In-situ crop root system monitoring device and root system phenotype parameter analysis system

Abstract

An in-situ crop root monitoring device and root phenotypic parameter analysis system are disclosed. Microroot tubes and microroot hair observation tubes are pre-installed at an angle in the cultivation substrate. During data collection, a slide is inserted into the microroot tube. An image acquisition module and a supplementary lighting module are fixed to the output of a rotary stepper motor via a connecting and fixing module. The slide is engaged with a lead screw using a three-hole flange-type round nut. The microroot hair observation tube is equipped with a transparent observation plate and a limiting part for attaching a high-definition camera. The control unit controls the rotary stepper motor to complete one-cycle image acquisition at each axial position, then drives the slide to the next position until the observation range is covered. The microroot hair observation tube simultaneously acquires root hair images. The images are saved sequentially and transmitted to a PC. The PC automatically extracts the total root length, total root area, and stratification parameters after stitching and segmenting the sequential images. It also separates the root axis and root hairs to calculate parameters such as root hair length, area, and quantity, achieving automated analysis from image acquisition to phenotypic parameter output.

Description

Technical Field

[0001] This invention belongs to the field of crop phenotypic monitoring and agricultural information technology, specifically involving the cross-application of crop root in-situ image acquisition technology, electromechanical automation control technology and computer vision image processing technology. In particular, it relates to a system that uses pre-embedded micro-root tubes and micro-root hair observation tubes to collaboratively acquire root and root hair images, and uses image stitching algorithms, deep learning segmentation models and automatic phenotypic parameter calculation models to achieve high-throughput in-situ monitoring of root phenotypic characteristics. Background Technology

[0002] Roots, as vital organs for crops to absorb water and mineral nutrients, anchor the plant, and facilitate rhizosphere material exchange, directly influence the efficiency of crop utilization of soil resources through their morphology, structure, and spatial distribution. The overall spatial configuration, total root length, root area, and distribution ratio at different soil depths largely reflect a crop's potential for acquiring soil water and nutrients. Root hairs, as a key component of the root absorption interface, directly affect the contact efficiency between roots and soil particles and the rate of nutrient absorption through their length, density, and area. Therefore, continuous in-situ observation of the root system and root hairs of the same plant, without disrupting normal plant growth or the original soil structure, to obtain visualized images and further quantify relevant phenotypic parameters, has significant theoretical and practical value for optimizing crop cultivation management, selecting drought-resistant and infertile-tolerant varieties, studying root growth and development patterns, and formulating precise water and fertilizer control strategies.

[0003] Currently, commonly used methods for root research include excavation, soil drilling, root washing and scanning analysis, transparent root box observation, and microroot canal imaging. Excavation and soil drilling typically require removing the roots from the soil, inevitably disrupting their original spatial distribution and structural characteristics, and making repeated observation of the same plant impossible. While root washing and scanning can obtain relatively clear images of the root system and measure parameters such as root length and diameter, it also suffers from the fundamental drawback of destructive sampling, making it difficult to reflect the true growth dynamics of roots in their natural soil environment. The transparent root box method involves planting crops in transparent containers filled with soil, observing the root growth and development process through the container walls. However, due to the boundary effects of the container space and differences in soil compaction, aeration, and natural conditions, the observation results often deviate from actual field conditions.

[0004] Microroot canal imaging involves embedding a transparent tube in the soil of the target root zone and inserting an image acquisition device inside the tube to photograph the roots attached to the outer wall of the tube. This allows for long-term, continuous, and repeated observation of the same root system without damaging the soil structure or the plant, making it a widely recognized and reliable in-situ root monitoring method. However, current conventional microroot canal observation devices are mainly used to acquire images of root distribution over a large scale, typically focusing on the taproot, lateral roots, and the overall spatial configuration of the root system. Limited by camera resolution and depth of field, it is difficult to clearly image the microscopic structure of root hairs, which are precisely the parts of the root system in closest contact with the soil nutrient interface, making it challenging to obtain their growth and development information in situ. Furthermore, existing microroot canal devices generally suffer from several technical shortcomings during image acquisition. For example, the stability of the acquisition component's rotation and axial movement within the tube is poor, the overlap ratio between adjacent acquisition positions is difficult to control, and improper lighting can easily lead to localized overexposure or reflection, affecting image quality and subsequent stitching results. Although some imported microroot canal monitoring devices on the market can achieve relatively comprehensive root image acquisition and preliminary analysis, these devices are expensive and can usually only be configured with a small number of monitoring points in practical applications. This makes it difficult to meet the breeding and cultivation research needs of simultaneous monitoring of multiple plants under different genotypes and treatment conditions, which greatly restricts the throughput and efficiency of root phenotypic data acquisition.

[0005] In response to the above situation, it is necessary to design an in-situ monitoring scheme that combines the ability to observe the root distribution on a large scale with the ability to image the root hairs at a high level. This scheme should be equipped with a corresponding automatic image analysis process that integrates the stitching of local root images, target segmentation, calculation of overall root parameters, and automatic extraction of root hair parameters into the same system. This would reduce the workload of manual processing, improve the efficiency and standardization of data processing, and thus better serve crop root-related research. Summary of the Invention

[0006] Technical problems solved: To address the challenges of existing microroot canal monitoring devices in simultaneously acquiring microscopic images of the overall root system distribution and corresponding root hair depths, unstable rotation and positioning during acquisition, and overexposure during supplemental lighting leading to stitching difficulties, as well as the high cost of imported equipment making batch monitoring difficult, and the reliance on manual processing for image analysis and phenotypic parameter extraction, this invention provides an in-situ crop root system monitoring device and root phenotypic parameter analysis system. Through the collaborative acquisition of pre-embedded microroot canals and microroot hair observation tubes, electromechanical linkage control, and automatic image stitching, segmentation, and parameter calculation, it can achieve high-throughput automatic analysis of the entire process from root and root hair image acquisition to phenotypic parameter output.

[0007] Technical Solution: An in-situ crop root monitoring device includes micro-root canals, micro-root hair observation tubes, a sliding table, and a control unit. The micro-root canals and micro-root hair observation tubes are pre-installed at a predetermined angle to the horizontal plane in the cultivation substrate before crop planting. The sliding table is detachably installed inside the micro-root canals. An image acquisition module, a supplementary lighting module, and a rotary stepper motor are mounted on the sliding table. The image acquisition module and the supplementary lighting module are fixedly connected to the output end of the rotary stepper motor via a connecting and fixing module. The rotary stepper motor drives the image acquisition module and the supplementary lighting module to rotate one revolution at each axial position to acquire a circumferential sequence of images. The sliding table is threadedly engaged with the lead screw of the transmission module via a three-hole flange-type round nut. The micro-root hair observation tube has a transparent observation plate and a limiting part inside for mounting a high-definition camera. The limiting part maintains a fixed distance between the lens of the high-definition camera and the transparent observation plate. The control unit is electrically connected to the transmission module, the rotary stepper motor, the image acquisition module, the supplementary lighting module, and the high-definition camera.

[0008] A rolling support assembly is provided on the outer periphery of the slide, and the rolling support assembly makes rolling contact with the inner wall of the micro-root canal.

[0009] The image acquisition module described above includes at least one camera, which is connected to the rotary stepper motor via the connection and fixing module.

[0010] The aforementioned supplementary lighting module includes four LED light-emitting elements, which are symmetrically distributed on both sides of the image acquisition module. The LED light-emitting elements are arranged in a backlighting manner.

[0011] The image acquisition module and the supplementary lighting module are fixed to the connecting and fixing module by screws and nuts, and the connecting and fixing module is fixedly connected to the output end of the rotary stepper motor.

[0012] The aforementioned rotary stepper motor is fixedly mounted on the slide, and the three-hole flange-type round nut is fixed on the slide. The three-hole flange-type round nut is threadedly engaged with the lead screw of the transmission module.

[0013] The aforementioned transmission module includes a lead screw and a lead screw stepper motor for driving the lead screw to rotate, and the lead screw stepper motor is electrically connected to the control unit.

[0014] A root phenotypic parameter analysis system is applied to a PC terminal that is communicatively connected to the control unit of the in-situ crop root monitoring device. The system comprises: an image receiving module for receiving root image information transmitted by the control unit and imported root images stored locally; a root image stitching module for stitching together sequentially arranged local root images obtained from micro-root canals to obtain a holistic root image; a model segmentation module for segmenting the holistic root image into root targets to generate corresponding binary root images; a root stitched image parameter extraction module for extracting the skeleton and performing depth layering processing on the binary root image to obtain at least one holistic root parameter among total root length, total root area, layered root length, and layered root area; a root hair image parameter extraction module for separating root axes and root hairs, extracting skeletal lines, and calculating root hair parameters on the binary root hair image to obtain at least one root hair parameter among root axis length, root hair area, root hair length, and root hair number; and a result output module for exporting and saving the holistic root parameters, the root hair parameters, and the images processed during the process.

[0015] The above-mentioned model segmentation module includes a sliding window segmentation unit and a prediction fusion unit. The sliding window segmentation unit is used to divide the original root system image into multiple overlapping image segments. The prediction fusion unit is used to fuse the prediction results of each image segment in the overlapping region to obtain a prediction probability map corresponding to the size of the original root system image.

[0016] An electronic device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program to implement the functions of the root phenotypic parameter analysis system described above.

[0017] Beneficial Effects: In this invention, the microroot canals and microroot hair observation tubes are pre-embedded. The acquisition components are then installed into the tubes when data collection is needed. Compared to traditional post-installation methods, this design significantly reduces disturbance to the crop root zone and soil structure during installation, allowing for long-term, continuous, and repeated observation of the same plant's root system without damage. The microroot canal acquisition component and the microroot hair observation tube work together. The former acquires overall distribution images of the main and lateral roots over a large scale, while the latter provides high-definition imaging of root hairs attached to the transparent observation plate. Their collaborative operation enables this invention to simultaneously acquire both macroscopic root system spatial configuration and microscopic root hair developmental status information within the same system, overcoming the limitation of conventional microroot canal devices in achieving clear root hair imaging. The rolling support component on the outer periphery of the slide significantly reduces the resistance to movement of the acquisition module within the tube, mitigating tube wall scratches and reducing image jitter during movement. Simultaneously, the connecting and fixing module stably connects the camera and supplementary lighting element to the rotary motor, effectively preventing the offset of adjacent image overlap caused by relative sliding during rotation, thus ensuring quality for subsequent stitching. The supplementary lighting module employs backlighting or indirect illumination, avoiding the light spots and overexposed areas caused by traditional forward lighting. This improves the brightness uniformity and stitching fusion effect between sequential images. The miniature root hair observation tube uses a limiting structure to maintain a fixed distance between the high-definition camera and the transparent observation plate, ensuring consistency in scale and focal length for root hair images acquired at different tube positions and times. This makes subsequent comparisons of root hair parameters highly comparable. The entire system forms a complete automated chain from acquisition to analysis. During acquisition, the control unit triggers images in an orderly manner according to the set axial step and rotation angle and saves them by sequence number. After receiving the data on the PC, image stitching, model segmentation, overall root system parameter calculation, and root hair parameter calculation are completed within the same interface, significantly reducing manual intervention. Furthermore, one acquisition component and control box can be used with multiple pre-embedded tubes, effectively reducing the unit cost of batch monitoring and providing a feasible technical path for high-throughput in-situ root phenotypic monitoring. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall structure of an in-situ crop root monitoring device and root phenotypic parameter analysis system of the present invention. The left side shows the control box, the middle shows the transmission, sliding table, image acquisition and supplementary lighting structure installed in the micro root canal, and the right side shows the micro root hair observation tube and its high-definition camera and magnetic supplementary lighting structure.

[0019] The following are explanations of the reference numerals in the diagram: 1. Micro-root canal; 2. Transmission module; 31. Camera; 32. Illumination module; 33. Connection and fixing module; 4. Slide table; 41. Rolling support assembly; 42. Three-hole flange round nut; 51. Control box; 52. Power supply module; 53. Stepper motor drive module; 54. Control unit; 55. Storage and communication module; 6. Miniature root hair observation tube; 61. High-definition camera; 62. Transparent observation plate; 63. Magnetic illumination element.

[0020] Figure 2 This is a schematic diagram of the microroot canal section and the root hair observation tube section in an in-situ crop root monitoring device. A represents the microroot canal section; B represents the root hair observation tube section.

[0021] Figure 3 This is a schematic diagram of the root image analysis system part of the in-situ crop root monitoring device and root phenotypic parameter analysis system.

[0022] Figure 4 This is a flowchart illustrating the operation of the in-situ crop root monitoring device and the acquisition of root parameters.

[0023] Figure 5 The process for segmenting root system splicing diagrams and extracting root system skeleton diagrams and layered diagrams.

[0024] Figure 6 This document outlines the process for root hair image segmentation and the separation of root hairs, root axes, and their skeletons. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment uses a tomato pot experiment conducted under laboratory conditions as an example. The in-situ crop root monitoring device and root phenotypic parameter analysis system provided by the present invention are used to acquire images and analyze phenotypic parameters of tomato roots and root hairs under different root zone oxygen content conditions. It should be understood that this embodiment uses tomato as the specific experimental object and is only used to illustrate the implementation process of the device and parameter measurement system of the present invention, and not to limit the application of the present invention to tomatoes. Unless otherwise specified, the device dimensions, materials, acquisition parameters, algorithm thresholds, and software parameters can be adjusted according to the tomato variety, growth stage, pot container size, cultivation substrate, observation tube size, and image resolution.

[0026] In this embodiment, tomatoes were grown in laboratory pots, and three treatment groups were set up according to the oxygen content in the root zone: an oxygen-enhancing group, a control group, and a hypoxia group. The oxygen-enhancing group increased the oxygen content in the tomato root zone according to the experimental protocol, the control group maintained conventional cultivation conditions, and the hypoxia group decreased the oxygen content in the tomato root zone according to the experimental protocol. All treatment groups used the same tomato planting conditions and cultivation management measures, and all experimental conditions were kept consistent except for the root zone oxygen content treatment. After one month of continuous treatment with the three oxygen treatments, root system images and root hair images of tomatoes in each treatment group were collected using the device described in this invention. The collected images were then processed and phenotypic parameters were calculated using a PC-based root phenotypic parameter analysis system.

[0027] Example 1: In-situ root system and root hair monitoring device and image acquisition

[0028] like Figure 1 As shown, the tomato in-situ root monitoring device in this embodiment includes a micro root canal 1, a transmission module 2, a camera 31 (USB industrial driverless camera), a supplementary lighting module 32, a connection and fixing module 33, a slide table 4, a rolling support assembly 41 (BCHA non-standard stainless steel bearing sliding wheel transport and transmission parts), a three-hole flange type round nut 42, a control box 51, and a micro root hair observation tube 6. The control box 51 contains a power supply module 52 (5V 6000mAh mobile power lithium battery module (18650 lithium battery pack with 5V boost and voltage regulation output module); 12V power adapter), a stepper motor drive module 53 (digital HPD322 driver), a control unit 54 (Raspberry Pi 4 Model B), and a storage and communication module 55 (Micro SD card (32GB) + Raspberry Pi 4B built-in WiFi wireless module); a miniature root hair observation tube 6 is equipped with a high-definition camera 61 (4K 12MP IMX258 high-definition miniature camera), a transparent observation plate 62, and a magnetic supplementary lighting element 63 (LED magnetic light).

[0029] Before conducting the tomato pot experiment, microroot canals 1 and microroot hair observation tubes 6 were pre-embedded in the cultivation substrate of each tomato planting pot. Both microroot canals 1 and microroot hair observation tubes 6 were embedded at a 45° angle to the horizontal plane, allowing the tomato roots to gradually approach and adhere to the outer wall of the microroot canals 1 and the surface of the transparent observation plate 62 of the microroot hair observation tubes 6 during growth. During this process, the microroot hair observation tubes 6 in different pots were kept at the same depth, ensuring that the acquired root hair images were at the same depth level. After pre-embedding, the cultivation substrate was filled to the predetermined depth before tomato planting, thus avoiding root damage and disturbance to the cultivation substrate structure caused by inserting the observation tubes after the tomato roots had formed.

[0030] The micro-root cannula 1 is a transparent tube made of acrylic material, with a removable end cap at the upper end and a closed end with a base at the lower end. The transmission module 2 is arranged along the axial direction of the micro-root cannula 1 and includes a lead screw and a lead screw stepper motor (20T type lead screw stepper motor) that drives the lead screw to rotate. The slide 4 is threadedly engaged with the lead screw via a three-hole flange-type round nut 42 fixed on it. When the lead screw stepper motor drives the lead screw to rotate, the slide 4 moves along the axial direction of the micro-root cannula 1.

[0031] The slide 4 is equipped with six omnidirectional wheels or other rolling support components 41 on its outer periphery. The rolling support components 41 roll in contact with the inner wall of the micro-root canal 1, keeping the slide 4 relatively centered during movement. The rolling contact method reduces friction between the slide 4 and the inner wall of the micro-root canal 1, reduces movement resistance, avoids scratches on the inner wall of the micro-root canal 1 caused by direct sliding of the slide 4, and reduces the shaking of the camera 31 during movement and shooting.

[0032] The camera 31 and the supplementary lighting module 32 are mounted on the slide table 4 via a connecting and fixing module 33, and are connected to a rotary stepper motor (20HS24 stepper motor) mounted on the slide table 4. The connecting and fixing module 33 is used to keep the relative positions of the camera 31, the supplementary lighting module 32 and the output end of the rotary stepper motor fixed, preventing the camera 31 from sliding between the camera and the rotary stepper motor during rotation, thereby ensuring that the rotation angle and overlapping area of ​​the images acquired at different circumferential positions are relatively stable.

[0033] The camera 31 is a USB strip camera capable of close-up imaging. The supplementary lighting module 32 includes four LED light-emitting elements, which are symmetrically arranged on both sides of the camera 31 and soldered onto the PCB board. The LED light-emitting elements adopt a back-facing indirect lighting method, so that the emitted light is reflected and illuminates the tomato root observation area, reducing the strong reflection, local bright spots and image overexposure caused by direct light illuminating the wall of the micro-root canal 1.

[0034] The power supply module 52 inside the control box 51 supplies power to the transmission module 2, camera 31, supplementary lighting module 32, rotary stepper motor, and control circuit. The stepper motor drive module 53 drives the lead screw stepper motor in the transmission module 2 and the rotary stepper motor on the slide table 4. The control unit 54 generates motor control signals, controls the camera 31 to acquire images, and numbers, stores, and transmits the acquired images. The storage and communication module 55 saves the acquired images of the tomato's local root system and root hairs, and transmits the image data to the PC.

[0035] The tomato pot experiment included three treatment groups: an oxygen-enriched group, a control group, and a hypoxia group. The oxygen content in the tomato root zone was regulated according to a predetermined protocol in each treatment group. After one month of treatment, images of the tomato roots and root hairs in each of the three treatment groups were acquired. Before acquisition, it was confirmed that the camera 31, supplemental lighting module 32, slide table 4, transmission module 2, and rotary stepper motor were functioning correctly. The acquisition components were then installed into the corresponding microroot canals 1 and microroot hair observation tubes of the tomato pots under test.

[0036] When acquiring images of the tomato root system, the control unit 54 first controls the transmission module 2 to move the slide 4 to a preset starting position. At the current axial position, the control unit 54 controls the supplementary lighting module 32 to turn on and controls the rotary stepper motor to rotate at preset angle intervals. Each time the rotary stepper motor rotates through an acquisition angle, the control unit 54 triggers the camera 31 to acquire a partial image of the tomato root system.

[0037] After completing the circumferential image acquisition within a circumferential range at the current position, the control unit 54 controls the lead screw stepper motor to drive the slide 4 to move one preset step distance along the axis of the microroot canal 1. After the slide 4 reaches the next axial position, it completes another circumferential image acquisition. This process is repeated sequentially until the camera 31 has completed the acquisition of all tomato root images within the preset observation range of the microroot canal 1. A certain range of image overlap is maintained between adjacent circumferential acquisition positions and adjacent axial acquisition positions to facilitate subsequent tomato root image stitching.

[0038] Each tomato root system image is named or written into image metadata according to at least one of the following: treatment group, tomato pot number, collection time, axial position sequence number, and circumferential angle sequence number. The image name includes identifiers for the oxygenated group, control group, and hypoxic group, ensuring accurate correspondence between tomato root system images collected under different oxygen treatment conditions. The control unit 54 saves the tomato root system images arranged in the collection order to the storage and communication module 55 and transmits them to the PC via a wired or wireless communication interface.

[0039] The miniature root hair observation tube 6 includes a 3D-printed tubular body, a transparent observation plate 62, a high-definition camera 61, and a magnetic light-adhesive element 63. The transparent observation plate 62 is disposed in the observation area of ​​the miniature root hair observation tube 6 to form a transparent observation surface. When the tomato roots grow to the vicinity of the transparent observation plate 62, some tomato root axes and root hairs can approach or attach to the outside of the transparent observation plate 62, thereby allowing image acquisition from inside the miniature root hair observation tube 6.

[0040] When collecting images of tomato root hairs, a high-definition camera 61 is inserted into a miniature root hair observation tube 6, and the high-definition camera 61 is engaged with the limiting part inside the tube. The limiting part maintains a fixed distance between the lens of the high-definition camera 61 and the transparent observation plate 62, thereby ensuring that the root hair images obtained by the oxygenated group, the control group, and the hypoxic group in different observation tubes and at different collection times have basically the same imaging distance, magnification, and focal plane.

[0041] A magnetic illumination element 63 is installed around the observation end of the miniature root hair observation tube 6 to provide uniform illumination for the tomato rootstock and root hairs near the transparent observation plate 62. After adjusting the high-definition camera 61 to a clear image, root hair images of tomatoes in each treatment group are acquired sequentially. The acquired root hair images are numbered according to the treatment group, pot number, and acquisition time, and transmitted to the control unit 54 or directly to the PC.

[0042] Example 2: Image stitching, target segmentation, and result output on PC

[0043] In this embodiment, the PC-based root phenotypic parameter analysis system is implemented using Python, and the graphical user interface is built using PyQt6. The upper left corner of the interface houses the tomato root mosaic image analysis module, the lower left corner houses the tomato root hair image analysis module, and the right side displays a shared result image area and parameter result table for both analysis modules.

[0044] After one month of treatment, local root images and root hair images of tomatoes collected from the oxygenated group, control group, and hypoxic group were transmitted to the PC. The image receiving module can receive local root images transmitted by the control unit 54, and can also import overall root mosaic images, root binary images, original images of tomato root hairs, or root hair binary images from local folders.

[0045] For the tomato root system images acquired by the micro-root canal 1, the system first sorts them according to the image file name, axial position number, and circumferential angle number, so that the images acquired at the same axial position are arranged in the circumferential acquisition order, and the images at different axial positions are arranged in order according to the observation depth.

[0046] Because the reflective conditions, tomato root distribution, and local brightness may vary at different locations within the microroot canal 1, the system performs brightness uniformity processing on adjacent local root images before stitching. Exposure correction employs a combination of grayscale mean correction, gamma correction, and linear compensation for overlapping areas to reduce the brightness differences between adjacent tomato root images.

[0047] After exposure correction is completed, the system determines the positional relationship between adjacent images based on the preset acquisition step size and phase-related feature point matching results of adjacent images, and processes the overlapping areas of adjacent images using a linear weighting method.

[0048] Let the width of a single local root system image be W, and the overlap ratio of adjacent images be ρ. Then the stitching step size of adjacent images is:

[0049] W×(1-ρ)

[0050] Specifically, ten identical images of the tomato circumferential root system were acquired at the same axial position, with an overlap ratio ρ1 of 0.18. The system first unfolded and stitched the ten images sequentially according to their circumferential angles into a single circumferential unfolded image. Then, the circumferential unfolded images at different axial positions were stitched vertically according to their depth, with an overlap ratio ρ2 of 0.2. This yielded complete root system mosaic images for each tomato potted plant in the aeration group, control group, and low-oxygen group, respectively.

[0051] For the stitched tomato color root image and the original tomato root hair image acquired by the micro root hair observation tube 6, the system loads a pre-trained root image segmentation model for target segmentation. Specifically, the segmentation model uses the U^2-Net variant. The segmentation module divides the large image into multiple overlapping image blocks using a sliding window method, and inputs each block into the segmentation model for prediction.

[0052] Specifically, the image patch size is 480×480 pixels, and the sliding window step size is 240 pixels. After each image patch obtains its corresponding prediction probability map, the prediction fusion unit accumulates the prediction results into a probability matrix of the same size as the original image, according to the position of the image patch in the original image. Then, it averages the predictions based on the number of times each pixel position is covered by the image patch to obtain a fused probability map of the same size as the original image.

[0053] P(x,y) = Σk Pk(x,y) / max[Σk Wk(x,y), ε]

[0054] Where P(x,y) is the predicted probability of the fused pixel position (x,y), Pk(x,y) is the predicted probability of the k-th image block at the corresponding position, Wk(x,y) represents the coverage of the k-th image block at that position, and ε is a minimum value set to prevent the denominator from being zero.

[0055] The fused probability map is compared with a preset segmentation threshold T to generate a binary image with tomato roots or root hairs as the white foreground and a black background. Specifically, the segmentation threshold T is set to 0.5.

[0056] The model segmentation task is executed in a background thread and outputs the model loading status and image processing progress to the graphical user interface to keep the interface responsive.

[0057] The result image display area is used to display the original image, stitched image, segmented image, preprocessed image, skeleton image, layered image, root axis image, root hair image, and root axis / root hair comparison image of tomato roots or root hairs. It also supports zooming in and out of the image by holding down the Ctrl key and using the mouse wheel.

[0058] After the analysis is completed, the system creates independent result folders according to the oxygen treatment group, tomato pot number, source image name, analysis module type, and analysis time. The system also saves the process images and parameter results in Excel, CSV, or text format to the corresponding folders for subsequent processing and comparison of tomato root system and root hair phenotypic parameters in the oxygenated group, control group, and hypoxic group.

[0059] Example 3: Measurement of root system and root hair phenotypic parameters

[0060] This embodiment examines the impact of changes in root zone oxygen content on tomato root growth, comparing root images from the oxygenated group, control group, and hypoxic group one month after treatment. Total root length and total root area were selected to characterize the overall growth scale of the tomato root system. Root length, root area, and their proportion at depths of 0–10 cm, 10–20 cm, and 20–30 cm were selected to characterize the vertical distribution of tomato roots in the potting substrate, thus reflecting the influence of different oxygen treatments on root growth and spatial distribution.

[0061] For tomato root hairs, based on previous observations of root hair images, root axis length was selected as the baseline reference for root hair occurrence area. Total root hair length, total root hair area, and root hair number were selected to characterize the elongation, overall growth scale, and number of root hairs, respectively. These phenotypic parameters can comprehensively evaluate the differences in tomato root and root hair growth under different oxygen content conditions in root zones, considering overall root growth, vertical distribution, and root hair occurrence and elongation.

[0062] When measuring the overall root system parameters of a tomato, the system reads the segmented binary image of the tomato root system, representing the tomato root region as white foreground pixels and the background as black background pixels. The system first removes noisy connected regions with an area smaller than a preset threshold of 10 pixels, and then performs skeletonization processing on the denoised tomato root region to obtain a tomato root skeleton with a width of one pixel.

[0063] Let the actual pixel length of the image be s, and the number of adjacent pixel connections in the horizontal, vertical, and two diagonal directions of the tomato root system skeleton be Nh, Nv, Nd1, and Nd2, respectively. Then the total root length L of the tomato is calculated according to the following formula:

[0064] L = (Nh + Nv)·s + (Nd1 + Nd2)· ·s

[0065] Let Na be the number of foreground pixels of the tomato root system. Then the total root area A of the tomato is calculated according to the following formula:

[0066] A = Na·s²

[0067] When performing depth stratification, microroot canals 1 are embedded at a 45° angle to the horizontal plane, and the longitudinal direction of the tomato root system mosaic corresponds to the axial observation direction of microroot canal 1. In this embodiment, the total vertical observation depth of the soil is 30 cm, divided into three depth layers every 10 cm: 0–10 cm, 10–20 cm, and 20–30 cm. Let the vertical depth be d, and its corresponding observation distance along the microroot canal axis be l, then:

[0068] l = d / sin45° = ·d

[0069] Let the height of the complete tomato root mosaic be H pixels. Then, the three depth layers of 0–10 cm, 10–20 cm, and 20–30 cm correspond to pixel height ranges of 0–H / 3, H / 3–2H / 3, and 2H / 3–H, respectively. The tomato root length of each depth layer is obtained by counting the number of horizontal, vertical, and diagonal connections of the root skeleton within the corresponding pixel height range, and then multiplying the result by the actual pixel length after weighted summation according to the corresponding connection lengths. The tomato root area of ​​each depth layer is obtained by counting the number of foreground pixels within the corresponding pixel height range and multiplying the result by the actual area of ​​a single pixel.

[0070] Let Li be the root length and Ai be the root area of ​​the tomato plant in the i-th layer, then the proportions of root length and root area in this layer are respectively:

[0071] Li / L×100%

[0072] and

[0073] Ai / A×100%

[0074] Specifically, the actual pixel length of the tomato root system mosaic was obtained through calibration using a microroot canal device. During calibration, a 2.5 mm grid calibration paper was used, and the paper was placed tightly against the wall of the microroot canal 1. The calibration image was then captured by camera 31. Two points with an actual spacing of 10 mm were selected in the calibration image, and the pixel distance between the two points was measured to be 275 pixels. Based on this, the actual length of a single pixel was approximately 0.0363 mm / pixel.

[0075] Based on the above calibration and depth stratification settings, the system outputs the image width, image height, pixel size, total observation depth, total root length, total root area, and root length, root area, and corresponding proportion of each depth layer for each tomato potted plant in the oxygenation group, control group, and low oxygen group.

[0076] When measuring tomato root hair parameters, the system first removes noisy connected regions with an area less than 10 pixels from the binary image of the tomato root hair and fills in holes with an area less than 10 pixels to obtain a foreground region R composed of the tomato root axis and root hairs. A Euclidean distance transformation is then performed on the foreground region R, and the core region C of the tomato root axis with a relatively large local width is identified based on the distance from the foreground pixels to the background region.

[0077] The core region C of the tomato root axis is morphologically expanded and intersected with the original foreground region R to obtain the coarse root axis region A0. The coarse root axis region A0 is then subtracted from the original foreground region R to obtain the tomato root hair region H for subsequent root hair parameter measurements. The root hair region H excludes interference from the coarse root axis region and the root hair bases and local impurities contained therein.

[0078] C = {p∈R | D(p) ≥ r0}

[0079] A0 = dilate(C, rd) ∩ R

[0080] H = R \ A0

[0081] Where D(p) represents the Euclidean distance from the foreground pixel p to the nearest background pixel, r0 is the local radius threshold of the root axis, which is set to 4.0 pixels in this embodiment; rd is the root axis dilation radius, which is set to 4 pixels in this embodiment.

[0082] To improve the root axis boundary of tomatoes, the system extracts the center position and half-width of the thick root axis region row by row, interpolates missing rows, and smooths the root axis center sequence and half-width sequence respectively. Then, the smoothed half-width is multiplied by a root axis width contraction coefficient of 0.90 to eliminate interference from root hair matrix and impurities, and the smoothed root axis region A is reconstructed.

[0083] The smooth root axis region A is skeletonized and an eight-neighborhood skeleton graph is constructed. The path with the largest weighted distance between the skeleton endpoints is selected as the main skeleton line of the tomato root axis. The actual length of the root axis is obtained by multiplying the weighted pixel length of the main skeleton line by the actual pixel length of the root hair image.

[0084] The total area of ​​tomato root hairs is calculated by multiplying the number of foreground pixels in the root hair region H by the actual area of ​​each pixel. The total length of tomato root hairs is obtained by skeletonizing the root hair region H and accumulating the skeleton connection lengths in the horizontal, vertical, and diagonal directions.

[0085] The number of tomato root hairs was estimated using a combination of connected regions of the root hair skeleton and endpoint positions. First, connected regions of the root hair skeleton were marked, and those with a skeleton length less than 8 pixels were removed. Then, the smooth root axis region A was expanded by a root axis nearest neighbor radius of 12 pixels to form root axis nearest neighbor regions. For each retained connected region of the root hair skeleton, the number of root hairs corresponding to that region was determined based on whether its skeleton endpoints were located in root axis nearest neighbor regions and the number of endpoints on the side furthest from the root axis. Finally, these values ​​were summed to obtain the total number of tomato root hairs.

[0086] In one specific implementation, the actual pixel length of the tomato root hair image is obtained through calibration using a miniature root hair observation tube 6. During calibration, a 2.5 mm grid calibration paper is used, and it is placed tightly against the transparent observation plate 62 of the observation area of ​​the miniature root hair observation tube 6. A high-definition camera 61 is installed in a preset fixed position to acquire the calibration image. Two points with an actual spacing of 10 mm are selected in the calibration image, and the pixel distance between the two points is measured to be 746 pixels. Based on this, the actual length corresponding to a single pixel is approximately 0.0134 mm / pixel.

[0087] The system outputs the root axis length, total root hair area, total root hair length, and root hair number of each tomato potted plant in the oxygenated group, control group, and low-oxygen group. Based on the tomato root system and root hair phenotypic parameters obtained from each treatment group, the changes in total root length, root area, vertical root distribution, root axis length, root hair length, root hair area, and root hair number of tomatoes under different root zone oxygen content conditions can be further compared.

[0088] The above three embodiments constitute a complete implementation process in a tomato pot experiment, from the pre-embedding of observation tubes, oxygen treatment, acquisition of tomato root and root hair images, image transmission, PC-side stitching and target segmentation, to the output of tomato root and root hair phenotypic parameters. Those skilled in the art, without departing from the concept of this invention, can apply equivalent substitutions to the tomato variety, potting container, oxygen control method, device structure, number of micro-root tubes and micro-root hair observation tubes, driving method, segmentation model, parameter thresholds, or software implementation method to this invention.

[0089] Comparative Example 1: Fisheye Lens Hemispherical Imaging Root System Image Acquisition

[0090] This comparative example uses the same tomato pot experiment and oxygenation group, control group, and hypoxia group treatment settings as Example 1. The difference is that the tomato root system image acquisition method adopts the method described in the prior art (CN118018852A): a camera with a fisheye lens is set in a pre-embedded pipe. The camera moves at a fixed distance in the pipe and acquires a hemispherical image covering a circumference of the pipe wall at each axial position. Then, a 360° panoramic image of the root system is obtained by image unfolding, phase correlation, RANSAC abnormal displacement removal, and multi-level stitching and fusion. The supplementary lighting uses ring LED strong light illumination, and light-absorbing cloth is used to reduce the interference of pipe wall reflection.

[0091] Comparing the panoramic image of tomato roots obtained in Comparative Example 1 with the stitched image of tomato roots obtained in Example 1, the following differences exist: (1) The imaging method of the fisheye lens, which acquires a hemispherical image at a time, has obvious imaging distortion in the edge area of ​​the field of view. After the image is unfolded and corrected, the root morphology in the edge area still has a certain degree of distortion, which affects the overall accuracy of the stitched root image. In contrast, the present invention uses a rotating stepper motor to drive the camera 31 to acquire multiple small-area local root images at each axial position and then stitches them together. The distortion of a single image is smaller, and the root morphology after stitching is more realistic. (2) The ring-shaped LED strong light direct illumination is prone to forming local light spots and overexposed areas on the tube wall. Even with the light-absorbing cloth, the brightness uniformity between the sequence images is still not easy to guarantee, which has an adverse effect on the stitching and fusion. In contrast, the supplementary light module 32 of the present invention adopts a back-facing indirect illumination method. After the light is reflected, it illuminates the tomato root observation area, and the brightness of the sequence image is more uniform, which reduces the overexposure phenomenon. (3) In the fisheye lens imaging mode, the rotational positioning and axial movement stability of the acquisition component in the tube are not easy to control, and the image overlap ratio between adjacent acquisition positions is prone to shift. However, the present invention improves the movement stability of the acquisition component in the tube and the stability of the overlap ratio between adjacent images through the stepper motor, the rolling support component 41 and the connecting and fixing module 33, providing quality assurance for subsequent stitching. (4) This method can only acquire root distribution images within a large scale range and cannot simultaneously acquire root hair micro images. However, the present invention adds a micro root hair observation tube 6, which can realize the coordinated in-situ monitoring of the entire root system and root hairs.

[0092] Comparative Example 2: Observation of root hairs on a glass slide culture medium

[0093] This comparative example uses tomato seeds of the same variety as in Example 1 for root hair observation. The difference is that the root hair observation method adopts the method described in the prior art (CN109943619B): 1 / 2 MS solid culture medium is set on the surface of a glass slide, tomato seeds are sown on the culture medium block, covered with a biaxially stretched polypropylene film, and after being cultured in a sterile humid chamber, the root hairs are observed using a regular microscope or a fluorescence microscope.

[0094] The root hair observation results of Comparative Example 2 were compared with the root hair observation and parameter measurement results of Examples 1 and 3. The following differences were found: (1) The slide culture medium method can avoid root hair being compressed, deformed and dehydrated during slide preparation. However, the root growth environment is a solid culture medium rather than cultivation soil. The observation results are difficult to reflect the true development state of root hair under soil conditions. Moreover, the volume of the slide culture system is limited, and the plant can only be cultured to the seedling stage. In contrast, the present invention pre-embeds micro root hair observation tubes 6 in the cultivation substrate. Tomato root hair grows in the real soil environment and is attached to the surface of the transparent observation plate 62 for observation. The observation results are closer to the actual cultivation conditions. (2) The slide culture medium method is a one-time slide observation, which cannot be used for long-term and repeated observation of the same plant. In contrast, the present invention does not require the preparation of slide culture medium slides. It can be used to conduct long-term and repeated observation of the root hair of the same plant without damaging the root system and soil structure. (3) The observation results of the slide method rely on manual microscopic examination and interpretation, and it is difficult to obtain quantitative results automatically. However, the present invention keeps the imaging distance between the high-definition camera 61 and the transparent observation plate 62 fixed by the limiting part, and combined with the root axis and root hair separation and parameter calculation method (Example 3), the root axis length, root hair length, root hair area and root hair number can be obtained automatically, realizing the extension from root hair image observation to root hair phenotypic quantitative analysis.

Claims

1. An in-situ crop root monitoring device comprising a micro root tube (1), a micro root hair observation tube (6), a sliding table (4) and a control unit (54), characterized in that, The microroot canal (1) and the microroot hair observation tube (6) are pre-installed at a set angle to the horizontal plane in the cultivation substrate before crop planting. The slide (4) is detachably installed inside the microroot canal (1). The slide (4) is equipped with an image acquisition module, a supplementary lighting module (32), and a rotary stepper motor. The image acquisition module and the supplementary lighting module (32) are fixedly connected to the output end of the rotary stepper motor through a connecting and fixing module (33). The rotary stepper motor drives the image acquisition module and the supplementary lighting module (32) to rotate one revolution at each axial position to acquire a circumferential sequence. The slide (4) is threaded with the screw of the transmission module (2) through a three-hole flange round nut (42). The micro hair observation tube (6) is provided with a transparent observation plate (62). The micro hair observation tube (6) is provided with a limiting part for clamping the high-definition camera (61). The limiting part keeps a fixed distance between the lens of the high-definition camera (61) and the transparent observation plate (62). The control unit (54) is electrically connected to the transmission module (2), the rotary stepper motor, the image acquisition module, the supplementary light module (32) and the high-definition camera (61) respectively.

2. The in situ crop root monitoring apparatus of claim 1, wherein, A rolling support assembly (41) is provided on the outer periphery of the slide (4), and the rolling support assembly (41) makes rolling contact with the inner wall of the microroot canal (1).

3. The in situ crop root monitoring apparatus of claim 1, wherein, The image acquisition module includes at least one camera (31), and the camera (31) is connected to the rotary stepper motor through the connection fixing module (33).

4. The in situ crop root monitoring apparatus of claim 1, wherein, The supplementary lighting module (32) includes four LED light-emitting elements, which are symmetrically distributed on both sides of the image acquisition module. The LED light-emitting elements are set in a backlighting manner.

5. The in situ crop root monitoring apparatus of claim 1, wherein, The image acquisition module and the supplementary lighting module (32) are fixed to the connection and fixing module (33) by screws and nuts. The connection and fixing module (33) is fixedly connected to the output end of the rotary stepper motor.

6. The in situ crop root monitoring apparatus of claim 1, wherein, The rotary stepper motor is fixedly installed on the slide (4), and the three-hole flange round nut (42) is fixed on the slide (4). The three-hole flange round nut (42) is threadedly engaged with the lead screw of the transmission module (2).

7. The in situ crop root monitoring apparatus of claim 1, wherein, The transmission module (2) includes a lead screw and a lead screw stepper motor for driving the lead screw to rotate, and the lead screw stepper motor is electrically connected to the control unit (54).

8. A root phenotypic parameter analysis system, applied to a PC terminal communicatively connected to the control unit (54) of the in-situ crop root monitoring device according to any one of claims 1 to 7, characterized in that, include: The image receiving module is used to receive root image information transmitted by the control unit (54) and to import root images stored locally; the root image stitching module is used to stitch together the local root images obtained by the microroot canal (1) in sequence to obtain the overall root image; the model segmentation module is used to perform root target segmentation on the overall root image to generate the corresponding binary root image; the root stitching image parameter extraction module is used to perform skeleton extraction and depth layering processing on the binary root image to obtain at least one overall root parameter among the total root length, total root area, layered root length and layered root area of ​​the root stitching image; the root hair image parameter extraction module is used to perform root axis and root hair separation, skeletal line extraction and root hair parameter calculation on the root hair binary image to obtain at least one root hair parameter among the root axis length, root hair area, root hair length and root hair number; the result output module is used to export and save the overall root parameters and the root hair parameters as well as the images in the processing process.

9. The root phenotypic parameter analysis system according to claim 8, characterized in that, The model segmentation module includes a sliding window segmentation unit and a prediction fusion unit. The sliding window segmentation unit is used to divide the original root system image into multiple overlapping image blocks. The prediction fusion unit calls the segmentation model to predict each image block and performs overlapping region fusion on the prediction results of each image block to obtain a prediction probability map corresponding to the size of the original root system image.

10. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program to implement the functions of the root phenotypic parameter analysis system as described in claim 8.

Images