Search

CN-121998884-A - Axial resolution enhancement method and system for plant leaf chloroplast three-dimensional imaging

CN121998884ACN 121998884 ACN121998884 ACN 121998884ACN-121998884-A

Abstract

The invention provides an axial resolution enhancement method and an axial resolution enhancement system for plant leaf chloroplast three-dimensional imaging, wherein the method comprises the steps of constructing an in-vitro model sample by using fluorescent microspheres, the in-vitro model sample comprises a non-scattering truth image and an axial stretching distortion image caused by scattering of the same visual field imaging, pre-training a deep learning network by using the in-vitro model sample to establish a pre-training model, constructing a plant true sample of plant leaf chloroplast, training the pre-training model to obtain an axial resolution enhancement model, inputting a chloroplast three-dimensional fluorescent image which is to be processed and is acquired in a complete leaf into the axial resolution enhancement model, and outputting a chloroplast image which is subjected to axial resolution enhancement and is used for accurate three-dimensional reconstruction. According to the invention, hardware is not required to be modified, and the actual three-dimensional morphology of the chloroplast is recovered by specifically correcting the axial deformation of the plant leaf chloroplast in three-dimensional imaging through innovative training data preparation strategy and supervised deep learning.

Inventors

  • HE MUBIN

Assignees

  • 东莞理工学院

Dates

Publication Date
20260508
Application Date
20260128

Claims (10)

  1. 1. The axial resolution enhancement method for plant leaf chloroplast three-dimensional imaging is characterized by comprising the following steps: S1, constructing an in-vitro model sample by using fluorescent microspheres, wherein the in-vitro model sample comprises a non-scattering truth image and an axial stretching distortion image caused by scattering of the same visual field imaging; S2, pre-training the deep learning network by using an in-vitro model sample, and establishing a pre-training model which is mapped from an axial stretching distortion image caused by scattering to a non-scattering true value image S3, constructing a plant real sample of the chloroplast of the plant leaf, wherein the plant real sample comprises a non-scattering truth image of the chloroplast and an axial stretching distortion image caused by scattering of the same visual field imaging, and the scattering condition of the axial stretching distortion image caused by scattering is consistent with that of an in-vitro model sample; S4, training the pre-training model by using a plant true sample to obtain an axial resolution enhancement model; S5, inputting the chloroplast three-dimensional fluorescence image to be processed and collected in the complete blade into the axial resolution enhancement model, and outputting to obtain a chloroplast image for accurate three-dimensional reconstruction after the axial resolution enhancement.
  2. 2. The method for enhancing axial resolution of three-dimensional imaging of plant leaf chloroplasts according to claim 1, wherein in step S1, fluorescent microspheres are used having a size similar to that of chloroplasts, and the diameter of the fluorescent microspheres is within the range of chloroplast diameters.
  3. 3. The method for enhancing axial resolution of three-dimensional imaging of plant leaf chloroplasts according to claim 1, wherein in step S1, the method for obtaining a non-scattering truth image of an in vitro model sample comprises dispersing fluorescent microspheres in a low scattering medium for three-dimensional fluorescent imaging.
  4. 4. The method for enhancing axial resolution of three-dimensional imaging of plant leaf chloroplasts according to claim 3, wherein in step S1, obtaining an image of axial stretching distortion caused by scattering of an in vitro model sample comprises replacing a low scattering medium with a high scattering medium, and performing three-dimensional fluorescence imaging on the same field of view to obtain an image of axial stretching.
  5. 5. The method of claim 4, wherein the high scattering medium is a lipid drop solution having a concentration configured such that the reduced scattering coefficient of the high scattering medium matches the reduced scattering coefficient of the whole plant leaf under study.
  6. 6. The method for enhancing the axial resolution of plant leaf chloroplast three-dimensional imaging according to claim 1, wherein in the step S2, the deep learning network is based on a U-Net coding and decoding architecture, an encoder performs layer-by-layer downsampling on an input axial stretching distortion image caused by scattering, each layer of downsampling feature image is transmitted into a stacked Transformer module, three-dimensional long-range dependence is captured through multi-head attention, global context modeling is enhanced, a CBAM attention module is integrated at a jump joint of the encoder and the decoder, channel weights are generated by channel attention to enhance channel responses of key features, then spatial attention is used for generating spatial weights, a core region is positioned, the double weighted core region is fused with the upsampling feature of the decoder to inhibit scattering noise, and the decoder performs layer-by-layer upsampling on the deep feature processed by the Transformer and is fused with the encoder feature optimized by CBAM to restore the resolution.
  7. 7. The method for enhancing axial resolution of plant leaf chloroplast three-dimensional imaging of claim 6 wherein training of the deep learning network employs a mixed loss function: Wherein Representing the mean square error loss of the signal, Representation of Weight coefficient of (2); Representing a loss of similarity in the multi-scale structure, Representation of Weight coefficient of (2); updating parameters to the cosine annealing learning rate through an Adam optimizer and back propagation iteration The loss converges.
  8. 8. The method for enhancing axial resolution of three-dimensional imaging of plant leaf chloroplasts according to claim 1, wherein the method for constructing a plant true sample of plant leaf chloroplasts in step S3 comprises: Performing epidermis stripping on plant leaves, exposing mesophyll cells, and performing three-dimensional fluorescence imaging under the condition of no epidermis scattering to obtain a non-scattering truth image of chloroplast; And then immersing the plant leaf into the same high scattering medium used in the in-vitro model sample construction, and carrying out three-dimensional fluorescence imaging on the same region to obtain an axial stretching distortion image caused by the scattering of chloroplasts.
  9. 9. Use of the axial resolution enhancement method of three-dimensional imaging of chloroplasts of plant leaves according to any one of claims 1 to 8 for axial resolution enhancement of other spheroid-like intracellular structures in plant leaves marked by fluorescent proteins.
  10. 10. An axial resolution enhancement system for three-dimensional imaging of plant leaf chloroplasts, comprising: An in-vitro model sample module, which is used for constructing an in-vitro model sample by using fluorescent microspheres, wherein the in-vitro model sample comprises a non-scattering truth image and an axial stretching distortion image caused by scattering of the same visual field imaging; the pre-training module is used for pre-training the deep learning network by using the in-vitro model sample and establishing a pre-training model which is mapped from the scattering-induced axial stretching distortion image to the non-scattering truth image The plant real sample module is used for constructing a plant real sample of the chloroplast of the plant leaf, wherein the plant real sample comprises a non-scattering truth image of the chloroplast and an axial stretching distortion image caused by scattering of the same visual field imaging; training the pre-training model by using a plant true sample to obtain an axial resolution enhancement model; And the enhancement module is used for inputting the chloroplast three-dimensional fluorescent image which is to be processed and is acquired in the complete blade into the axial resolution enhancement model and outputting the chloroplast image which is subjected to axial resolution enhancement and is used for accurate three-dimensional reconstruction.

Description

Axial resolution enhancement method and system for plant leaf chloroplast three-dimensional imaging Technical Field The invention belongs to the technical field of computational optical imaging and plant phenotype analysis, and particularly relates to an axial resolution enhancement method and an axial resolution enhancement system for plant leaf chloroplast three-dimensional imaging. Background Plant leaves are key organs for photosynthesis, wherein the three-dimensional structure and spatial distribution of chloroplasts directly determine the photosynthetic efficiency of plants. However, leaf tissue (e.g., epidermis waxy layer, stomatal complex, and deep mesophyll cells) has very strong light scattering and absorption properties, severely limiting imaging depth and resolution of three-dimensional fluorescence microscopy (e.g., three-photon microscopy). When deep imaging is carried out, excitation light and fluorescent signals are subjected to severe scattering, so that the axial resolution of an image is obviously inferior to the transverse resolution, obvious axial stretching phenomenon occurs, and meanwhile, the signal to noise ratio is rapidly reduced along with the depth. This makes the reconstructed chloroplast morphology distorted, spatially distributed information inaccurate, and difficult to perform accurate three-dimensional quantitative analysis. In the prior art, there are hardware solutions and software solutions to the above problems. Hardware solutions such as the computational deconvolution method, which usually assume that the imaging medium is isotropic, cannot effectively cope with tissues such as blades with strongly anisotropic scattering properties, require complex and expensive optical path modifications, and can only be aimed at thin samples, although isotropic resolution can be improved by using 4PI microscopy. Therefore, there is an urgent need in the art for a high-precision computing method capable of specifically correcting axial deformation in three-dimensional imaging of plant leaf chloroplasts without modifying hardware. Disclosure of Invention The invention aims to provide an axial resolution enhancement method and an axial resolution enhancement system for plant leaf chloroplast three-dimensional imaging, which are free from modifying hardware, and can specifically correct axial deformation of plant leaf chloroplast three-dimensional imaging and recover real three-dimensional morphology of chloroplast through innovative training data preparation strategy and supervised deep learning. In order to achieve the above object, the technical scheme of the present invention is as follows: An axial resolution enhancement method for plant leaf chloroplast three-dimensional imaging, comprising: S1, constructing an in-vitro model sample by using fluorescent microspheres, wherein the in-vitro model sample comprises a non-scattering truth image and an axial stretching distortion image caused by scattering of the same visual field imaging; S2, pre-training the deep learning network by using an in-vitro model sample, and establishing a pre-training model which is mapped from an axial stretching distortion image caused by scattering to a non-scattering true value image S3, constructing a plant real sample of the chloroplast of the plant leaf, wherein the plant real sample comprises a non-scattering truth image of the chloroplast and an axial stretching distortion image caused by scattering of the same visual field imaging, and the scattering condition of the axial stretching distortion image caused by scattering is consistent with that of an in-vitro model sample; S4, training the pre-training model by using a plant true sample to obtain an axial resolution enhancement model; S5, inputting the chloroplast three-dimensional fluorescence image to be processed and collected in the complete blade into the axial resolution enhancement model, and outputting to obtain a chloroplast image for accurate three-dimensional reconstruction after the axial resolution enhancement. Further, in step S1, the size of the fluorescent microsphere used is similar to that of chloroplast, and the diameter of the fluorescent microsphere is within the diameter range of chloroplast. In step S1, the method for obtaining the non-scattering truth image of the in-vitro model sample comprises dispersing fluorescent microspheres in a low-scattering medium, and performing three-dimensional fluorescence imaging. Further, in step S1, the obtaining of the axial stretching distortion image caused by the scattering of the in vitro model sample comprises the steps of replacing a low scattering medium with a high scattering medium, and carrying out three-dimensional fluorescence imaging on the same visual field to obtain the axial stretching image. Preferably, the high scattering medium is a lipid drop solution, the concentration of which is configured such that the reduced scattering coefficient of the high scattering medium matc