Plant leaf stomatal phenotype nondestructive acquisition method based on microscopic hyperspectral super-resolution

Abstract

The invention discloses a plant leaf stomatal phenotype nondestructive acquisition method based on microscopic hyperspectral super-resolution. The method comprises the following steps: step 1, collecting a low-resolution microscopic hyperspectral image of a plant living leaf; 2, carrying out the preprocessing of the low-resolution microscopic hyperspectral image; step 3, inputting the preprocessed low-resolution image into a pre-trained super-resolution reconstruction model, and outputting a high-fidelity high-resolution hyperspectral image; step 4, based on the high-resolution hyperspectral image obtained by reconstruction, performing threshold segmentation by using optical characteristics of plant tissues in a characteristic wave band to obtain a binary mask of pores; and step 5, performing connected domain analysis and ellipse fitting on the binarized mask, and calculating pore opening and pore area microscopic phenotype parameters. The problems that in the prior art, the spatial resolution is insufficient, the spectral fidelity is poor, and dynamic monitoring of a living body cannot be achieved are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of plant phenomics, and particularly relates to a plant leaf stomata phenotype nondestructive acquisition method based on microscopic hyperspectral super resolution. BACKGROUND

[0002] Plant stomata are micron-sized pores formed by a pair of guard cells on the epidermis of plant leaves, and are important channels for the exchange of H2O, CO2 and O2 between plants and the outside world, affecting physiological activities such as photosynthesis, respiration and transpiration of plants, and further affecting the yield and quality of plants. Therefore, exploring the plant stomata phenotype (such as stomata opening, area, long and short axes and stomata conductance) has important significance for in-depth analysis of its role in plant growth and environmental response, and elucidation of the regulation mechanism of stomata on plant growth and development. Current plant stomata phenotype acquisition methods are mainly based on leaf microscopic image acquisition for phenotype extraction, which can be divided into destructive sampling and non-destructive sampling. Commonly used destructive leaf stomata image acquisition methods such as hand tearing and imprinting rely on manual measurement, are time-consuming and prone to errors. Non-destructive leaf stomata observation methods such as in-situ observation use portable microscopes without sample preparation, and have high imaging efficiency, but the image quality obtained by this method is low. In current technology, the destructive method can only obtain static morphological phenotype, and the non-destructive in-situ observation cannot be associated with physiological function, so it is necessary to use microscopic hyperspectral imaging to realize the synchronous in-vivo acquisition of morphological-spectral-physiological multi-dimensional phenotype.

[0003] In recent years, microscopic hyperspectral imaging has made it possible to measure plant microphenotypes. However, due to the bottleneck of optical system light flux, its spatial resolution is often lower than that of the same magnification visible light microscope, resulting in that the stomata opening cannot be identified. If only two-dimensional visible light super resolution network (such as RCAN, EDSR) is used, although the spatial clarity can be improved, the reconstructed image spectral curve is distorted, which cannot meet the physiological quantitative requirements. Therefore, there is an urgent need for a nondestructive detection method that can simultaneously obtain high-definition morphological structure and high-fidelity spectral information of plant leaf stomata. SUMMARY

[0004] The present application aims to provide a plant leaf stomata phenotype nondestructive acquisition method based on microscopic hyperspectral super resolution. The present application breaks through the physical limit of hardware imaging through algorithm, reconstructs low-resolution microscopic hyperspectral images into high-resolution images, and realizes accurate segmentation and parameter calculation of stomata by using spectral specificity, solving the problems of insufficient spatial resolution, poor spectral fidelity and inability to realize in-vivo dynamic monitoring in the prior art.

[0005] The technical scheme of the present application: a plant leaf stomata phenotype nondestructive acquisition method based on microscopic hyperspectral super resolution, comprising the following steps: Step 1, collecting low-resolution microscopic hyperspectral images of plant in-vivo leaves; Step 2, preprocessing the low-resolution microscopic hyperspectral image; Step 3, inputting the preprocessed low-resolution image into a pre-trained super-resolution reconstruction model to output a high-fidelity high-resolution hyperspectral image; Step 4, based on the reconstructed high-resolution hyperspectral image, using the optical properties of plant tissues at characteristic wavelengths to perform threshold segmentation to obtain a binary mask of stomatal pores; Step 5, performing connected component analysis and ellipse fitting on the binary mask to calculate stomatal aperture and stomatal area microphenotype parameters.

[0006] The above-mentioned plant leaf stomatal phenotype non-destructive acquisition method based on microscopic hyperspectral super-resolution, in step 1, when collecting the low-resolution microscopic hyperspectral image, select healthy, undamaged and avoid the main vein of the leaf, clean and lay on the stage; use the microscopic hyperspectral imaging system to perform reflection push-broom imaging in the spectral range of 400-1000nm, and use a standard white plate for reflectance normalization before collection.

[0007] The aforementioned plant leaf stomatal phenotype non-destructive acquisition method based on microscopic hyperspectral super-resolution, in step 2, the preprocessing includes: Savitzky-Golay smoothing filtering on the collected low-resolution microscopic hyperspectral image, and intercepting the effective spectral interval of 470-880nm; then using sliding window cropping to obtain training image blocks, and constructing low-resolution-high-resolution image pairs for model training.

[0008] The aforementioned plant leaf stomatal phenotype non-destructive acquisition method based on microscopic hyperspectral super-resolution, in step 3, the super-resolution reconstruction model is a PhytoSR network based on decoupled parallel architecture, which extracts spatial geometric features and spectral features through parallel 2D texture enhancement branch and 3D space-spectrum joint branch, and finally performs feature fusion through an adaptive fusion module to complete image reconstruction.

[0009] The aforementioned plant leaf stomatal phenotype non-destructive acquisition method based on microscopic hyperspectral super-resolution, the 2D texture enhancement branch includes a stacked Res-MDTA module and a cascaded MPAE module; the Res-MDTA module combines residual structure and multi-head transpose attention mechanism for shallow feature extraction, and the MPAE module uses three parallel strategies of compression distillation, information preservation and attention enhancement to enhance and sharpen the features.

[0010] The aforementioned plant leaf stomatal phenotype nondestructive acquisition method based on microscopic hyperspectral super-resolution, the 3D space spectrum joint branch successively includes a convolution module for feature projection, a cascaded SM3D module, a bottleneck layer and an up-sampling module; the SM3D module integrates a bidirectional state space model for bidirectional global scanning of a spectral band sequence to capture long-distance spectral dependence.

[0011] The aforementioned plant leaf stomatal phenotype nondestructive acquisition method based on microscopic hyperspectral super-resolution, in step 4, the threshold segmentation using the optical characteristics of plant tissues at characteristic bands is specifically: selecting a single-band image at a 550nm band, identifying regions with reflectivity lower than 0.5 as stomatal pores, and generating a binary mask.

[0012] The aforementioned plant leaf stomatal phenotype nondestructive acquisition method based on microscopic hyperspectral super-resolution, in step 5, the calculation of stomatal opening is specifically: elliptical fitting of stomatal pores by a least square method, and taking the short axis length of the fitted ellipse as the stomatal opening; the calculation of stomatal area is specifically: calculating the total number of pixels in the connected domain, and converting into a physical area according to the spatial resolution of the imaging system.

[0013] Compared with the prior art, the present application has the following beneficial effects: 1. The present application does not need to tear, dye and other destructive treatment of plant leaves, and realizes in-situ nondestructive detection of plant leaf stomatal phenotype through in-vivo microscopic hyperspectral image acquisition and super-resolution reconstruction technology. It not only avoids the damage of traditional destructive methods to plant samples, but also continuously and real-time captures the morphological and physiological dynamic changes of stomata in the growth process.

[0014] 2. The present application synchronously reconstructs high-resolution morphological structure and high-fidelity spectral information through the PhytoSR network, closely combines the microscopic morphology (opening, area) of stomata with the spectral physiological characteristics, and provides technical support for in-depth analysis of stomatal structure and plant growth.

[0015] 3. The present application proposes a decoupled parallel architecture PhytoSR network. Through a 2D texture enhancement branch, spatial geometric features are accurately mined, a 3D space spectrum joint branch efficiently captures long-distance spectral dependence, and self-adaptive fusion and hybrid loss function constraints are combined to break through the hardware light flux limit, improve the image spatial resolution, ensure the high fidelity of the spectral curve, and meet the needs of stomatal phenotype quantitative analysis.

[0016] 4. The present application greatly reduces the time consumption and human error of traditional methods, significantly improves the detection efficiency, and has wide applicability and promotional value. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a step flowchart of the present application; Figure 2 is a structural diagram of PhytoSR network; Figure 3 is a structural diagram of MPAE module; Figure 4 is a structural diagram of SM3D module; Figure 5 is a schematic diagram of high-resolution hyperspectral image; Figure 6 is a schematic diagram of spectral reflectance curve.

[0018] Figure 7 is a schematic diagram of binary mask. DETAILED DESCRIPTION

[0019] The application will be further described below in conjunction with the drawings and examples, but not as the basis for limiting the application.

[0020] Example: Non-destructive acquisition method of plant leaf stomatal phenotype based on microscopic hyperspectral super-resolution.

[0021] The application is based on microscopic hyperspectral super-resolution technology to realize non-destructive acquisition of plant leaf stomatal phenotype, and the core is completed through an end-to-end process of image acquisition, data preprocessing, super-resolution reconstruction, stomatal segmentation and parameter calculation. The application will be described in detail below in conjunction with specific equipment, parameters and operation steps.

[0022] I. Experimental preparation and device configuration (I) Experimental materials Select healthy, disease-free, and mechanically undamaged plant leaves, and preferentially avoid the main vein area (stomata are sparsely distributed near the main vein and are easily blocked). In this embodiment, tomato leaves are used as the research object, and other dicotyledonous plants (such as peppers, cucumbers) or monocotyledonous plants (such as corn, rice) can be directly adapted without adjusting the core process.

[0023] (II) Core equipment and software Microscopic hyperspectral imaging system: FHY-MICROHIS-VNIR10H-OP01 developed by Xi'an Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, matched with transmission reflection metallographic microscope L3230; Image acquisition software: MicroSpectrograph version V2.3 (matched with the above imaging system); Standard white board: reflectivity is 99%, used for reflectivity normalization calibration; Computing device: workstation equipped with NVIDIA GTX-A4000 (16 GB video memory) graphics card, supporting CUDA acceleration environment; Deep learning framework and dependent library: build the model based on PyTorch framework, and match the data processing library such as NumPy and OpenCV.

[0024] II. Detailed implementation steps, such as Figure 1 Step 1, collect low-resolution microscopic hyperspectral images of plant live leaves, specifically including: 1.1, leaf pretreatment: gently wipe the leaf surface with a dust-free paper towel to remove dust, water stains and other impurities, and avoid affecting the spectral reflectance and imaging quality; place the cleaned leaf on the microscope stage to ensure that the leaf surface is wrinkle-free and completely adheres to the stage, and use a transparent pressing sheet to fix it (avoid damaging the leaf tissue or squeezing the stomata).

[0025] 1.2, imaging system calibration: before image acquisition, place a standard white board at the original leaf position on the stage, and collect the hyperspectral image of the white board through MicroSpectrograph V2.3 software, and perform reflectance normalization processing based on this as a reference to eliminate the interference of factors such as light source fluctuation and optical system attenuation.

[0026] 1.3, imaging parameter setting and image acquisition: start the imaging system and configure the parameters as follows: imaging mode is reflective push-broom imaging, spectral range is 400-1000nm, number of bands is 252, spectral resolution is ≤5nm (center wavelength error ±1nm), and spatial resolution is 600×700 pixels. After parameter configuration is completed, move back the leaf and keep the position unchanged, start push-broom imaging, and collect low-resolution microscopic hyperspectral images (i.e. low-resolution hyperspectral cube).

[0027] Step 2, pre-process the low-resolution microscopic hyperspectral image; specifically including: 2.1, spectral smoothing and effective interval cutting: use Savitzky-Golay smoothing filter algorithm to process the collected low-resolution microscopic hyperspectral image, set the smoothing window size to 11 and the polynomial order to 3, and eliminate spectral noise; then cut off the effective spectral interval of 470-880nm, and eliminate the edge redundant bands.

[0028] 2.2, training image block cropping: use sliding window method to crop the processed low-resolution microscopic hyperspectral image, set the cropping window size to 256×256 pixels, and the window overlap degree to 5%, crop along the image plane from top to bottom and from left to right, ensure that there is enough spatial redundancy between adjacent windows, and finally get 576 training image blocks.

[0029] ​2.3, Constructing low-resolution-high-resolution image pairs: using a bicubic interpolation algorithm to down-sample the 256x256 pixel training image block by 4, obtaining a 64x64 pixel low-resolution image block, forming a one-to-one low-resolution-high-resolution image pair, which is used for training the super-resolution reconstruction model.

[0030] Step 3, input the pre-processed low-resolution image into the pre-trained super-resolution reconstruction model, output high-fidelity high-resolution hyperspectral image; specific including: 3.1, construct a super-resolution reconstruction model based on decoupled parallel architecture (named PhytoSR network), which includes parallel 2D texture enhancement branch, 3D space-spectrum joint branch and adaptive fusion module, as shown in Figure 2 , the specific structure is as follows: 2D texture enhancement branch: including stacked Res-MDTA module and cascaded MPAE module (multi-path attention enhancement module) in turn. Among them, Res-MDTA module combines residual structure and multi-head transpose attention mechanism, which is used for shallow feature extraction; MPAE module adopts three parallel strategies of compression distillation, information preservation and attention enhancement to enhance and sharpen the features, and improve the accuracy of spatial geometric feature extraction. Specifically, the initial features are extracted through the Conv layer, and then the spatial texture features are mined through 6 stacked Res-MDTA modules (residual and multi-head transpose attention), and then processed in two ways: one way is directly up-sampled (Upsample) to get shallow spatial features (weight w1); the other way is to sharpen the details through 2 MPAE modules (multi-path attention enhancement) and then up-sample to get deep spatial features (weight w2).

[0031] The module results are shown in Figure 3 , which enhances spatial details through three parallel strategies: Distill Path (compression distillation): compressing channels with Conv+MDTA and preserving core features; Remain Path (information preservation): directly pass the original features to avoid information loss; Enhance Path (attention enhancement): use Conv+MDTA to enhance high-frequency details (such as stomatal edges); The three features are added, spliced (Concat), and convolved to output, achieving accurate enhancement of spatial texture.

[0032] Spectrum-Convolution Branch: sequentially includes a convolution module for feature projection, a cascaded SM3D module (Spectrum Mamba 3D module), a bottleneck layer, and an up-sampling module. Among them, the SM3D module integrates a bidirectional state space model for bidirectional global scanning of the spectrum band sequence to capture long-distance spectral dependence; the input features are projected to a 3D space through the convolution module, processed by the SM3D module, and then reduced in channel dimension and fused by the bottleneck layer, and then projected back to a 2D space by the up-sampling module and the convolution. Specifically, the 2D image is first projected to a 3D spectrum space by the Conv2D to 3D module, then the long-distance spectral dependence across the bands is captured by the 4 SM3D modules (Spectrum Mamba 3D), and then the features are concatenated (Concat) and reduced by the Bottleneck layer, and then up-sampled and projected back to the 2D space by the Conv3D to 2D to obtain high-fidelity spectral features (weight w3).

[0033] The structure and principle of the SM3D module are shown in Figure 4 The spatial / spectral local features are extracted by Spatial / Spectral Conv (spatial-spectral convolution module) respectively; the spatial / spectral information is fused by Conv Mix&CA and the channel attention is added; the spatial correlation is strengthened by MDTA (multi-head transposed attention) and the long-distance spectral dependence is captured by Mamba (state space model); finally, the residual connection is output to realize the dual optimization of spatial details and spectral fidelity.

[0034] Adaptive Fusion Module: dynamically integrates the spatial geometric features extracted by the 2D texture enhancement branch and the spectral features extracted by the 3D spectrum-convolution branch, and outputs a high-fidelity high-resolution hyperspectral image. Specifically, the 3 groups of features (w1 / w2 / w3) output by the double branches are adaptively weighted and fused, and finally a high-resolution hyperspectral image (SR) is output; at the same time, the residual connection of the original LR feature with "Bicubic Upsample" is added to further improve the reconstruction accuracy.

[0035] 3.2, Model training parameter configuration: Loss function: a hybrid loss function L total =L1+λL SAM , wherein the L1 loss is used to constrain the reconstruction accuracy of the spatial geometric edge, and the L SAM (spectral angle mapping) loss is used to constrain the waveform consistency of the reconstructed spectrum and the real spectrum, and the weight coefficient λ is set to 0.3.

[0036] Optimizer: Adam algorithm is adopted, momentum factor β1=0.9, β2=0.999, and weight decay coefficient is 1×10 -5 .

[0037] Learning rate strategy: the initial learning rate is 1 x 10 -4 , and a multi-step decay strategy is adopted, with a learning rate decay of 0.5 every 50 epochs, and a total training epoch number of 200.

[0038] Training process: the low-resolution-high-resolution image pairs constructed in step 2 are divided into training set and validation set according to the conventional ratio, and the model training is performed with a batch size of 8, and the training loss and validation loss are monitored in real time. When the validation loss does not decrease for 20 consecutive epochs, the training is stopped, and the pre-trained super-resolution reconstruction model is saved.

[0039] 3.3, high-resolution hyperspectral image output: input the low-resolution image (complete low-resolution hyperspectral cube) preprocessed in step 2 into the above pre-trained super-resolution reconstruction model, and obtain a high-fidelity high-resolution hyperspectral image through model inference. When a low-resolution image of 256x256 pixels is input, the resolution of the output high-resolution hyperspectral image is 1024x1024 pixels. As shown in Figure 5 , Figure 5 The red and green areas in the figure respectively mark the two target areas of stomatal pores and guard cells, which are used for subsequent spectral feature extraction and comparison.

[0040] Step 4, based on the reconstructed high-resolution hyperspectral image, the threshold segmentation is performed on the optical properties of plant tissues at the characteristic wavelength to obtain the binary mask of stomatal pores. In this step, the optical properties of plant tissues are analyzed based on the high-resolution hyperspectral image reconstructed in step 3: at the characteristic wavelength of 550 nm (green wavelength), the guard cells are rich in chlorophyll and water, showing high reflectivity (Reflectance> 0.5), while the stomatal pores show low reflectivity (Reflectance<0.5) due to the light trapping effect of the physical cavity structure, as shown in Figure 6 . Figure 6 The spectral reflectivity curves (horizontal coordinate: wavelength, vertical coordinate: reflectivity) of the red (stomatal pores) and green (guard cells) areas are shown: the difference between the two curves at different wavelengths is obvious (such as the different jump amplitudes of reflectivity near 700 nm), which reflects the difference in optical properties between stomatal pores (physical cavity) and guard cells (containing chlorophyll / water), which is the basis for subsequent segmentation of stomata based on spectral features.

[0041] Based on this characteristic, threshold segmentation is performed, and the specific operation is as follows: select the single-band image at 550 nm wavelength, set the reflectivity threshold T=0.5, identify the area with reflectivity less than 0.5 as stomatal pores and mark it as white; mark the area with reflectivity≥0.5 as background and mark it as black, and finally obtain the binary mask of stomatal pores. Figure 7is based on Figure 6 The binary mask (black background and white stomata target) after threshold segmentation of the hyperspectral image, and the red circle is the identified "stomatal pore area", corresponding to the red labeled area in Figure 5 , which verifies the effectiveness of stomatal segmentation based on spectral features.

[0042] Step 5, performing connected component analysis and ellipse fitting on the binary mask to calculate the stomatal opening and stomatal area microphenotype parameters; specifically including: 5.1, connected component analysis: using computer image processing algorithm to perform connected component analysis on the binary mask obtained in step 4, extracting the connected component corresponding to each independent stomata, eliminating small noise connected components (to avoid interference with the measurement results), and ensuring accurate extraction of independent stomata targets.

[0043] 5.2, ellipse fitting: using least squares method to perform ellipse fitting on the connected component of each independent stomata, obtaining the geometric parameters of the fitted ellipse such as the short axis length and the long axis length.

[0044] 5.3, microphenotype parameter calculation: according to the spatial resolution calibration value (μm / pixel) of the microscopic imaging system, perform physical parameter conversion: Stomatal opening: taking the short axis length of the fitted ellipse as the stomatal opening, and combining the spatial resolution calibration value, converting it into the corresponding physical length; Stomatal area: counting the total number of pixels within each stomata connected component, and combining the spatial resolution calibration value, converting it into the corresponding physical area.

[0045] Through the above steps, the complete process from collecting low-resolution microscopic hyperspectral images of plant live leaves to calculating stomatal opening and stomatal area microphenotype parameters is completed, realizing non-destructive acquisition of plant leaf stomatal phenotype.

[0046] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A non-destructive method for obtaining plant leaf stomatal phenotypes based on microscopic hyperspectral super-resolution, characterized in that: Includes the following steps: Step 1: Acquire low-resolution microscopic hyperspectral images of living plant leaves; Step 2: Preprocess the low-resolution microscopic hyperspectral image; Step 3: Input the preprocessed low-resolution image into the pre-trained super-resolution reconstruction model to output a high-fidelity, high-resolution hyperspectral image; Step 4: Based on the reconstructed high-resolution hyperspectral image, threshold segmentation is performed using the optical properties of plant tissue in the characteristic bands to obtain a binarized mask of stomatal pores. Step 5: Perform connected component analysis and ellipse fitting on the binarized mask to calculate the micro-phenotypic parameters of pore aperture and pore area.

2. The method for non-destructive acquisition of plant leaf stomatal phenotypes based on microscopic hyperspectral super-resolution as described in claim 1, characterized in that: In step 1, when acquiring the low-resolution microscopic hyperspectral image, healthy, undamaged leaves that avoid the midrib are selected, cleaned, and laid flat on the stage; a microscopic hyperspectral imaging system is used to perform reflectance push-broom imaging in the 400-1000nm spectral range, and a standard white board is used to normalize the reflectance before acquisition.

3. The method for non-destructive acquisition of plant leaf stomatal phenotypes based on microscopic hyperspectral super-resolution as described in claim 1, characterized in that: In step 2, the preprocessing includes: performing Savitzky-Golay smoothing filtering on the acquired low-resolution microscopic hyperspectral images and truncating the effective spectral range of 470-880nm; then using a sliding window to crop to obtain training image patches and constructing low-resolution-high-resolution image pairs for model training.

4. The method for non-destructive acquisition of plant leaf stomatal phenotypes based on microscopic hyperspectral super-resolution as described in claim 1, characterized in that: In step 3, the super-resolution reconstruction model is a PhytoSR network based on a decoupled parallel architecture. The PhytoSR network extracts spatial geometric features and spectral features through parallel 2D texture enhancement branches and 3D spatial-spectral joint branches, respectively, and finally performs feature fusion through an adaptive fusion module to complete image reconstruction.

5. The method for non-destructive acquisition of plant leaf stomatal phenotype based on microscopic hyperspectral super-resolution as described in claim 4, characterized in that: The 2D texture enhancement branch includes stacked Res-MDTA modules and cascaded MPAE modules. The Res-MDTA module combines residual structure and multi-head transposed attention mechanism to extract shallow features, and the MPAE module uses a three-way parallel strategy of compression distillation, information preservation and attention enhancement to enhance and sharpen features.

6. The method for non-destructive acquisition of plant leaf stomatal phenotypes based on microscopic hyperspectral super-resolution as described in claim 4, characterized in that: The 3D spatial-spectral joint branch sequentially includes a convolution module for feature projection, a cascaded SM3D module, a bottleneck layer, and an upsampling module; the SM3D module integrates a bidirectional state-space model for bidirectional global scanning of spectral band sequences to capture long-range spectral dependencies.

7. The method for non-destructive acquisition of plant leaf stomatal phenotype based on microscopic hyperspectral super-resolution as described in claim 1, characterized in that, In step 4, the threshold segmentation using the optical properties of plant tissue in the characteristic band is specifically as follows: select a single-band image of 550nm band, identify areas with reflectivity below 0.5 as stomata, and generate a binarized mask.

8. The method for non-destructive acquisition of plant leaf stomatal phenotype based on microscopic hyperspectral super-resolution as described in claim 1, characterized in that, In step 5, the calculation of pore opening specifically involves: fitting an ellipse to the pore size using the least squares method, and taking the length of the minor axis of the fitted ellipse as the pore opening; the calculation of pore area specifically involves: calculating the total number of pixels in the connected region, and converting it into physical area according to the spatial resolution calibration of the imaging system.

Images