Plant photosynthetic physiology three-dimensional imaging device and method

Abstract

The invention discloses a plant photosynthetic physiology three-dimensional imaging device and method, and belongs to the field of plant chlorophyll fluorescence three-dimensional detection and analysis. The plant photosynthetic physiology three-dimensional imaging device and method aim at solving the problems that in the prior art, plant chlorophyll fluorescence three-dimensional imaging is complex, single-view reconstruction of a complex geometric structure and a shielding part is limited, and the photosynthetic efficiency heterogeneity analysis capacity is poor. According to the method, the uniformity of an exciting light source is improved by adopting a free-form surface reflection cup and a staggered light source array; two groups of precise three-dimensional point cloud information of the plant are obtained by adopting double-path structured light, and matching, segmentation and complementation are executed to realize high-quality reconstruction of the three-dimensional form of the plant; and finally fusing the plant color image and the chlorophyll fluorescence image to realize three-dimensional visual display of plant photosynthetic physiology. The plant photosynthetic physiology three-dimensional imaging precision and speed can be improved, and technical support is provided for crop photosynthetic physiology heterogeneity analysis.

Description

Technical Field

[0001] The present application relates to the technical field of plant chlorophyll fluorescence, and in particular to a three-dimensional imaging device and method for plant photosynthetic physiology. Background Art

[0002] In natural and agricultural production environments, photosynthetic physiological information is an important indicator for assessing plant health, growth potential, and environmental adaptability. In order to fully understand the photosynthetic physiological information of crops, it is necessary not only to consider traditional growth parameters, such as crop morphological characteristics and chlorophyll content, but also to pay attention to multiple parameters such as chlorophyll fluorescence. As an effective probe for photosynthesis research, chlorophyll fluorescence has become an important means to measure crop photosynthetic efficiency. Plant photosynthetic physiology three-dimensional imaging can intuitively display plant growth and photosynthetic information, and realize real-time monitoring of plant growth and development. In addition, the use of a three-dimensional model of plant photosynthetic physiology can provide technical support for the heterogeneity analysis of crop photosynthetic physiology. Therefore, fast and accurate three-dimensional imaging of plant photosynthetic physiology is crucial for research on plant growth and development, disease prevention and control, and response to adverse stress.

[0003] Common three-dimensional imaging methods include voxel grids, neural radiation fields, point clouds, etc. For example, the Chinese invention patent document with publication number CN116678862A discloses a plant chlorophyll fluorescence three-dimensional imaging device and method. The device uses a blue-violet LED light as an excitation light source to detect chlorophyll fluorescence, obtains a single-view image of the plant under light sources at different positions, reconstructs the plant in three dimensions through the plant neural radiation field, generates a detailed three-dimensional model of the plant, and optimizes the detailed three-dimensional model of the plant in combination with rendering to obtain a three-dimensional image of the plant and a three-dimensional image of the plant chlorophyll fluorescence.

[0004] Most of the fluorescence imaging systems in the prior art use two-dimensional imaging, which cannot accurately reflect the three-dimensional structure of plant leaves and the spatial heterogeneity of photosynthetic efficiency. Summary of the invention

[0005] 1. Problem to be solved Based on this, it is necessary to provide a plant photosynthetic physiology three-dimensional imaging device, imaging method and system that can improve the accuracy and speed of plant photosynthetic physiology three-dimensional imaging in order to address the above technical problems.

[0006] 2. Technical solution In the first aspect, the present application provides a plant photosynthetic physiological three-dimensional imaging device. The device includes: a dark box, a sealing cloth is provided on the dark box, a plant rotating tray is provided at one end of the dark box away from the sealing cloth, the sealing cloth and the plant rotating tray are used to close the dark box, a light source module and a structured light projection module are provided at one end of the dark box facing the plant rotating tray, the light source module includes a white light source composed of a plurality of white LED lamp beads and a blue light source composed of a plurality of blue LED lamp beads, the white light source and the blue light source are arranged in an alternating manner, a free-form reflector cup is provided on the LED lamp beads, at least two shooting modules are provided on the dark box, a filter wheel is provided on the shooting module, the filter wheel includes at least RGB three-band filters, a near-infrared filter and a lightless filter, and the shooting module is connected to a host computer.

[0007] In a second aspect, the present application also provides a method for three-dimensional imaging of plant photosynthetic physiology. The method comprises: Switch between white light source and blue light source, and collect plant color images under white light source and plant chlorophyll fluorescence images under blue light source respectively based on the shooting module and the filter wheel; Collect plant structured light stripe images through a shooting module; Based on the three-frequency four-step phase-shift structured light algorithm, plant structured light stripe images are processed and three-dimensional point cloud information is obtained; Extract the two sets of generated 3D point cloud information, reconstruct the plant in 3D by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; Render plant color images and plant chlorophyll fluorescence images into high-precision three-dimensional plant models and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

[0008] In one of the embodiments, before switching the white light source and the blue light source, the method includes: Initialize an LED staggered array; Presetting particle swarm algorithm parameters, wherein the particle swarm algorithm parameters at least include a range of inertia weight and a learning factor; Calculate the standard deviation of the initial array light intensity distribution corresponding to each particle; Based on the simulated light intensity distribution, the light intensity of each LED lamp bead is superimposed on the target plane according to the distance attenuation formula to obtain the light intensity distribution diagram; Calculate the standard deviation of the initial array light intensity distribution and dynamically adjust the value of the inertia weight w by adopting nonlinear decreasing. The formula of nonlinear decreasing inertia weight is as follows: ; in, is the initial inertia weight, which is the maximum value. is the final inertia weight, which is the minimum value. is the current iteration number, is the maximum number of iterations.

[0009] In one embodiment, initializing an LED staggered array includes: A free-form surface reflector is constructed based on the LED staggered array, wherein the free-form surface reflector comprises a first curved surface and a second curved surface, wherein the first curved surface is used for small-angle light to directly emit to a preset target surface, and the second curved surface is used for large-angle light to be reflected by the inner surface of the reflector to reach the preset target surface; Construct a light source intensity distribution model, the light source intensity distribution model formula is as follows: ; in, is the angle between the light source and the optical axis; The distance from the light source in the direction of the optical axis is The radiance at is the radiation pattern of the light source. When the light source is a perfect ideal Lambertian light source, =1; The non-Lambertian properties are corrected based on the non-ideal light source characteristic correction model. The intensity distribution of the corrected light source is as follows: ; in, is the correction factor, is the angle between the light source and the optical axis; The distance from the light source in the direction of the optical axis is The radiance at is the radiation pattern of the light source; Modeling of free-form reflector, any point on the focal plane of free-form reflector The irradiation intensity is as follows: ; in, is the brightness of the LED chip, in units of ; is the reflective cup area; is the position coordinate of the reflective cup; If exists When the free-form reflectors are illuminated at the same time, the total intensity formula is as follows: = ; The total strength formula of the free-form reflector is as follows: ; in, and are the number of light sources in the rectangular array and the circular array, and The area of ​​the reflector cups for rectangular and circular arrays, and The first The radius and angle of the light source.

[0010] In one embodiment, processing a plant structured light stripe image based on a three-frequency four-step phase-shift structured light algorithm includes: Based on the three-frequency four-step phase shift structured light algorithm, the plant structured light stripe image is processed to generate accurate three-dimensional point cloud information and ; The improved maximum clique 3D registration method is used to align two sets of 3D point cloud information. and Perform registration to produce highly accurate point cloud information; The two network models BIOneFormer3D and RSVDFormer are used to segment and complete the 3D point cloud information respectively to generate a high-precision 3D model of the plant; By fusing plant color images and chlorophyll fluorescence images, high-precision three-dimensional rendering of plants can be achieved, and plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information can be obtained.

[0011] In one embodiment, an improved maximum clique 3D registration method is used to register two sets of 3D point cloud information. and The point cloud information produced by the registration includes: Extract local features of point clouds based on the deep learning model PointNet++; Based on the local features, the graph neural network is inputted for processing, and neighborhood information is aggregated to update the feature representation of each node; Based on the feature representation output by the graph neural network, the quality of the matching point pairs is re-evaluated, and more reliable matching point pairs are screened out to form an initial corresponding set; Construct first-order graph and second-order graph to realize the construction of compatibility graph and search all maximal cliques in the graph based on Bron-Kerbosch algorithm; Sort the maximum clusters based on the preset node weights and intra-cluster consistency, giving priority to clusters with higher weights and better consistency; The singular value decomposition algorithm is used to calculate the transformation hypothesis for the selected group, and two sets of point clouds are realized based on the optimal hypothesis. and of the registration.

[0012] In one embodiment, two network models, BIOneFormer3D and RSVDFormer, are used to segment and complete the three-dimensional point cloud information respectively, including: BIOneFormer3D specifically includes: sparse 3D U-Net backbone network, super point pooling operation, BI-Transformer Decoder Layer and output layer; The sparse 3D U-Net backbone network includes an encoder and a decoder. The encoder part is composed of multiple layers of sparse convolutional layers to extract higher-level features layer by layer. The features of the encoder are directly transferred to the corresponding layer of the decoder using skip connections to achieve feature fusion; The decoder outputs point-by-point features with the number of channels C. The point-by-point features extracted by the sparse 3D U-Net are aggregated into super-point features through super-point pooling operation; The output super-point features are input into the BI-Transformer Decoder Layer as keys and values, and the learnable semantic queries and instance queries are passed as input to the BI-Transformer Decoder Layer; The RSVDFormer network model specifically includes: input data, feature extraction backbone, feature fusion module, and decoding and upsampling module; The input data includes part of the low-resolution point cloud data and a two-dimensional image generated by multi-view mapping of the 3D point cloud data; The feature extraction backbone is divided into 3D Backbone and 2D Backbone; 3D Backbone uses PointNet++ network to extract features from input 3D point cloud data and generate feature vectors ; 2D Backbone uses the MobileNetV3 network to extract features from two-dimensional images and generate feature vectors .

[0013] The feature fusion module constructs a new multi-modal cyclic feature fusion module; The decoding and upsampling module generates a rough point cloud using a 1D convolutional transpose layer and a self-attention layer. , merged with the original point cloud data and resampled to get a rough result ; Generated rough point cloud Entering the detail optimization stage, the SDG module is used for refinement and upsampling, and finally a high-precision complete point cloud is generated. .

[0014] In a third aspect, the present application also provides a plant photosynthetic physiology three-dimensional imaging system. The system includes: The plant color image and chlorophyll fluorescence image acquisition module is used to switch between the white light source and the blue light source, and to respectively acquire the plant color image under the white light source and the plant chlorophyll fluorescence image under the blue light source based on the shooting module and the filter wheel; A plant structured light stripe image acquisition module is used to collect plant structured light stripe images through a shooting module; A three-dimensional point cloud acquisition module is used to process plant structured light stripe images and obtain three-dimensional point cloud information based on a three-frequency four-step phase-shift structured light algorithm; A high-precision 3D model generation module is used to extract the two sets of generated 3D point cloud information, perform 3D reconstruction of the plant by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; The three-dimensional image and photosynthetic physiological acquisition module is used to render plant color images and plant chlorophyll fluorescence images into a high-precision three-dimensional model of the plant and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

[0015] In a fourth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program: Switch between white light source and blue light source, and collect plant color images under white light source and plant chlorophyll fluorescence images under blue light source respectively based on the shooting module and the filter wheel; Collect plant structured light stripe images through a shooting module; Based on the three-frequency four-step phase-shift structured light algorithm, plant structured light stripe images are processed and three-dimensional point cloud information is obtained; Extract the two sets of generated 3D point cloud information, reconstruct the plant in 3D by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; Render plant color images and plant chlorophyll fluorescence images into high-precision three-dimensional plant models and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

[0016] In a fifth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented: Switch between white light source and blue light source, and collect plant color images under white light source and plant chlorophyll fluorescence images under blue light source respectively based on the shooting module and the filter wheel; Collect plant structured light stripe images through a shooting module; Based on the three-frequency four-step phase-shift structured light algorithm, plant structured light stripe images are processed and three-dimensional point cloud information is obtained; Extract the two sets of generated 3D point cloud information, reconstruct the plant in 3D by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; Render plant color images and plant chlorophyll fluorescence images into high-precision three-dimensional plant models and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

[0017] 3. Beneficial effects (1) The present invention proposes a plant photosynthetic physiology three-dimensional imaging device, which uses an improved particle swarm algorithm to optimize the LED array to form a staggered LED array, thereby overcoming the problem of uneven light intensity caused by most LED arrays. The improved LED array greatly improves the uniformity of light intensity. The light energy emitted by the free-form reflector cup can compensate for the uniformly strong illumination area on the receiving surface, converge and collect the large-angle emitted light, so that more light reaches the target surface, and finally forms a light spot distribution with uniform light intensity, which greatly improves the uniform light spot distribution.

[0018] (2) The present invention proposes a three-dimensional imaging method for plant photosynthetic physiology. The structured light stripes projected by the structured light projector are clear and low-noise on the crop, which is easy to reconstruct the plant in three dimensions. The point cloud information is generated by using a dual camera and a three-frequency four-step phase-shifted surface structured light algorithm, thereby collecting two groups of seven structured light stripe images of the plant, and the two groups of point cloud information generated are registered, segmented, and completed, which overcomes the problem that the complex geometric structure and occluded parts in the single-view input scene are difficult to accurately reconstruct, thereby greatly improving the accuracy of the plant point cloud information, and finally obtaining a high-precision plant three-dimensional model. It also overcomes the problems of low efficiency of plant three-dimensional reconstruction caused by single-camera image acquisition and single point cloud information with low accuracy, and greatly improves the accuracy and efficiency of plant three-dimensional point cloud information reconstruction.

[0019] (3) The present invention proposes a network model named BIOneFormer3D. In general network models, the point cloud feature extraction and decoder parts usually rely on the standard Transformer architecture. The BI-TransformerDecoder Layer decoder used in this model improves the network's performance on point cloud segmentation tasks and further reduces the computational complexity of the self-attention module. Through this combination, this method enhances the contextual perception and segmentation accuracy of the original OneFormer3D network when segmenting plant 3D point clouds, while improving the attention to difficult-to-classify samples and the overall segmentation performance.

[0020] (4) This study proposed a network model called RSVDFormer to overcome the problem of incomplete point clouds caused by occlusion. In terms of phenotypic extraction, since the detailed geometry of the missing area is difficult to recover, a multimodal recurrent feature fusion module is embedded to fully integrate multiple modal information to explicitly encourage the network to pay more attention to the missing area information. By using the RSVDFormer point cloud completion model, the point cloud completeness of occluded parts such as leaves and stems has been greatly improved, further improving the accuracy of extracting phenotypic information such as leaf area and stem width. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of the structure of a plant photosynthetic physiology three-dimensional imaging device according to an embodiment of the present invention; Figure 2 This is a flow chart of a method for acquiring three-dimensional photosynthetic physiological information of plants according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the framework of a three-dimensional imaging method for plant photosynthetic physiology according to an embodiment of the present invention; Figure 4 This is a structural diagram of the BI-OneFormer3D segmentation model according to an embodiment of the present invention; Figure 5 This is a structural diagram of the RSVDFormer point cloud completion model according to an embodiment of the present invention; Figure 6 A structural block diagram of a plant photosynthetic physiology three-dimensional imaging system in one embodiment; Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment.

[0022] Figure numerals: 1. dark box; 2. sealing cloth; 3. plant rotating tray; 4. light source module; 41. white light source; 42. blue light source; 5. structured light projection module; 51. structured light projector; 6. LED staggered array; 7. free-form reflective cup; 8. shooting module; 9. filter wheel; 10. host computer. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0024] Example 1 like Figure 1As shown, a plant photosynthetic physiological three-dimensional imaging device provided in this embodiment. The imaging device comprises: a dark box 1, a sealing cloth 2 is provided on the dark box 1, a plant rotating tray 3 is provided at one end of the dark box 1 away from the sealing cloth 2, the sealing cloth 2 and the plant rotating tray 3 are used to close the dark box 1, a light source module 4 and a structured light projection module 5 are provided at one end of the dark box 1 facing the plant rotating tray 3, the light source module 4 comprises a white light source 41 composed of a plurality of white LED lamp beads and a blue light source 42 composed of a plurality of blue LED lamp beads, the white light source 41 and the blue light source 42 are arranged in a staggered manner, the LED lamp beads are provided with a free-form reflective cup 7, at least two shooting modules 8 are provided on the dark box 1, the shooting module 8 is provided with a filter wheel 9, the filter wheel 9 comprises at least filters of three bands of RGB, a near-infrared filter and a lightless filter, and the shooting module 8 is connected to a host computer 10.

[0025] Among them, the structured light projection module 5 is composed of a single structured light projector 51, and a shooting module 8 is installed on the left and right side walls of the dark box 1 respectively. The shooting module 8 can be a camera. A filter wheel 9 is installed in front of the lens of each shooting module 8, and the required filter can be switched by the filter wheel 9. It also includes a host computer 10, which is connected to the dark box 1.

[0026] In one embodiment, specifically in this embodiment, the shape of the dark box 1 includes a polygonal shape, a cylindrical shape, and the like. In this embodiment, preferably, the shape of the dark box 1 is a polygonal shape, and the polygonal shape has certain advantages over the cylindrical dark box 1 in terms of space utilization, heat dissipation performance, electromagnetic wave absorption performance, quiet zone characteristics, and economy. The height of the dark box 1 is between 35 cm and 70 cm, so that more plants of different heights can be collected. In this embodiment, preferably, the height of the dark box 1 is set to 55 cm; in this embodiment, the color of the dark box 1 is black, and black can effectively reduce the reflection of light, thereby reducing interference from the image background. At the same time, the black dark box 1 also provides a dark adaptation environment for plants to measure chlorophyll fluorescence.

[0027] The plant rotating tray 3 at the bottom of the dark box 1 is set as a circular tray, which is used to support and rotate the plants to be measured to form multi-angle collection, and the electric lifting facilitates the adjustment of the relevant distance of the plants to be measured. During the measurement process, the shooting modules 8 on both sides of the dark box 1 are set at a fixed height. If the collected plants are short, the electric lifting of the plant rotating tray 3 is required to adjust the height, so that the shooting module 8 can collect better image effects.

[0028] It is worth noting that the dark box 1 and the plant rotating tray 3 are both treated with black paint, and the materials are frosted. The frosted material can effectively reduce the reflection and transmission of light, and make the light emitted by the light source module 4 evenly distributed inside the dark box 1, thereby reducing the impact of uneven light distribution on the three-dimensional imaging of plant chlorophyll fluorescence and improving the stability of system measurement. In addition, the surface of the frosted material is rough. This surface characteristic causes the sound to produce viscosity and friction when it contacts the frosted material, achieving the effect of absorbing sound, thereby reducing the reflection of sound to achieve the effect of reducing background noise, improving the signal-to-noise ratio, and helping to improve the detection efficiency and measurement accuracy of chlorophyll fluorescence signals.

[0029] The light source module 4 is installed in the dark box 1 at one end close to the sealing cloth 2, and is used to emit two light source environments to the plants to be tested. The light source module 4 includes a white light source 41 and a blue light source 42, using LED blue light in the 440-450nm band and white light in the 400-700nm band, and the light source environment can be switched freely.

[0030] The light source module 4 assembled in the above-mentioned method optimizes the array of the LED staggered array 6 by using an improved particle swarm algorithm to improve the uniformity of light distribution. First, an LED staggered array 6 is initialized, which is mainly obtained by combining and staggering rectangular and circular arrays. Then, 46 particles are randomly generated and arranged in the initial staggered array, where the position of each particle represents the coordinates of an LED, and the algorithm parameters such as the range of the inertia weight and the learning factor are set; then, the standard deviation of the initial array light intensity distribution corresponding to each particle is calculated, and the light intensity of each LED is superimposed on the target plane according to the distance attenuation formula by simulating the light intensity distribution to obtain a light intensity distribution diagram, and then the standard deviation of the initial array light intensity distribution is calculated; the value of the inertia weight w is dynamically adjusted by nonlinear decreasing, and the global search and local search capabilities are balanced, the global search capability is enhanced in the early stage, and the local search capability is enhanced in the later stage. The nonlinear decreasing inertia weights used are as follows: ; Among them is the initial inertia weight, which is the maximum value. is the final inertia weight, which is the minimum value. is the current iteration number, is the maximum number of iterations.

[0031] In addition, improve individual cognitive factors and group cognitive factors , in the initial control stage control Take the larger value, Take a smaller value to enhance the global search capability, and enhance the local search capability in the final iteration. Use sine and cosine functions to control , ,let The value of can decrease nonlinearly, while The value of increases nonlinearly, and the improved formula is as follows: ; ; in is the individual cognitive factor, is the group cognitive factor, is the current iteration number, is the maximum number of iterations.

[0032] Through the above improved particle swarm algorithm, the initial LED array coordinates are iteratively updated, and the coordinate values ​​of the LEDs in the light source array are continuously adjusted and optimized to obtain the minimum standard value. The smaller the standard value, the optimal spacing and arrangement of the matrix and the annular multi-shape combination array are finally determined, thereby generating an improved staggered LED array. In addition, the improved LED staggered array 6 is optimized, and each LED bulb in the array is equipped with a free-form reflector cup 7, so that more light can reach the target surface evenly. First, a free-form reflector cup 7 is designed, which is mainly divided into a first surface and a second surface design, wherein the first surface does not transmit light, and the second surface mainly reflects the large-angle light emitted by the light source; the light source light in the free-form reflector cup 7 system is mainly divided into two parts, one part is the small-angle light that can be directly emitted to the target surface, and the other part is the large-angle light that needs to be reflected by the inner surface of the free-form reflector cup 7 to reach the target surface. In the light source of a single free-form reflector cup 7, the light source intensity distribution is: ; in, is the light source intensity distribution, is the angle between the light source and the optical axis; The distance from the light source in the direction of the optical axis is The radiance at is the radiation pattern of the light source. When the light source is a perfect ideal Lambertian light source, =1, that is, its light intensity distribution has nothing to do with the observation angle, but in actual situations The value is greater than 30, that is, the light intensity distribution of the actual light source often changes with the angle, resulting in non-uniformity of the light source. Therefore, the non-ideal light source characteristic correction model is used to correct the non-Lambertian property, and a correction factor is introduced to adjust the light intensity distribution. The intensity distribution of the light source after correction is: ; in, is the intensity distribution of the corrected light source, is the angle between the light source and the optical axis; The distance from the light source in the direction of the optical axis is The radiance at is the radiation pattern of the light source, is a correction factor to compensate for non-Lambertian properties.

[0033] Finally, the free-form surface reflector 7 is modeled. According to the inverse square law, in the spatial coordinate system, any point on the focal plane of the free-form surface reflector 7 The irradiation intensity can be expressed as: ; In the formula: Any point on the focal plane of the free-form reflector 7 The radiation intensity, is the brightness of the LED chip, in units of ; is the reflective cup area; is the position coordinate of the reflective cup, is the radiation pattern of the light source. If it exists When the free-form reflective cups 7 are irradiated simultaneously, the total intensity can be expressed as: = ; The LED staggered array 6 mentioned above is obtained by staggering a combination of a rectangular array and a circular array. The optimal spacing and arrangement of the multi-shape combination array are determined by an improved particle swarm algorithm. Therefore, the total intensity formula of the array equipped with a free-form reflector cup is: ; in, Any point on the focal plane of the free-form reflector 7 The radiation intensity, is the brightness of the LED chip, in units of ; is the number of light sources in the rectangular array, is the number of light sources in the circular array; The area of ​​the rectangular array equipped with the free-form reflector cup 7, The area of ​​the annular array is provided with a free-form reflective cup 7; The first The radius of the light source, The first The angle of the light source.

[0034] By optimizing the design of the LED staggered array 6 through the above formula, an optimized array and total light intensity can be obtained. The greater the total light intensity, the better the uniformity of light distribution. By designing the free-form surface reflector 7, the large-angle emitted light is converged and collected, so that more light reaches the target surface. The final LED staggered array 6 has good distribution uniformity, thereby greatly improving the uniformity of illumination.

[0035] Since the light source system of the free-form surface reflector 7 generally has a large part of light emitted through the inner wall, it is necessary to select a material with low light absorption and high reflectivity for production. Therefore, the free-form surface reflector 7 designed by the device uses aluminum material with low cost and good plasticity, and uses vacuum electroplating to form a layer of pure aluminum electroplating film on the inner wall of the free-form surface reflector 7, so that the reflectivity of the free-form surface reflector 7 can generally reach more than 90%, which can effectively reflect and converge light. It is assembled with each LED lamp of the LED staggered array 6. The free-form surface reflector 7 can compensate for the uniform strong illumination area on the receiving surface, converge and collect the large-angle emitted light, so that more light reaches the target surface, and finally forms a light spot distribution with uniform illumination intensity, which greatly improves the uniform light spot distribution. The semi-aperture of the free-form surface reflector 7 processed by the device is 5mm, the radius of curvature of the second surface is 1, the cone constant of the second surface is -1.2, the thickness is 1mm, and the inner wall of the free-form surface reflector 7 is electroplated, and the surface has good reflection characteristics. For the structured light projection module 5: the structured light of this module has a light source with a wavelength of 420nm-700nm and a resolution of 912*1140. It can project monochromatic light with a color depth of 8bit. The projected structured light stripes are clear on the crops with low noise, and are easy to reconstruct the crops in three dimensions.

[0036] The image acquisition module is installed on the left and right side walls of the dark box 1 to collect chlorophyll fluorescence images and RGB color images of the plants to be tested; the image acquisition module includes a shooting module 8 and a filter wheel 9, and the filter wheel 9 can be switched to an RGB three-band filter, a near-infrared band filter and a lightless filter; the RGB three-band filters are a 450nm blue light band filter with a bandpass of 50K, a 550nm green light band filter with a bandpass of 50K, and a 650nm red light band filter with a bandpass of 50K. The central transmittance reaches 95%, which can well collect RGB three-color information for perfectly restoring the shape of real crops.

[0037] The near-infrared band filter uses a narrow-band filter with a 690nm bandpass of 20K, a transmittance of up to 90%, and a cutoff depth of OD5, which perfectly transmits the fluorescence of the chlorophyll fluorescence peak band. Specifically, when the LED white light source is turned on to light up the plant, the host computer 10 controls the shooting module 8 to turn on the red light band, the green light band, and the blue light band at the same time, and the filter wheel 9 rotates the filter in a clockwise direction in sequence, and the plant images and plant shadow images taken at the same time by the red light band, the green light band, and the blue light band are superimposed to obtain a plant color image.

[0038] Further, when the blue light source 42 is turned on, the host computer 10 controls the shooting module 8 while the filter wheel 9 rotates to the near-infrared band filter to collect the plant chlorophyll fluorescence image. Further, to collect the stripe image, the shooting module 8 is installed on both sides. The structured light stripe image of the plant collected by a single camera is too simple, and the three-frequency four-step phase-shifted surface structured light algorithm is used. The accuracy of the generated point cloud information is low and the efficiency is low. The shooting modules 8 of this device are respectively installed on the opposite sides of the left and right walls of the dark box 1, and the three-frequency four-step phase-shifted surface structured light algorithm is used to generate point cloud information, thereby collecting two groups of 7 structured light stripe images of the plant, and the two groups of point cloud information generated are registered, segmented, and completed, thereby greatly improving the accuracy of the plant point cloud information, and finally obtaining a high-precision plant three-dimensional model.

[0039] Furthermore, the host computer 10 is connected to the shooting module 8, and can generate two sets of accurate crop three-dimensional point clouds by processing the two sets of structured light stripe images collected by the two cameras, align, segment, and complete the two sets of point cloud information generated, and output high-precision three-dimensional crop perspective pictures, and then process and analyze the chlorophyll fluorescence image and RGB color image information, and finally render the three-dimensional point cloud data with the photosynthesis efficiency information and the color image information to generate a three-dimensional photosynthetic phenotype image of the plant.

[0040] In this embodiment, the host computer 10 can control the dark box 1 and the plant rotating tray 3 inside the dark box 1 to lift and adjust the required height of the experiment, control the light source module 4 to turn on and switch and turn off the light source, control the start of the shooting module 8 and the rotation of the filter wheel 9 to collect multi-band images, and control the start of the structured light projector 51 and the synchronous data collection of the shooting module 8. The host computer 10 can also process and analyze the collected image data, and visualize the three-dimensional imaging of plant chlorophyll fluorescence. In addition, the host computer 10 analyzes and processes the plant color images, plant chlorophyll fluorescence images, and structured light stripe images collected from the shooting module 8, uses point cloud information to perform three-dimensional reconstruction, and renders them with fluorescence and color images.

[0041] Therefore, the present embodiment provides a plant chlorophyll fluorescence and photosynthetic physiology three-dimensional imaging device, in which a white light source 41, a blue light source 42 and a structured light projector 51 are arranged in a dark box 1, and the plants are photographed from multiple angles through the dual cameras in the shooting module 8, so as to quickly obtain plant color images, plant chlorophyll fluorescence images and plant structured light stripe images, thereby overcoming the problems of complex geometric structures and limited occlusion parts in single-view reconstruction and low accuracy of point cloud information, and effectively improving the accuracy and efficiency of plant three-dimensional point cloud information reconstruction.

[0042] Example 2 In this embodiment, a plant chlorophyll fluorescence and photosynthetic physiology three-dimensional imaging device described in Example 1 is used to perform chlorophyll fluorescence three-dimensional imaging of plants. Figure 2 and Figure 3 As shown, the specific steps include: A plant chlorophyll fluorescence and photosynthetic physiology three-dimensional imaging device based on the above is constructed. Further, the plant is placed in the plant rotating tray 3, and the camera angles on both sides of the shooting module 8 are adjusted to an angle of 30 degrees with the vertical side wall.

[0043] Furthermore, plant color images under the white light source 41, chlorophyll fluorescence images under the blue light source 42, and plant structured light stripe images under structured light are collected respectively. Specifically, the host computer 10 controls the light source module 4 to switch to the white light source 41, and controls the filter wheel 9 to rotate the three-band filters of the red light band, the green light band, and the blue light band in a clockwise direction under the white light source 41. The shooting module 8 synchronizes the filter wheel 9 to collect plant images and plant shadow images of the red light band, the green light band, and the blue light band, wherein the RGB three-band filters are respectively a 450nm blue light band filter with a bandpass of 50K, a 550nm green light band filter with a bandpass of 50K, and a 650nm red light band filter with a bandpass of 50K.

[0044] Furthermore, the plant images and plant shadow images captured at the same time in the red light band, green light band and blue light band are superimposed to obtain a plant color image. Furthermore, the blue light source 42 is turned on, the filter wheel 9 is switched to the near-infrared band filter, and the shooting module 8 synchronizes the filter wheel 9 to collect the plant chlorophyll fluorescence image.

[0045] Specifically, the plants are first dark-adapted for 5-10 minutes in a dark box 1 environment. It should be noted that the purpose of dark adaptation before collecting plant chlorophyll fluorescence images is to reduce the photochemical damage of photosynthetic pigments and ensure that all PSII reaction centers are in an open state, so that the minimum fluorescence and maximum fluorescence can be accurately measured. In addition, for most plants, 5-10 minutes of dark adaptation treatment time is usually sufficient to re-oxidize the electron transport chain back to the initial dark state. The adaptation time of some plants is more special, and the dark adaptation time can be improved according to actual experimental needs.

[0046] Further, pulse-modulated blue light to low-intensity blue light is used as the measurement light to illuminate the plant, and the shooting module 8 uses a long integration time to collect the minimum fluorescence image. Further, pulse-modulated blue light to high-intensity blue light is used as the saturation light, and the shooting module 8 uses a short integration time to collect the maximum fluorescence image. It should be noted that after the photosynthesis of the plant reaches a certain light intensity, the photosynthetic rate no longer increases. At this time, this light intensity is called the light saturation point. When the saturation point is exceeded, the photosynthetic efficiency of the plant will not continue to increase, but may decrease. Long-term saturated light irradiation of plant leaves will cause the loss of photosynthetic activity of the plant leaves and even hinder the growth of the plant. Therefore, saturated light irradiation of plants generally only takes 1 second to 3 seconds.

[0047] Furthermore, the structured light projector 51 is turned on to project structured light, and the shooting module 8 shoots the plant stripe structured light image without a filter, wherein the structured light used has a light source with a wavelength of 420nm-700nm and a resolution of 912*1140, and can project monochromatic light with a color depth of 8bit. The projected structured light stripes are clear on the crops, with low noise, and are easy to reconstruct the crops in three dimensions.

[0048] Furthermore, the 7 structured light stripe images collected by the left and right shooting modules 8 are processed by using the three-frequency four-step phase-shifted surface structured light algorithm to obtain all the point cloud information collected by the left and right cameras of the crop shooting module 8. The two sets of point cloud information are registered to obtain more accurate point cloud information. The specific steps are as follows: the point cloud information on both sides is registered using the improved maximum clique 3D registration method called IMAC. and The registration is performed according to the following algorithm: First, the deep learning model PointNet++ is used to extract local features of the point cloud. Then, a graph structure is constructed based on these features. Specifically, each point in the point cloud corresponds to a node in the graph, and the feature similarity or geometric distance between any two points is calculated to determine the edge connection relationship between the nodes, thereby completing the construction of the graph. The above-constructed graph is processed as the input of the graph neural network (GNN), and the neighborhood information is aggregated to update the feature representation of each node; based on the feature representation output by the GNN, the quality of the matching point pairs is re-evaluated, and more reliable matching point pairs are screened out to form an initial corresponding set. ; Then, the compatibility graph is constructed by constructing the first-order graph and the second-order graph. Specifically, based on the matching points ) to construct a first-order graph (FOG) based on the rigid distance constraints between them and calculate its compatibility score: ; ; in, represents a point in the source space, represents the corresponding point in the target space, is the distance difference between two corresponding point pairs under transformation, is the distance parameter, is the compatibility score; if ) is greater than the threshold ,but and Form an edge , )for The weight of ) will be set to 0; the second-order graph (SOG) is derived from FOG, and the weight matrix It can be calculated as: ; in, is the weight matrix of the second-order graph (SOG), represents the element-wise product between two matrices, is the weight matrix of the first-order graph (FOG).

[0049] Furthermore, after constructing the compatibility graph, the improved Bron-Kerbosch algorithm is used to search for all maximal cliques in the graph by calling the igraph library function `igraph_maximal_cliques` in C++. Each maximal clique represents a consensus set. Subsequently, the maximal cliques are sorted based on the preset node weights and intra-clique consistency, and cliques with higher weights and better consistency are prioritized. The singular value decomposition (SVD) algorithm is used to calculate the transformation hypothesis for the selected cliques, and two sets of point clouds are realized based on the optimal hypothesis. and of the registration.

[0050] Furthermore, the singular value decomposition algorithm is used to calculate the transformation hypothesis for the selected clusters, and the best hypothesis is used to achieve the transformation of the two sets of point clouds. and After the point cloud registration is completed, the improved BIOneFormer3D network model is used to segment the point cloud information, such as Figure 4 As shown in the figure, the BI-Transformer Decoder Layer decoder is used to replace the general standard Transformer architecture in the improved BIOneFormer3D network model, thereby improving the network performance in point cloud segmentation tasks and further reducing the computational complexity of the self-attention module. Through this combination, this method enhances the context perception ability and segmentation accuracy of the original OneFormer3D network when segmenting plant three-dimensional point clouds, while improving the attention to difficult-to-classify samples and the overall segmentation performance.

[0051] It is worth mentioning that, first, sparse 3D U-net is used to extract point-by-point features of the point cloud. Sparse 3D U-Net consists of two parts: encoder and decoder. The encoder part consists of multiple layers of sparse convolutional layers, which extract higher-level feature representations layer by layer, and gradually reduce the spatial dimension through convolution operations with a step size of 2; the initial input contains 6-channel point cloud data containing RGB and XYZ information. After the first layer of sparse convolution, the number of output channels increases to 32; the subsequent convolutional layers expand the number of channels to 64, 128 and 256 respectively; the decoder part has three layers, and the spatial resolution is gradually restored through sparse deconvolution operations. Each layer is upsampled by a 2×2×2 deconvolution kernel, with a step size of 2 and a padding of 1; the deconvolution operation is followed by a ReLU activation function to restore the feature map size to the original point cloud feature size. The features of the encoder and the decoder are fused using skip connections, which directly transfer the features of the encoder to the corresponding layers of the decoder to restore the spatial resolution. Finally, the point-by-point features with C channels output by the decoder are aggregated into super-point features through super-point pooling operations, and the point-by-point features extracted by the sparse 3DU-Net are aggregated into super-point features through super-point pooling operations.

[0052] Specifically, the input point-by-point features are first divided into several super-points through the super-point graph algorithm, and each super-point contains a set of semantically consistent points. Through the average pooling operation, the point-by-point features within each super-point are aggregated into super-point features. The above-output super-point features are input into the BI-Transformer Decoder Layer as keys and values, and the learnable semantic queries and instance queries are passed as input to the BI-Transformer Decoder Layer for further feature decoding and semantic parsing. The bidirectional routing attention mechanism is a key component of the BI-Transformer Decoder Layer, which uses a 6-layer Transformer structure to capture long-distance dependencies in point cloud data.

[0053] Specifically, during the decoding process, the model uses a 3x3 depthwise separable convolution (DWConv) and a standard normalization layer (LN). The bidirectional routing attention mechanism is another key component of the decoder layer. First, the feature map Divide into S×S non-overlapping regions, and transform the features of each region into: ; and linearly project the query (Q), key (K), and value (V), represents the rth region feature, express regions, H, W and C represent the height, width and number of channels of the feature map respectively. represents a set of real numbers, which is used to describe the dimension and shape of a tensor. By constructing a directed graph to determine the participation relationship between regions, the average values ​​of Q and K in each region are calculated to obtain , , then calculate and The adjacency matrix of the regional correlation The formula is as follows: ; in, for and The adjacency matrix of the regional correlation, is the average value of Q in each region, The correlation graph is pruned by keeping only the top k connections in each region, averaging over K in each region: ; in, represents the pruned correlation graph, (·) function indicates that the i-th row contains the k indices of the most relevant regions in the i-th region, Represents the adjacency matrix; first aggregate the tensors of K and V for the k most relevant regions, and then use the attention operation on the aggregated QKV.

[0054] ; ; in, represents K and V after clustering the most relevant regions, represents an aggregate function, represents the attention function, represents the pruned correlation graph, represents the output after the operation; the function LCE(·) is parameterized using deep convolution and the kernel size is set to 3. Finally, the data processed by the attention mechanism passes through the MLP layer to generate the final output.

[0055] Furthermore, after all the point cloud information is segmented, the point cloud needs to be completed. Since the detailed geometry of the missing area is difficult to recover, a multi-modal cyclic feature fusion module is embedded to fully integrate multiple modal information to explicitly encourage the network to pay more attention to the missing area information. By using the RSVDFormer point cloud completion model, the point cloud integrity of occluded parts such as leaves and stems has been greatly improved, further improving the accuracy of extracting phenotypic information such as leaf area and stem width.

[0056] It is worth noting that in terms of phenotypic extraction, the occlusion between leaves and stems affects the accuracy of phenotypic analysis. In order to solve the problem of incomplete point cloud caused by occlusion, the present invention designs the RSVDFormer point cloud completion model. Figure 5 As shown in Figure 1, the input data of this model consists of two parts: one is the 3D point cloud data processed by 3D Backbone. , and the second is a two-dimensional image generated by multi-view mapping of 3D point cloud data. 3D Backbone uses PointNet++ network to extract features from the input 3D point cloud data and generate feature vectors 2D Backbone uses the MobileNet V3 network to extract features from the two-dimensional depth map and generate feature vectors . A new multimodal cyclic feature fusion module (MCFFM) was constructed.

[0057] Furthermore, MCFFM generates a random scaling factor of 0.5≤(α,β)≤1 for the 3D point cloud features. and 2D View Features Perform multi-scale scaling, fuse these multi-scale features, and feed them back to To generate enhanced features ; This process is repeated n times, and the multimodal features are finally fused Passed to the decoder to generate a rough point cloud . Such as the formula: ; in, is a rough point cloud, is the final fused multimodal feature, and For the The random scaling factor, is the number of accumulations. In the global shape generation stage, the enhanced features generated by MCFFM is used in the decoder to generate a more accurate global shape.

[0058] For example, if , then the final fused multimodal feature formula is as follows:

[0059] Furthermore, a rough point cloud is generated through a 1D convolutional transpose layer and a self-attention layer. , and Merge and resample to get a rough result The rough point cloud generated Entering the detail optimization stage, the SDG module is used for refinement and upsampling, and finally a high-precision complete point cloud is generated. ,This improved model structure further improves the performance of point cloud ,completion while making full use of multimodal information.

[0060] Therefore, the present embodiment provides a plant chlorophyll fluorescence three-dimensional imaging method, which obtains crop chlorophyll fluorescence information under a blue light source 42, obtains crop color images under a white light source 41, and processes 7 structured light stripe images collected by the left and right cameras under structured light to obtain all point cloud information collected by the left and right cameras of the crop, aligns the two sets of point cloud information to obtain more accurate point cloud information, generates a plant fine three-dimensional model, and finally fuses the plant color image and the chlorophyll fluorescence image to achieve a three-dimensional visualization display of plant photosynthetic physiology, providing technical support for the heterogeneity analysis of crop photosynthetic physiology. Using a fine three-dimensional model to better obtain plant chlorophyll fluorescence three-dimensional imaging, not only using a spatial three-dimensional angle to analyze plant photosynthetic information, but also using dual cameras to improve the accuracy of plant chlorophyll fluorescence three-dimensional imaging, reduce the difficulty of plant chlorophyll fluorescence three-dimensional imaging, can provide more accurate and comprehensive information for research on plant generation, photosynthetic information, diseases, etc., and has strong practicality and wide applicability.

[0061] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0062] Based on the same inventive concept, the embodiment of the present application also provides a plant photosynthetic physiology 3D imaging device for implementing the plant photosynthetic physiology 3D imaging method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more plant photosynthetic physiology 3D imaging device embodiments provided below can refer to the limitations of the plant photosynthetic physiology 3D imaging method above, and will not be repeated here.

[0063] In one embodiment, Figure 6 As shown, a plant photosynthetic physiology 3D imaging system is provided, including: a plant color image and chlorophyll fluorescence image acquisition module, a plant structured light stripe image acquisition module, a 3D point cloud acquisition module, a high-precision 3D model generation module and a 3D image and photosynthetic physiology acquisition module, wherein: A plant color image and chlorophyll fluorescence image acquisition module, which is used to switch between the white light source 41 and the blue light source 42, and to respectively acquire the plant color image under the white light source 41 and the plant chlorophyll fluorescence image under the blue light source 42 based on the shooting module 8 and the filter wheel 9; A plant structured light stripe image acquisition module is used to acquire plant structured light stripe images through the shooting module 8; A three-dimensional point cloud acquisition module is used to process plant structured light stripe images and obtain three-dimensional point cloud information based on a three-frequency four-step phase-shift structured light algorithm; A high-precision 3D model generation module is used to extract the two sets of generated 3D point cloud information, perform 3D reconstruction of the plant by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; The three-dimensional image and photosynthetic physiological acquisition module is used to render plant color images and plant chlorophyll fluorescence images into a high-precision three-dimensional model of the plant and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

[0064] Each module in the above system can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module above.

[0065] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a three-dimensional imaging method of plant photosynthetic physiology is implemented.

[0066] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a three-dimensional imaging method of plant photosynthetic physiology is implemented.

[0067] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0068] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0069] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0070] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0071] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0072] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0073] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0074] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A three-dimensional imaging device for plant photosynthetic physiology, characterized in that: The plant photosynthetic physiological three-dimensional imaging device comprises: a dark box (1), a sealing cloth (2) is provided on the dark box (1), a plant rotating tray (3) is provided at one end of the dark box (1) away from the sealing cloth (2), the sealing cloth (2) and the plant rotating tray (3) are used to seal the dark box (1), a light source module (4) and a structured light projection module (5) are provided at one end of the dark box (1) facing the plant rotating tray (3), the light source module (4) comprises a white light source (41) composed of a plurality of white LED lamp beads and a blue light source (42) composed of a plurality of blue LED lamp beads, the white light source (41) and the blue light source (42) are arranged in a staggered manner, the LED lamp beads are provided with a free-form reflective cup (7), at least two shooting modules (8) are provided on the dark box (1), the shooting module (8) is provided with a filter wheel (9), the filter wheel (9) comprises at least three-band filters of RGB, a near-infrared filter and a lightless filter, and the shooting module (8) is connected to a host computer (10).

2. A three-dimensional imaging method for plant photosynthetic physiology, characterized in that: The plant photosynthetic physiology three-dimensional imaging method comprises: Switch between white light source and blue light source, and collect plant color images under white light source and plant chlorophyll fluorescence images under blue light source respectively based on the shooting module and the filter wheel; Collect plant structured light stripe images through a shooting module; Based on the three-frequency four-step phase-shift structured light algorithm, plant structured light stripe images are processed and three-dimensional point cloud information is obtained; Extract the two sets of generated 3D point cloud information, reconstruct the plant in 3D by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; Render plant color images and plant chlorophyll fluorescence images into high-precision three-dimensional plant models and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

3. The plant photosynthetic physiology three-dimensional imaging method according to claim 2, characterized in that: Before the switching of the white light source and the blue light source, the method includes: Initialize an LED staggered array; Presetting particle swarm algorithm parameters, wherein the particle swarm algorithm parameters at least include a range of inertia weight and a learning factor; Calculate the standard deviation of the initial array light intensity distribution corresponding to each particle; Based on the simulated light intensity distribution, the light intensity of each LED lamp bead is superimposed on the target plane according to the distance attenuation formula to obtain the light intensity distribution diagram; Calculate the standard deviation of the initial array light intensity distribution and dynamically adjust the value of the inertia weight w by adopting nonlinear decreasing. The formula of nonlinear decreasing inertia weight is as follows: ； in, is the initial inertia weight, which is the maximum value. is the final inertia weight, which is the minimum value. is the current iteration number, is the maximum number of iterations.

4. The plant photosynthetic physiology three-dimensional imaging method according to claim 3, characterized in that: Initializing an LED staggered array includes: A free-form surface reflector is constructed based on the LED staggered array, wherein the free-form surface reflector comprises a first curved surface and a second curved surface, wherein the first curved surface is used for small-angle light to directly emit to a preset target surface, and the second curved surface is used for large-angle light to be reflected by the inner surface of the reflector to reach the preset target surface; Construct a light source intensity distribution model, the light source intensity distribution model formula is as follows: ； in, is the angle between the light source and the optical axis; The distance from the light source in the direction of the optical axis is The radiance at is the radiation pattern of the light source. When the light source is a perfect ideal Lambertian light source, =1; The non-Lambertian properties are corrected based on the non-ideal light source characteristic correction model. The intensity distribution of the corrected light source is as follows: ； in, is the correction factor, is the angle between the light source and the optical axis; The distance from the light source in the direction of the optical axis is The radiance at is the radiation pattern of the light source; Modeling of free-form reflector, any point on the focal plane of free-form reflector The irradiation intensity is as follows: ； in, is the brightness of the LED chip, in units of ; is the reflective cup area; is the position coordinate of the reflective cup; If exists When the free-form reflectors are illuminated at the same time, the total intensity formula is as follows: = ； The total strength formula of the free-form reflector is as follows: ； in, and are the number of light sources in the rectangular array and the circular array, and The area of ​​the reflector cups for rectangular and circular arrays, and The first The radius and angle of the light source.

5. The plant photosynthetic physiology three-dimensional imaging method according to claim 2, characterized in that: The processing of plant structured light stripe images based on a three-frequency four-step phase-shift structured light algorithm includes: Based on the three-frequency four-step phase shift structured light algorithm, the plant structured light stripe image is processed to generate accurate three-dimensional point cloud information and ; The improved maximum clique 3D registration method is used to align two sets of 3D point cloud information. and Perform registration to produce highly accurate point cloud information; The two network models BIOneFormer3D and RSVDFormer are used to segment and complete the 3D point cloud information respectively to generate a high-precision 3D model of the plant; By fusing plant color images and chlorophyll fluorescence images, high-precision three-dimensional rendering of plants can be achieved, and plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information can be obtained.

6. The three-dimensional imaging method of plant photosynthetic physiology according to claim 5, characterized in that: The improved maximum clique 3D registration method is used to register two sets of 3D point cloud information. and The point cloud information produced by the registration includes: Extract local features of point clouds based on the deep learning model PointNet++; Based on the local features, the graph neural network is inputted for processing, and neighborhood information is aggregated to update the feature representation of each node; Based on the feature representation output by the graph neural network, the quality of the matching point pairs is re-evaluated, and more reliable matching point pairs are screened out to form an initial corresponding set; Construct first-order graph and second-order graph to realize the construction of compatibility graph and search all maximal cliques in the graph based on Bron-Kerbosch algorithm; Sort the maximum clusters based on the preset node weights and intra-cluster consistency, giving priority to clusters with higher weights and better consistency; The singular value decomposition algorithm is used to calculate the transformation hypothesis for the selected group, and two sets of point clouds are realized based on the optimal hypothesis. and of the registration.

7. The plant photosynthetic physiology three-dimensional imaging method according to claim 6, characterized in that: The two network models BIOneFormer3D and RSVDFormer are used to segment and complete the three-dimensional point cloud information respectively, including: BIOneFormer3D specifically includes: sparse 3D U-Net backbone network, super point pooling operation, BI-TransformerDecoder Layer and output layer; The sparse 3D U-Net backbone network includes an encoder and a decoder. The encoder part is composed of multiple layers of sparse convolutional layers to extract higher-level features layer by layer. The features of the encoder are directly transferred to the corresponding layer of the decoder using skip connections to achieve feature fusion; The decoder outputs point-by-point features with the number of channels C. The point-by-point features extracted by the sparse 3D U-Net are aggregated into super-point features through super-point pooling operation; The output super-point features are input into the BI-Transformer Decoder Layer as keys and values, and the learnable semantic queries and instance queries are passed as input to the BI-Transformer Decoder Layer; The RSVDFormer network model specifically includes: input data, feature extraction backbone, feature fusion module, and decoding and upsampling module; The input data includes part of the low-resolution point cloud data and a two-dimensional image generated by multi-view mapping of the 3D point cloud data; The feature extraction backbone is divided into 3D Backbone and 2D Backbone; 3D Backbone uses PointNet++ network to extract features from input 3D point cloud data and generate feature vectors ; 2D Backbone uses the MobileNet V3 network to extract features from two-dimensional images and generate feature vectors ; The feature fusion module constructs a new multi-modal cyclic feature fusion module; The decoding and upsampling module generates a rough point cloud using a 1D convolutional transpose layer and a self-attention layer. , merged with the original point cloud data and resampled to get a rough result ; Generated rough point cloud Entering the detail optimization stage, the SDG module is used for refinement and upsampling, and finally a high-precision complete point cloud is generated. .

8. A three-dimensional imaging system for plant photosynthetic physiology, characterized in that: The system comprises: The plant color image and chlorophyll fluorescence image acquisition module is used to switch between the white light source and the blue light source, and to respectively acquire the plant color image under the white light source and the plant chlorophyll fluorescence image under the blue light source based on the shooting module and the filter wheel; A plant structured light stripe image acquisition module is used to collect plant structured light stripe images through a shooting module; A three-dimensional point cloud acquisition module is used to process plant structured light stripe images and obtain three-dimensional point cloud information based on a three-frequency four-step phase-shift structured light algorithm; A high-precision 3D model generation module is used to extract the two sets of generated 3D point cloud information, perform 3D reconstruction of the plant by registering, segmenting and completing the 3D point cloud information, and generate a high-precision 3D model of the plant; The three-dimensional image and photosynthetic physiological acquisition module is used to render plant color images and plant chlorophyll fluorescence images into a high-precision three-dimensional model of the plant and obtain plant chlorophyll fluorescence three-dimensional images and photosynthetic physiological information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 2 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 2 to 7 are implemented.

Images