lines in generation-wise breeding through its accuracy and efficiency.

[0003] Traditional methods for measuring soybean seedling height often rely on manual measurement with a measuring tape, which suffers from large errors, low efficiency, and strong subjectivity, hindering subsequent breeding research on soybean plant height. In recent years, with the rapid development of smart agriculture and high-throughput phenomics technologies, non-destructive plant phenotyping techniques based on deep learning and image recognition have been increasingly applied to crop phenotyping. Current vision-based plant height measurement methods primarily use a monocular RGB camera with a ruler, capturing two-dimensional RGB images and establishing a pixel-to-physical-size ratio to calculate plant height. However, this method is highly sensitive to image resolution, lighting conditions, shooting angle, and ruler placement. Ruler tilt, changes in shooting distance, cluttered backgrounds, or leaf obstruction significantly reduce measurement accuracy. Furthermore, relying solely on RGB images for phenotypic data lacks depth spatial information, resulting in limited feature extraction and issues such as inaccurate baseline positioning and height calculation errors. Existing image recognition and phenotypic acquisition technologies still cannot effectively solve the technical challenges of high-throughput, high-precision, and automated measurement of soybean seedling height. There is an urgent need for a method to acquire plant height that is not limited by scales and two-dimensional images, has a unified measurement benchmark, and is more accurate, so as to achieve stable, objective, and high-precision automated acquisition of soybean plant height and provide technical support for high-throughput phenotypic identification and precision breeding of soybeans. Summary of the Invention

[0004] To address the shortcomings of existing soybean plant height measurements, which rely on manual methods or two-dimensional RGB images, resulting in low accuracy and a lack of depth information, this invention proposes a high-throughput soybean plant height phenotypic acquisition method and system in a plant factory environment. This method acquires soybean plant height data by recognizing parameters through depth images, while simultaneously introducing ground calibration methods and height correction coefficients to standardize and correct the measurement benchmark, eliminating systematic errors caused by placement distance, angle, and environmental noise. Through individual plant marking and feature recognition, visual information is transformed into quantitative data, significantly improving the accuracy and objectivity of soybean seedling height acquisition and laying a high-quality phenotypic data foundation for subsequent in-depth analysis. By acquiring soybean plant growth dynamics, this provides growers with forward-looking management guidance.

[0005] This application discloses a method for obtaining the high-throughput plant height phenotype of soybean in a plant factory environment, including the following steps: S1. Build a high-throughput phenotypic identification platform to obtain depth images and RGB images of soybean plants and potted platforms; S2. Extract feature points from the RGB image for image registration, and simultaneously stitch the depth image based on the registration parameters to obtain a depth stitched image. S3. The depth mosaic image is sequentially converted into a pseudo-color depth image and an actual grayscale image to obtain a depth grayscale image; S4. Extract the reference gray value of the ground area from the depth gray map, and calculate the correction coefficient based on the actual physical height of the potted plant platform from the ground and the gray value of the calibration reference point. S5. Based on the gray value at the top of the plant and the minimum gray value of the image, combined with the reference gray value, the fixed height of the camera and the correction coefficient, the actual plant height of the soybean plant is obtained by mapping the gray value difference ratio.

[0006] Preferably, S2 includes: The feature point recognition function of RGB images is used to extract the feature points of the overlapping areas of adjacent RGB images. Image registration is achieved through homography matrix, and multiple RGB images are stitched into a complete RGB stitched image. Then, using the registration parameters of the RGB stitched image as a reference, multiple corresponding depth images are simultaneously stitched together to obtain a depth stitched image corresponding to the spatial location.

[0007] Preferably, S3 includes: Original depth map normalization: according to the formula Depth values ​​in the original depth map Normalization is performed to obtain normalized gray values. ;in, The normalized grayscale value. This represents the depth value of a pixel in the original depth map. This represents the minimum depth value in the original depth map. The table shows the maximum depth values ​​in the original depth map.

[0008] Pseudo-color mapping: The normalized grayscale values ​​are mapped using a color lookup table. Mapped to RGB three-color values, generating a pseudo-color depth map; Grayscale image restoration: Reverse reconstruction of RGB to grayscale values The lookup table is used to restore the pseudo-color depth map to the actual grayscale map, thus obtaining the depth grayscale map.

[0009] Preferably, the formula for calculating the correction coefficient is as follows:

[0010]

[0011]

[0012] in, This is a calibration item for the grayscale stability of the calibration reference point. This is the initial compensation term for system error. For correction factor, For the first The grayscale value of each ground calibration reference point For the first The grayscale value of each ground calibration reference point This refers to the actual physical height of the potted plant platform from the ground. This is the theoretical calculation distance, that is, the theoretical calculation distance from the potted plant stand to the ground. .

[0013] Preferably, the extraction of the reference grayscale value of the ground area in S4 includes: The depth grayscale image is divided into multiple sub-areas according to preset rules. Within the potted plant platform area of ​​each sub-area, the HSL color features of the RGB image and the standard deviation features of the depth image are used for dual screening to remove areas containing plants and retain pure ground areas. Calculate the median grayscale value of the pure ground area in each cell, and then take a weighted average of the median ground grayscale values ​​of all cells to obtain the baseline grayscale value for the entire platform. . Preferably, the formula for calculating the actual plant height of the soybean plant is as follows:

[0014]

[0015]

[0016]

[0017]

[0018] in, This represents the grayscale difference between the ground and the top of the plant. The grayscale difference between the ground and the minimum grayscale value in the image. This is the gray-scale difference normalization ratio term. For camera height calibration and grayscale fusion, For the final plant height, The ground reference gray value, This represents the grayscale value of the topmost part of a single soybean plant. This represents the minimum grayscale value of the entire depth image. Indicates the relative proportion of grayscale difference. For camera height calibration, This represents the actual vertical physical distance between the camera and the ground reference.

[0019] Preferably, the grayscale value of the topmost part of the soybean plant is obtained through the following steps: Manually click on the growth point of the target plant in the RGB mosaic image. Determine the region of interest centered on the click location. Extract the grayscale values ​​of the depth grayscale image within this region. After removing outliers using the 3σ principle, take the minimum grayscale value as the base value. .

[0020] Preferably, the method further includes the following steps: generating a table of plant height data for the soybean plants to be tested, a grayscale stability box plot for verifying the stability of the ground calibration grayscale values, and a heatmap of the average plant height for representing the fluctuation of plant height within the population.

[0021] This application also discloses a high-throughput soybean plant height phenotypic acquisition system in a plant factory environment, used to perform the above-described method, including: The image acquisition module is used to acquire depth images and RGB images containing soybean plants and potted platforms using a depth camera; The image stitching module is used to stitch together multiple acquired depth images and RGB images to obtain a depth stitched image and an RGB stitched image; The image preprocessing module is used to preprocess the depth mosaic image, converting the original depth image into a pseudo-color depth image and an actual grayscale image in sequence to obtain a depth grayscale image. The ground calibration module is used to extract the ground reference gray value and the gray value of the ground calibration reference point in the potted plant platform area based on the RGB mosaic image and the depth grayscale image. The correction factor calculation module is used to calculate the correction factor based on the gray value of the ground calibration reference point and the actual physical height of the potted plant platform from the ground. The plant height extraction and calculation module is used to determine the region of interest at the top of the plant under test on the RGB mosaic image, extract the minimum effective gray value of the corresponding region in the depth grayscale image and the minimum gray value of the whole image, and calculate the plant height of soybean plants. The data output module is used to output the plant height data of all plants to be tested.

[0022] The beneficial effects of this invention are: (1) The present invention uses a depth camera to acquire depth images and combines RGB images to perform multi-source information fusion, making full use of the spatial distance features in the depth information, avoiding the defects of traditional two-dimensional RGB images that rely on scales and are easily affected by shooting angles and background interference, thus improving the robustness and accuracy of plant height measurement.

[0023] (2) This invention uses a ground calibration method, with a potted plant platform as a fixed reference, to extract the gray value of the ground reference and introduce a correction coefficient to compensate for system errors, thereby eliminating the influence of factors such as camera installation angle, lens distortion, depth sensor accuracy deviation and uneven lighting, and realizing the unification and standardization of measurement reference.

[0024] (3) This invention proposes a complete depth image preprocessing process, including the conversion of the original depth image to a pseudo-color image and then the restoration of the grayscale image, which is convenient for human observation and computer processing. At the same time, combined with enhancement methods such as bilateral filtering and histogram equalization, the grayscale difference between the plant and the ground is highlighted, which is beneficial for subsequent feature extraction.

[0025] (4) This invention achieves high-throughput and automated plant height data extraction through steps such as automatic screening of pure ground areas, random multi-point calibration, and adaptive calculation of correction coefficients. It can also output results in various forms such as plant height Excel tables, grayscale stability box plots, and plant height mean heatmaps, providing intuitive data support for breeding research. Attached Figure Description

[0026] Figure 1 This is a flowchart of a method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a high-throughput phenotyping platform according to an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided with reference to the accompanying drawings and embodiments.

[0028] This application discloses a method for obtaining the high-throughput plant height phenotype of soybean in a plant factory environment, the process of which is as follows: Figure 1 As shown, it includes the following steps: S1. Build a high-throughput phenotypic identification platform to obtain depth images and RGB images of soybean plants and potted platforms.

[0029] Building such Figure 2 The high-throughput phenotyping platform shown includes a potted plant platform 6, on which a standard colorimetric card 5 and multiple trays 7 are mounted, each tray 7 containing a soybean plant 8. A slide rail 2 is mounted above the potted plant platform 6, and a depth camera 1 is mounted on the slide rail. The slide rail 2 is used to slide the depth camera 1; in this embodiment, the depth camera 1 is a D345. The depth camera 1 is connected to a computer 4 via a cable 3.

[0030] Specifically, in this embodiment, the potted plant platform 6 is 5m long and 1.0m wide. A total of 112 soybean materials were planted, with 3 pots planted per material, and 21 pots of 7 varieties were placed in each tray. There are a total of 16 trays 7. For ease of image processing and data calculation, the 16 trays are named 1-16 plots respectively.

[0031] A depth camera D345 is fixed on a sliding bracket, which is mounted on a slide rail 2 parallel to the potted plant platform 6. The slide rail 2 is driven by a conveyor belt to achieve uniform sliding. The depth camera D345 is connected to the industrial control computer 4 via a dedicated cable 3 to ensure stable transmission of image data. The camera lens is vertically oriented towards the potted plant platform 6. After calibration, the vertical physical distance between the depth camera 1 and the potted plant platform is fixed at 75cm. A standard colorimetric card (SpyderChecker24) is placed at a fixed position on the potted plant platform 6 for subsequent image color and grayscale calibration to eliminate image deviations caused by uneven lighting. The control computer 4 is pre-installed with a Python environment and image processing and data analysis libraries such as OpenCV, NumPy, and Matplotlib. An automatic plant height data extraction program is written and debugged, including functional modules such as image stitching, preprocessing, ground calibration, correction coefficient calculation, plant height extraction, and result output.

[0032] Depending on the specific experimental requirements, different data acquisition times were selected. In this embodiment, to determine the optimal time for acquiring soybean seedling height data under the conditions of this plant factory, depth cameras were used to take photos every morning at 9:00 AM for 5-25 days after soybean emergence. Simultaneously, plant height was manually measured on days 5, 10, 15, 20, and 25 after emergence to verify the accuracy of the data acquisition.

[0033] On the day the depth map is acquired, first, open the system connected to the depth camera 1. This can be done on-site using the provided computer 4 or remotely by opening the device. Then, set the specific operating procedure, including the shooting mode of the depth camera 1 (such as interval time and interval distance), the storage address of the depth images, and the storage method. The shooting mode of the depth camera 1 needs to be determined based on the size of the potted plant platform 6 and the spacing between the potted plant trays 7. This embodiment clearly demonstrates that, under the existing conditions of the potted plant platform 6, setting an interval of 60cm / 30s to capture a depth map yields the best results for subsequent image preprocessing and stitching.

[0034] After the program is set up, depth camera 1 will slide along slide rail 2 and take pictures, guided by the conveyor belt. The acquired images will be stored in a pre-set file address. The D345 depth camera can acquire depth images (TIF format) and RGB images. In this embodiment, the depth image is used to obtain the actual plant height distance, and the RGB image is used to extract feature points during stitching to better achieve depth map stitching and for region selection when acquiring plant height data.

[0035] After acquiring the image, the plant height was extracted following the process and steps of image preprocessing, threshold setting, ground calibration benchmark extraction, correction coefficient extraction, and plant height data extraction. All of the above steps were performed in Python.

[0036] S2. Extract feature points from the RGB image for image registration, and simultaneously stitch the depth image based on the registration parameters to obtain a depth stitched image.

[0037] Using OpenCV available in Python, and considering the data acquisition needs of 16 cells, depth camera 1 captured 16 depth images and 16 RGB images. Utilizing the feature point recognition function of the RGB images, feature points in overlapping areas of adjacent RGB images were extracted. Image registration was achieved using a homography matrix, and the 16 RGB images were stitched together into a single 8-column, 2-row RGB mosaic. Based on the registration parameters of the RGB mosaic, the corresponding 16 depth images were simultaneously stitched together to obtain an 8-column, 2-row depth mosaic corresponding to the spatial positions of the RGB mosaic, ensuring a one-to-one correspondence between the pixel positions of the depth images and the RGB images.

[0038] S3. Convert the depth mosaic image into a pseudo-color depth image and an actual grayscale image in sequence to obtain a depth grayscale image.

[0039] The 16x / 8x single-channel depth images acquired using depth camera 1 are difficult for the human eye to discern. Therefore, to facilitate subsequent image processing and data analysis, this embodiment uses a conversion method of "original depth image - pseudo-color depth image - grayscale image" to achieve depth image preprocessing. The specific conversion principle is as follows: Gray-level normalization of the original depth image. First, the gray-level values ​​in the original depth image are normalized to facilitate image conversion using OpenCV's apply Colormap tool.

[0040] The formula for normalizing depth values ​​is:

[0041] in, This represents the normalized grayscale value, with a range of 0 ≤ ≤255, This represents the depth value of a pixel in the original depth map. This represents the minimum depth value in the original depth map. This represents the maximum depth value in the original depth map.

[0042] Normalized grayscale values ​​are converted into pseudo-color depth maps. Each obtained value is then processed using a color lookup table (LUT). The values ​​are mapped to RGB three-color values ​​to form a pseudo-color map. The LUT included in the COLORMAP_JET function in OpenCV is used to generate the three-color values ​​of the corresponding pigments.

[0043] The pseudo-color depth map is converted into an actual grayscale image. Each RGB pixel value in the pseudo-color map is a unique mapping to a "normalized grayscale value G". Therefore, an "RGB-G" lookup table can be constructed in reverse, and the G value can be recovered through the lookup table to generate the actual grayscale value. First, the OpenCV JET color palette mapping rules are reproduced to generate a forward table of "G→RGB", ensuring consistency in the basis for the reverse lookup. The forward LUT is "G→BGR", and the reverse LUT needs to implement "BGR→G". Essentially, this involves establishing a three-dimensional index (B, G, and R each occupy one dimension, ranging from 0-255), with the index value being the corresponding G. The pseudo-color map is split into three single-channel arrays: B, G, and R. Using the (B, G, R) value of each pixel as the index, the corresponding G is directly read from the reverse LUT. Finally, the actual grayscale image is obtained.

[0044] S4. Extract the reference grayscale value of the ground area from the depth grayscale image, and calculate the correction coefficient based on the actual physical height of the potted plant platform from the ground and the grayscale value of the calibration reference point.

[0045] Bilateral filtering is applied to both the depth grayscale image and the depth mosaic image to eliminate Gaussian and salt-and-pepper noise while preserving edge features. Histogram equalization is then applied to the depth grayscale image to enhance contrast and highlight the grayscale difference between the plants and the ground, facilitating subsequent ground calibration and plant tip feature extraction.

[0046] Next, the threshold size is set. Different thresholds are set according to the different vertical physical distances of the potted plant platform 6 and the depth camera 1 from the ground calibration. In this embodiment, the minimum threshold is set to 0 and the maximum to 75cm. Then, the ground calibration reference is extracted. The extraction principle is to randomly select 3 positions from the potted plant platform 6 area within 1-16 cells and calculate the median value of the grayscale. The program is set to filter out the pure ground area without plants and automatically extract the ground calibration by using the HSL color features of the RGB image and the standard deviation features of the depth map. The preprocessed depth grayscale image and RGB stitched image are divided into 1-16 cells (grids) in 2 rows and 8 columns, corresponding one-to-one with the cell division of the potted plant platform, and the pixel boundary range of each cell is determined. Then, the pure ground area is filtered. Within the potted plant platform 6 area of ​​each cell, multiple candidate areas are randomly selected. Double filtering is performed by using the HSL color features of the RGB image and the standard deviation features of the depth map to remove areas containing soybean plants, leaves, and impurities, and retain the pure ground area containing only the potted plant platform 6. HSL color features are used to identify the baseline color of the ground, while depth map standard deviation features are used to identify areas with uniform grayscale values, ensuring that the selected areas are effective ground references. For each cell's selected pure ground area, the median grayscale value is calculated. Then, the median grayscale values ​​of the ground from the 16 cells are weighted and averaged to obtain the ground reference grayscale value for the entire platform. Meanwhile, three ground calibration benchmarks were randomly selected within each cell, and the grayscale values ​​of each benchmark were calculated to provide basic data for subsequent correction coefficient calculations.

[0047] The essence of the correction factor is to compensate for system errors, minimizing errors caused by factors such as the installation angle of depth camera 1, lens distortion, depth sensor accuracy deviation, and uneven lighting leading to grayscale value shifts in the depth map. This ensures that the plant height calculated from the depth grayscale values ​​matches the actual physical plant height. In this embodiment, the potted plant platform 6 is used as the reference for calculating the correction factor. Given that the actual height of the potted plant platform 6 from the ground is 5cm, three reference points are randomly selected from each image formed by the depth stitching map for ground calibration, and the correction factor is calculated.

[0048] First, calculate the mean and standard deviation of grayscale values ​​at the three calibration reference points to obtain the grayscale stability calibration item:

[0049] Recalculate the initial compensation term for system error:

[0050] Finally, the correction factor is obtained using the following formula:

[0051] in, This is a calibration item for the grayscale stability of the calibration reference point. This is the initial compensation term for system error. For correction factor, For the first The grayscale value of each ground calibration reference point For the first The grayscale value of each ground calibration reference point The actual physical height of the potted plant platform from the ground (in this embodiment) =5cm). This is the theoretical calculation distance, that is, the theoretical calculation distance from the potted plant stand to the ground. .

[0052] when hour, This indicates that no correction is needed.

[0053] when hour, This indicates that the theoretically calculated distance from the potted plant platform 5 to the ground has a huge error, and the system error cannot be compensated for by the correction coefficient.

[0054] Therefore, to avoid inaccurate plant height data due to extreme corrections, this embodiment limits... .

[0055] The program automatically displays the calculated correction coefficients in the Python operation panel and verifies whether the K value is within the valid range of [0.1, 1.0]. If it is outside the range, the ground calibration is performed again.

[0056] S5. Extract the gray value of the top of the plant and the minimum gray value of the image. Combine the reference gray value, the fixed height of the camera and the correction coefficient, and obtain the actual plant height of the soybean plant by mapping the gray value difference ratio.

[0057] The core principle of this embodiment for obtaining plant height based on depth images is to use the potted platform 6 as a reference. By combining the proportional relationship of grayscale differences in the depth images with a correction coefficient, pixel values ​​are converted into the actual height of the plant. The significance of the grayscale values ​​carried in the depth image is that the smaller the grayscale value, the closer the vertical distance from the plant to the depth camera 1, and the higher the physical height of the plant; conversely, the larger the grayscale value, the farther the vertical distance from the plant to the depth camera 1, and the lower the physical height of the plant. In this invention, the position of the depth camera 1 is fixed, and its vertical physical distance from the potted platform 6 is 75cm. Therefore, the threshold for obtaining plant height is specified as ≤75cm.

[0058] After the correction coefficients are calculated, the program automatically displays the RGB mosaic image of the plant population under test. By manually clicking on the highest point (ROI) of each soybean plant in the image, the program automatically determines a 60×60 pixel region of interest (ROItop) centered on the click location, serving as the feature region at the top of the plant. For each plant's ROItop region, the grayscale values ​​are extracted, and abnormal grayscale values ​​within the region are removed using the 3σ principle to obtain the minimum effective grayscale value at the top of the plant. Simultaneously extract the minimum grayscale value from the entire depth grayscale image. This provides parameters for calculating plant height.

[0059] Ground reference gray value Gray value of plant tip Minimum grayscale value of the image Correction coefficient Substituting the vertical distance from the camera into the plant height calculation formula, the program calculates the grayscale difference step by step. , Gray-scale difference normalization ratio term Camera height calibration and grayscale fusion item Finally, the actual plant height of each soybean plant was obtained. Obtain the actual plant height. Once the plant height data for the entire plant population is obtained, the program will automatically generate an Excel spreadsheet of plant height data, a grayscale stability box plot, and a heatmap of the mean plant height. The entire program will then be complete.

[0060] The specific formula for calculating plant height is as follows:

[0061]

[0062]

[0063]

[0064]

[0065] in, This represents the grayscale difference between the ground and the top of the plant. The grayscale difference between the ground and the minimum grayscale value in the image. This is the gray-scale difference normalization ratio term. For camera height calibration and grayscale fusion, For the final plant height, The ground reference gray value, This represents the grayscale value of the topmost part of a single soybean plant. This represents the minimum grayscale value of the entire depth image. For camera height calibration item ( ), The actual vertical physical distance between the camera and the ground reference (in this embodiment) =75 cm). This represents the relative proportion of grayscale difference, that is, the proportion of the grayscale difference between the top of the soybean and the ground to the grayscale difference of the furthest point from the ground. In other words, the pixel-level grayscale difference is converted into a relative height proportion of 0 to 1. Subsequently, through multi-dimensional calibration and fusion terms, it is mapped to the actual physical height and the system error is compensated to finally obtain the actual plant height.

[0066] Optionally, the program set in this embodiment will eventually output: (1) an Excel table of plant height samples of the plants to be tested, used to integrate soybean plant height data. (2) a grayscale stability box plot, used to verify the stability of the ground calibration grayscale values, in order to preliminarily determine the accuracy of the obtained plant height data. (3) a heatmap of the average plant height, used to visually represent the overall plant height fluctuation within the plant height population to be tested.

[0067] Another embodiment of this application discloses a high-throughput soybean plant height phenotypic acquisition system in a plant factory environment, used to perform the above-described method, including: The image acquisition module is used to acquire depth images and RGB images of soybean plants and potted platforms using a depth camera.

[0068] The image stitching module is used to stitch together multiple acquired depth images and RGB images to obtain a depth stitched image and an RGB stitched image.

[0069] The image preprocessing module is used to preprocess the depth mosaic image, converting the original depth image into a pseudo-color depth image and an actual grayscale image in sequence to obtain a depth grayscale image.

[0070] The ground calibration module is used to extract the ground reference grayscale value and the grayscale value of the ground calibration reference point within the potted plant platform area based on the RGB mosaic image and the depth grayscale image.

[0071] The correction factor calculation module is used to calculate the correction factor based on the gray value of the ground calibration benchmark and the actual physical height of the potted plant platform from the ground.

[0072] The plant height extraction and calculation module is used to determine the region of interest at the top of the plant under test on the RGB mosaic image, extract the minimum effective gray value of the corresponding region in the depth grayscale image and the minimum gray value of the entire image, and calculate the plant height of the soybean plant.

[0073] The data output module is used to output the plant height data of all plants to be tested.

[0074] In a specific embodiment, to demonstrate the feasibility of the method and system proposed in this application, the optimal time and accuracy for acquiring plant height data under the plant factory conditions used in this application are given. The results are shown in Table 1. Correlation analysis was performed between the plant height data acquired by machine vision and the control data measured manually with a measuring tape. The coefficient of determination (R²) and root mean square error (RMSE) were calculated to verify the measurement accuracy of the method in this application. Experimental verification showed that when plant height data was collected 15-20 days after soybean emergence, the R² under both normal and low light conditions was ≥0.84. On day 20 after emergence, the R² reached 0.95 / 0.93, and the RMSE was as low as 0.91 / 1.33, indicating that this period is the optimal time for acquiring soybean seedling height data under plant factory conditions. The measurement results of the method proposed in this application are highly consistent with the results of manual measurement, and the accuracy meets the requirements for high-throughput phenotypic identification of soybeans.

[0075] Table 1. Optimal time and accuracy for obtaining plant height data

[0076] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for obtaining the high-throughput plant height phenotype of soybean in a plant factory environment, characterized in that, Includes the following steps: S1. Build a high-throughput phenotypic identification platform to obtain depth images and RGB images of soybean plants and potted platforms; S2. Extract feature points from the RGB image for image registration, and simultaneously stitch the depth image based on the registration parameters to obtain a depth stitched image. S3. The depth mosaic image is sequentially converted into a pseudo-color depth image and an actual grayscale image to obtain a depth grayscale image; S4. Extract the reference gray value of the ground area from the depth gray map, and calculate the correction coefficient based on the actual physical height of the potted plant platform from the ground and the gray value of the calibration reference point. S5. Based on the gray value at the top of the plant and the minimum gray value of the image, combined with the reference gray value, the fixed height of the camera and the correction coefficient, the actual plant height of the soybean plant is obtained by mapping the gray value difference ratio.

2. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 1, characterized in that, S2 includes: The feature point recognition function of RGB images is used to extract the feature points of the overlapping areas of adjacent RGB images. Image registration is achieved through homography matrix, and multiple RGB images are stitched into a complete RGB stitched image. Then, using the registration parameters of the RGB stitched image as a reference, multiple corresponding depth images are simultaneously stitched together to obtain a depth stitched image corresponding to the spatial location.

3. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 2, characterized in that, S3 includes: Original depth map normalization: according to the formula Depth values ​​in the original depth map Normalization is performed to obtain normalized gray values. ;in, The normalized grayscale value. This represents the depth value of a pixel in the original depth map. This represents the minimum depth value in the original depth map. The table shows the maximum depth values ​​in the original depth map; Pseudo-color mapping: The normalized grayscale values ​​are mapped using a color lookup table. Mapped to RGB three-color values, generating a pseudo-color depth map; Grayscale image restoration: Reverse reconstruction of RGB to grayscale values The lookup table is used to restore the pseudo-color depth map to the actual grayscale map, thus obtaining the depth grayscale map.

4. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 3, characterized in that, The formula for calculating the correction factor is as follows: in, This is a calibration item for the grayscale stability of the calibration reference point. This is the initial compensation term for system error. For correction factor, For the first The grayscale value of each ground calibration reference point For the first The grayscale value of each ground calibration reference point This refers to the actual physical height of the potted plant platform from the ground. This is the theoretical calculation distance, that is, the theoretical calculation distance from the potted plant stand to the ground. .

5. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 4, characterized in that, The extraction of the reference grayscale value of the ground area described in S4 includes: The depth grayscale image is divided into multiple sub-areas according to preset rules. Within the potted plant platform area of ​​each sub-area, the HSL color features of the RGB image and the standard deviation features of the depth image are used for dual screening to remove areas containing plants and retain pure ground areas. Calculate the median grayscale value of the pure ground area in each cell, and then take a weighted average of the median ground grayscale values ​​of all cells to obtain the baseline grayscale value for the entire platform. .

6. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 5, characterized in that, The formula for calculating the actual plant height of soybean plants is as follows: in, This represents the grayscale difference between the ground and the top of the plant. The grayscale difference between the ground and the minimum grayscale value in the image. This is the grayscale difference normalization ratio term. For camera height calibration and grayscale fusion, For the final plant height, The ground reference gray value, This represents the grayscale value of the topmost part of a single soybean plant. This represents the minimum grayscale value of the entire depth image. Indicates the relative proportion of grayscale difference. For camera height calibration, This represents the actual vertical physical distance between the camera and the ground reference.

7. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 6, characterized in that, The grayscale value of the topmost part of the soybean plant was obtained through the following steps: Manually click on the growth point of the target plant in the RGB mosaic image. Determine the region of interest centered on the click location. Extract the grayscale values ​​of the depth grayscale image within this region. After removing outliers using the 3σ principle, take the minimum grayscale value as the base value. .

8. The method for obtaining high-throughput soybean plant height phenotype in a plant factory environment according to claim 7, characterized in that, It also includes the following steps: Generate a table of soybean plant height data, a grayscale stability box plot to verify the stability of ground calibration grayscale values, and a heatmap of mean plant height to represent the fluctuation of plant height within the population.

9. A high-throughput soybean plant height phenotypic acquisition system for a plant factory environment, characterized in that, For performing the method according to any one of claims 1-8, comprising: The image acquisition module is used to acquire depth images and RGB images containing soybean plants and potted platforms using a depth camera; The image stitching module is used to stitch together multiple acquired depth images and RGB images to obtain a depth stitched image and an RGB stitched image; The image preprocessing module is used to preprocess the depth mosaic image, converting the original depth image into a pseudo-color depth image and an actual grayscale image in sequence to obtain a depth grayscale image. The ground calibration module is used to extract the ground reference gray value and the gray value of the ground calibration reference point in the potted plant platform area based on the RGB mosaic image and the depth grayscale image. The correction factor calculation module is used to calculate the correction factor based on the gray value of the ground calibration reference point and the actual physical height of the potted plant platform from the ground. The plant height extraction and calculation module is used to determine the region of interest at the top of the plant under test on the RGB mosaic image, extract the minimum effective gray value of the corresponding region in the depth grayscale image and the minimum gray value of the whole image, and calculate the plant height of soybean plants. The data output module is used to output the plant height data of all plants to be tested.

Images