Method, device and electronic equipment for detecting cross-sectional structure of vascular plant stems

Abstract

The present invention provides a method, device, and electronic device for detecting the cross-sectional structure of a vascular plant stem. The method comprises: obtaining a fiber sheath prediction frame, fiber sheath positioning coordinates, a duct prediction frame, and duct positioning coordinates in the target cross-sectional image based on a target cross-sectional image and a trained vascular plant stem cross-sectional structure prediction model; and mapping the fiber sheath prediction frame and duct prediction frame to the target cross-sectional image based on the fiber sheath positioning coordinates and duct positioning coordinates; obtaining a fiber sheath image and a duct image of the target cross-sectional image based on the fiber sheath prediction frame and the duct prediction frame by a binarization method; and calculating the fiber sheath area and duct area of ​​the vascular plant stem based on the fiber sheath image, the duct image, and the image resolution. Thus, the cross-sectional structure of vascular plant stems can be quickly and accurately measured in large quantities of samples, large-sized samples, and within a large observation area.

Description

Technical Field

[0001] The present invention relates to the technical field of plant detection, and in particular to a method, device and electronic equipment for detecting the cross-sectional structure of a vascular plant stem. Background Art

[0002] Using cross-sectional structural images of vascular plant stems to quantitatively analyze anatomical features such as fibers, vessels, and parenchyma is a huge challenge faced by scientists in plant science.

[0003] Observing the cross-sectional structure of vascular plant stems typically involves conventional microscopic techniques, including optical microscopy, confocal laser scanning microscopy, scanning electron microscopy, and X-ray tomography. However, these methods struggle to rapidly and accurately determine the number and distribution of fiber sheaths (strands) and vessels in large samples, over large areas, and within a wide observation area. Summary of the Invention

[0004] The present invention provides a method, device and electronic equipment for detecting the cross-sectional structure of vascular plant stems, which are used to solve the defect in the prior art that it is difficult to quickly and accurately measure the cross-sectional structure of vascular plant stems, and realize the rapid and accurate measurement of the cross-sectional structure of vascular plant stems in large batches of samples, large-size samples and large observation areas.

[0005] The present invention provides a method for detecting the cross-sectional structure of a vascular plant stem, comprising:

[0006] According to the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinate, a duct prediction frame, and a duct positioning coordinate in the target cross-sectional image are obtained, and the fiber sheath prediction frame and the duct prediction frame are mapped to the target cross-sectional image according to the fiber sheath positioning coordinates and the duct positioning coordinates;

[0007] According to the fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method;

[0008] Calculating the fiber sheath area and the duct area of ​​the vascular plant stem according to the fiber sheath image, the duct image, and the image resolution;

[0009] The vascular plant stem cross-section structure prediction model is trained based on sample cross-section images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels, and duct positioning coordinate labels.

[0010] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, the cross-sectional structure prediction model of the vascular plant stem is trained according to the following method:

[0011] Segmenting the sample cross-sectional image into a plurality of sample sub-images, and synchronously segmenting the fiber sheath prediction frame label, the fiber sheath positioning coordinate label, the catheter prediction frame label, and the catheter positioning coordinate label in the sample cross-sectional image into the plurality of sample sub-images;

[0012] After the multiple sample subgraphs are subjected to data enhancement processing, they are sequentially passed through a target detection model with pre-trained weights, and trained in a freezing stage and a thawing stage to obtain the vascular plant stem cross-section structure prediction model.

[0013] According to a method for detecting the cross-sectional structure of a vascular plant stem provided by the present invention, the method of obtaining a fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame, and duct positioning coordinates in the target cross-sectional image based on a target cross-sectional image and a trained vascular plant stem cross-sectional structure prediction model specifically includes:

[0014] generating a plurality of cross-sectional sub-images according to the target cross-sectional image; wherein the size of the cross-sectional sub-images is consistent with the size of the sample sub-image;

[0015] According to the multiple cross-sectional subgraphs and the trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinate, a duct prediction frame, and a duct positioning coordinate in the multiple cross-sectional subgraphs are obtained;

[0016] The fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the multiple cross-sectional sub-images are merged to obtain the fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the target cross-sectional image.

[0017] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, the method further comprises:

[0018] Before generating a plurality of cross-sectional sub-images according to the target cross-sectional image, presetting overlapping areas of adjacent cross-sectional sub-images in the plurality of cross-sectional sub-images; and

[0019] Before merging the fiber sheath prediction frames, fiber sheath positioning coordinates, catheter prediction frames and catheter positioning coordinates in the multiple cross-sectional sub-images, the repeated fiber sheath prediction frames, repeated fiber sheath positioning coordinates, repeated catheter prediction frames and repeated catheter positioning coordinates in the multiple cross-sectional sub-images are removed based on the area of ​​the overlapping region.

[0020] According to a method for detecting cross-sectional structures of vascular plant stems provided by the present invention, before generating a plurality of cross-sectional sub-images based on the target cross-sectional image, the method further comprises:

[0021] Performing RGB conversion on the target cross-sectional image and adding gray bars before resizing;

[0022] An iterative sample size dimension is added to the resized target cross-sectional image.

[0023] According to a method for detecting a cross-sectional structure of a vascular plant stem provided by the present invention, obtaining a fiber sheath image and a duct image of the target cross-sectional image by a binarization method based on the fiber sheath prediction frame and the duct prediction frame specifically includes:

[0024] filtering the fiber sheath prediction frame and the catheter prediction frame;

[0025] According to the filtered fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method.

[0026] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, the method further comprises:

[0027] After obtaining the fiber sheath image and the catheter image of the target cross-sectional image, visualization processing is performed on the fiber sheath image and the catheter image.

[0028] The present invention also provides a device for detecting the cross-sectional structure of a vascular plant stem, comprising:

[0029] a prediction module for obtaining, based on a target cross-sectional image and a trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinates, a duct prediction frame, and a duct positioning coordinates in the target cross-sectional image, and mapping the fiber sheath prediction frame and the duct prediction frame to the target cross-sectional image based on the fiber sheath positioning coordinates and the duct positioning coordinates;

[0030] An image module, configured to obtain a fiber sheath image and a catheter image of the target cross-sectional image by a binarization method according to the fiber sheath prediction frame and the catheter prediction frame;

[0031] a calculation module, configured to calculate the fiber sheath area and the duct area of ​​the vascular plant stem based on the fiber sheath image, the duct image, and the image resolution;

[0032] The vascular plant stem cross-section structure prediction model is trained based on sample cross-section images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels, and duct positioning coordinate labels.

[0033] The present invention also provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above methods for detecting the cross-sectional structure of vascular plant stems is implemented.

[0034] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for detecting the cross-sectional structure of a vascular plant stem as described above is implemented.

[0035] The present invention provides a method, device and electronic equipment for detecting the cross-sectional structure of a vascular plant stalk. A cross-sectional structure prediction model of the vascular plant stalk is obtained by training based on sample cross-sectional images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels and duct positioning coordinate labels. Based on the target cross-sectional image and the trained cross-sectional structure prediction model of the vascular plant stalk, the fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame and duct positioning coordinates in the cross-sectional image are obtained, and the fiber sheath prediction frame and duct prediction frame are mapped to the cross-sectional image. Then, a fiber sheath image and duct image corresponding to the cross-sectional image are obtained by a binarization method. Finally, based on the fiber sheath image, duct image and image resolution, the fiber sheath area and duct area of ​​the vascular plant stalk are calculated, thereby being able to quickly and accurately measure the cross-sectional structure of the vascular plant stalk in large batches of samples, large-size samples and large observation areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 A schematic diagram of the process of the method for detecting the cross-sectional structure of a vascular plant stem provided by the present invention;

[0038] Figure 2 A schematic diagram of the relative position relationship between the target frame and the sub-image provided by the present invention;

[0039] Figure 3 A schematic structural diagram of a device for detecting cross-sectional structure of a vascular plant stem provided by the present invention;

[0040] Figure 4 This is a schematic structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0042] Vascular plants, including bamboo, rattan, and reeds, are characterized by their fibers, vessels, and parenchyma, key anatomical structures. These structures are interconnected: the fiber cavity, with its thin, thin walls, provides mechanical support; the vessel cavity, with its large, thin walls, transports liquids like water; and the parenchyma provides filling and bonding. Vascular plants experience macroscopic shrinkage and swelling in environments with fluctuating humidity. This is due to the fibers, vessels, and parenchyma contracting or expanding in response to changes in humidity. Vascular plants deform or even fail under external forces. At the microscopic level, their internal fibers act as a reinforcing skeleton, the vessel cavities provide space for deformation, and the parenchyma offers excellent compressive resistance. These three factors work together to give vascular plants both high strength and toughness. Therefore, quantitatively analyzing anatomical features such as fibers, vessels, and parenchyma using cross-sectional structural images of vascular plant stems is a significant challenge facing scientists in plant science.

[0043] Conventional microscopic techniques mostly use manual counting methods to measure the number or distribution of vascular bundles in small-sized samples or small observation areas. This makes it difficult to quickly and accurately measure the number and distribution of fiber sheaths (strands) and ducts in large batches of samples, large-sized samples, and large observation areas.

[0044] Figure 1 The schematic diagram of the process of the vascular plant stem cross-section structure detection method provided by the present invention is as follows: Figure 1 As shown, the method includes the following steps:

[0045] Step 100: Based on the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, the fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame and duct positioning coordinates in the target cross-sectional image are obtained, and the fiber sheath prediction frame and duct prediction frame are mapped to the target cross-sectional image based on the fiber sheath positioning coordinates and duct positioning coordinates.

[0046] Step 101: Obtain a fiber sheath image and a catheter image of a target cross-sectional image by a binarization method based on the fiber sheath prediction frame and the catheter prediction frame.

[0047] Step 102: Calculate the fiber sheath area and the vessel area of ​​the vascular plant stem based on the fiber sheath image, the vessel image, and the image resolution.

[0048] Among them, the vascular plant stem cross-sectional structure prediction model is trained based on sample cross-sectional images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels and duct positioning coordinate labels.

[0049] Specifically, in an embodiment of the present invention, the target cross-sectional image is a cross-sectional image of a vascular plant stem whose fiber sheath (strand) and duct number and distribution need to be measured, and the sample cross-sectional image is a cross-sectional image of a vascular plant stem with a fiber sheath prediction frame label, a fiber sheath positioning coordinate label, a duct prediction frame label, and a duct positioning coordinate label. It should be noted that the fiber sheath in the present invention can also be represented as a fiber strand, or a fiber sheath and a fiber strand. This concept remains consistent throughout the text and will not be repeated hereafter.

[0050] In the embodiments of the present invention, the target cross-sectional image and the sample cross-sectional image are cross-sectional images in which the boundaries between the fiber sheath (strand), vascular duct, and parenchyma tissue can be clearly distinguished by the naked eye. Such cross-sectional images can be obtained by cutting a vascular plant stem along the transverse direction and polishing the cross-sectional image using a blade (e.g., a Leica E0997 blade).

[0051] In some embodiments, before predicting the target cross-sectional image or training based on the sample cross-sectional image, the target cross-sectional image or the sample cross-sectional image can be ultrasonically processed using ultrasound to remove impurities (such as dust, powder, etc.) inside the cell cavity, for example, by performing ultrasonic processing with an ultrasonic frequency of 100-200W and an ultrasonic time of 5-10 minutes.

[0052] The target cross-sectional image and the sample cross-sectional image can be obtained by first shooting a video of the target cross-sectional image and then capturing a picture using, but not limited to, a scanning electron microscope, a flatbed imaging scanner, a charge coupled device (CCD) camera, etc., or by directly taking a photo.

[0053] The sample cross-sectional image can be annotated using software including but not limited to the Wizard Annotation Assistant software, Label-Img software, etc., to annotate the fiber sheath and catheter in the sample cross-sectional image with predicted frame labels and positioning coordinate labels respectively.

[0054] In some embodiments, since the catheter is circular and only has size differences in the cross section, the present invention takes catheters of different sizes into consideration and labels a total of 2,000 to 3,000 samples with the label value set to "vessel". Because the fiber sheath exhibits different morphologies in different vascular bundle types: the fiber sheath of undifferentiated bundles is a single unit; the outer fiber sheath of semi-differentiated / semi-open bundles is independent, with the lateral and inner fiber sheaths connected; the four fiber sheaths of open and tight-waisted bundles are mutually independent; the outer and lateral fiber sheaths of broken-waisted bundles are independent, with fiber strands present in the inner bundle; and the four fiber sheaths of double-broken-waisted bundles are mutually independent, with one fiber strand each in the inner and outer bundles. Since scattered bamboos only have fiber sheaths, while clumping bamboos have both fiber sheaths and fiber strands, all of which are aggregates of fibers, we uniformly label them "fiber." We annotated 400 to 600 samples of fiber sheaths (strands) for each type of vascular bundle (400 to 600 refers to the number of vascular bundles; each bundle contains 1 to 6 fiber sheaths (strands)). After annotation, the coordinate data can be saved as a Pascal-VOC file.

[0055] Based on the sample cross-sectional images and target detection models (such as the YOLO V7 target detection model) marked with the fiber sheath prediction frame label, fiber sheath positioning coordinate label, duct prediction frame label and duct positioning coordinate label, the vascular plant stem cross-sectional structure prediction model provided by the present invention can be trained.

[0056] After obtaining the trained vascular plant stalk cross-sectional structure prediction model, the fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame and duct positioning coordinates in the target cross-sectional image can be obtained according to the target cross-sectional image and the trained vascular plant stalk cross-sectional structure prediction model. Then, the fiber sheath prediction frame and duct prediction frame are respectively mapped to the target cross-sectional image according to the fiber sheath positioning coordinates and the duct positioning coordinates, that is, the fiber sheath prediction frame and the duct prediction frame are drawn on the target cross-sectional image.

[0057] Since each fiber sheath prediction frame and duct prediction frame contains a fiber sheath and a duct, the target within each prediction frame can be segmented out in the form of an image using a binarization method. It should be noted that due to the low contrast between the fiber sheath (strand), duct, and parenchyma in the cross section of a vascular plant stalk, directly binarizing the target cross-sectional image is not effective. The present invention chooses to perform binarization on the fiber sheath prediction frame and duct prediction frame of the target cross-sectional image annotated by a trained vascular plant stalk cross-sectional structure prediction model, thereby obtaining a fiber sheath image and a duct image.

[0058] In some embodiments, since the coordinates of each prediction box in the target cross-sectional image are unique, the obtained fiber sheath image and catheter image can be restored to the corresponding position of the target cross-sectional image and can be divided into multiple layers. The layers of the original target cross-sectional image can be hidden (only the outline is retained), so that the fiber sheath image and catheter image obtained by target binarization are more intuitive.

[0059] Among them, any existing binarization method can be used, including but not limited to the peak method, P parameter method, maximum inter-class variance method (Otsu method), maximum entropy threshold method, iterative method (optimal threshold method), etc. The present invention does not limit the binarization method used.

[0060] The actual size of a single pixel can be determined based on the image resolution of the target cross-sectional image. For example, in an image scanned at 9600 dpi, the side length of a single pixel is 2.64 μm, and the area of ​​a single pixel is 6.9696 μm. 2 .

[0061] Therefore, after binarization to obtain the fiber sheath image and duct image, the image areas of the fiber sheath and duct can be determined based on the number of pixels occupied by the fiber sheath image and the number of pixels occupied by the duct image (after binarization, black pixels represent the target, and white pixels represent the background) and the actual size of a single pixel calculated based on the image resolution. Specifically, by multiplying the number of pixels occupied by the fiber sheath image and duct image by the actual size of a single pixel, the fiber sheath area and duct area in the target cross-sectional image can be obtained, respectively. Furthermore, the parenchyma area can be obtained by subtracting the total area of ​​the fiber sheath (strand) and duct from the cross-sectional area of ​​the vascular plant stem.

[0062] The present invention provides a method for detecting the cross-sectional structure of a vascular plant stalk. The method obtains a cross-sectional structure prediction model of the vascular plant stalk by training a sample cross-sectional image with a fiber sheath prediction frame label, a fiber sheath positioning coordinate label, a duct prediction frame label, and a duct positioning coordinate label. The method obtains the fiber sheath prediction frame, the fiber sheath positioning coordinate, the duct prediction frame, and the duct positioning coordinate in the cross-sectional image according to the target cross-sectional image and the trained cross-sectional structure prediction model of the vascular plant stalk, and corresponds the fiber sheath prediction frame and the duct prediction frame to the cross-sectional image. The method then obtains the fiber sheath image and the duct image corresponding to the cross-sectional image by a binarization method. Finally, the fiber sheath area and the duct area of ​​the vascular plant stalk are calculated according to the fiber sheath image, the duct image, and the image resolution. This allows the cross-sectional structure of the vascular plant stalk to be quickly and accurately measured in large quantities of samples, large-size samples, and large observation areas.

[0063] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, a prediction model for cross-sectional structure of a vascular plant stem is trained according to the following method:

[0064] The sample cross-sectional image is segmented into a plurality of sample sub-images, and the fiber sheath prediction frame label, the fiber sheath positioning coordinate label, the catheter prediction frame label, and the catheter positioning coordinate label in the sample cross-sectional image are synchronously segmented into the plurality of sample sub-images;

[0065] After data augmentation processing, multiple sample sub-graphs are sequentially passed through the target detection model with pre-trained weights, and trained in the freezing and thawing stages to obtain a vascular plant stem cross-sectional structure prediction model.

[0066] Specifically, since the size of the fiber sheath (strand) and duct is very small compared to the cross-sectional size of the vascular plant stem, it is extremely difficult to directly detect the fiber sheath (strand) and duct from the entire stalk cross-sectional image. Therefore, before training the vascular plant stem cross-sectional structure prediction model, the present invention can perform simultaneous image label segmentation and data augmentation processing.

[0067] In some implementations, performing synchronous segmentation of image tags may include the following steps:

[0068] 1. Import each sample subgraph into the necessary Python libraries, including os, random, numpy, pandas, matplotlib, PIL, and torch.

[0069] 2. Define a function exist_objs, which is used to determine whether the intersection over union (IOU) of the new box and the old box meets the restriction conditions (which can be customized and the value range is 0.0 to 1.0). If so, the new box is retained as the target box of the sub-graph, where the old box represents the known target box in the sub-graph.

[0070] 3. Define the coordinates of the upper left corner and lower right corner of each target frame, and traverse all the target frames of the sub-image to determine the positional relationship between the target frame and the sub-image, that is, whether the target frame is within the range of the sub-image, and handle different situations respectively. Figure 2 The schematic diagram of the relative position relationship between the target frame and the sub-image provided by the present invention is as follows: Figure 2 As shown, there are 9 situations:

[0071] ① The target box is completely contained in the sub-image; ② The upper left corner of the target box is above the sub-image, and the lower right corner is in the sub-image; ③ The upper left corner of the target box is above the sub-image, and the lower right corner is on the right side of the sub-image; ④ The upper left corner of the target box is on the left side of the sub-image, and the lower right corner is in the sub-image; ⑤ The upper left corner of the target box is on the left side of the sub-image, and the lower right corner is below the sub-image; ⑥ The upper left corner of the target box is at the upper left of the sub-image, and the lower right corner is in the sub-image; ⑦ The upper left corner of the target box is in the sub-image, and the lower right corner is at the lower right of the sub-image; ⑧ The upper left corner of the target box is in the sub-image, and the lower right corner is below the sub-image; ⑨ The upper left corner of the target box is in the sub-image, and the lower right corner is on the right side of the sub-image.

[0072] 4. According to the positional relationship between the target frame and the sub-image, the position of the target frame in the sub-image is cropped.

[0073] 5. Return the retained target box to generate the annotation information of the sub-image.

[0074] In some implementations, data augmentation processing may include, but is not limited to, flipping, rotation, scaling, cropping, translation, Gaussian noise, changes in image brightness, saturation, and contrast, CutUp, MixUp, etc. As an optional embodiment, mosaic data augmentation processing may be selected, and the probability of using mosaic data augmentation for each step can be set (customizable, with a value range of 0% to 100%). The principle is to crop and mix four images into one image, and the length and width of the cropping position can be randomly varied.

[0075] The sample cross-sectional image is segmented into multiple sample sub-images, and the predicted box labels are segmented simultaneously to ensure a one-to-one correspondence between the sample sub-image image and the predicted box label. Cropping and data augmentation operations are performed on each sub-image to generate more training data and enhance the model's generalization capabilities.

[0076] After obtaining multiple data-enhanced sample subgraphs, the sample subgraphs can be sequentially passed through the target detection model with pre-trained weights, and a vascular plant stem cross-sectional structure prediction model can be obtained through training in freezing and thawing stages.

[0077] In some implementations, the open source yolov7_weights weights can be used as pre-trained weights for the model to reduce the randomness of the backbone feature extraction network, and use half-precision floating point format (FP16) to reduce the required amount of memory.

[0078] The training is divided into two phases: the freezing phase and the thawing phase.

[0079] In the freezing stage, the model backbone is frozen, the feature extraction network does not change, and only fine-tuning is performed on the network.

[0080] In some implementations, the number of frozen training epochs (Freeze_Epoch) can be set to 50-100, and the number of samples per iteration during freezing (Freeze_batch_size) can be set to a maximum of 4 or 8.

[0081] The unfreezing stage occupies a large amount of video memory and the feature extraction network will change.

[0082] In some implementations, the unfreeze training epoch (Unfreeze_Epoch) can be set to 100-500, and the sample size of each iteration during unfreezing (Unfreeze_batch_size) can be set to a smaller value of 2 or 4.

[0083] The selection of other training parameters of this model is related to the optimizer, including but not limited to the Adam optimizer, SGD optimizer, etc.

[0084] In some implementations, when using the Adam optimizer, the initial learning rate (Init_lr) of the model can be set to 0.001, the minimum learning rate of the model defaults to 0.01 of the initial learning rate, and the weight decay (weight_decay) to prevent overfitting is set to 0.

[0085] In some implementations, when using the SGD optimizer, the initial learning rate (Init_lr) of the model can be set to 0.0001, the minimum learning rate of the model defaults to 0.01 of the initial learning rate, and the weight decay (weight_decay) to prevent overfitting is set to 0.0005.

[0086] The learning rate descent method used by this model includes but is not limited to step type, cos type, etc. You can customize the weight saving every n generations (epochs).

[0087] The model calculates loss functions, including but not limited to binary cross entropy logistic loss (BCE_with_Logits_Loss) and intersection over union loss (IOU_Loss). Sufficient training epochs are helpful in finding the global or local minimum of the loss function.

[0088] In some implementations, based on 2000 "vessel" labels and 1000 "fiber" labels, the SGD optimizer is selected, the step type is selected as the learning rate descent method, no freezing training is performed, the sample size of each iteration during thawing is 2, and the thawing training generations are 300, resulting in an average precision (mAP) of 0.959, a recall rate of 0.947, and an accuracy of 0.920.

[0089] After training, a trained vascular plant stem cross-section structure prediction model can be obtained. After inputting the target cross-section image, the fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame and duct positioning coordinates of the target cross-section image can be obtained.

[0090] According to a method for detecting the cross-sectional structure of a vascular plant stem provided by the present invention, a fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame, and duct positioning coordinates in the target cross-sectional image are obtained based on a target cross-sectional image and a trained vascular plant stem cross-sectional structure prediction model, specifically comprising:

[0091] Generate multiple cross-sectional sub-images according to the target cross-sectional image; the size of the cross-sectional sub-image is consistent with the size of the sample sub-image;

[0092] According to the multiple cross-sectional subgraphs and the trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinate, a vessel prediction frame, and a vessel positioning coordinate in the multiple cross-sectional subgraphs are obtained;

[0093] The fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in multiple cross-sectional sub-images are merged to obtain the fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the target cross-sectional image.

[0094] Specifically, because the vascular plant stem cross-sectional structure model provided by the present invention is trained using sample sub-images segmented from a sample cross-sectional image, when using this model, the size of the input image must be consistent with the size of the sample sub-images used during training. Therefore, multiple cross-sectional sub-images can be generated based on the target cross-sectional image, and the sizes of the cross-sectional sub-images are consistent with the sizes of the sample sub-images.

[0095] In some implementations, a sliding window can be used to generate cross-sectional subimages. For example, for a sample subimage with a size of 768×768 pixels, a 768×768 pixel window is generated starting from the upper left corner of the target cross-sectional image file (for a video file, this can be the first frame or any frame) as the first cross-sectional subimage. This window is then translated rightward and downward to generate the remaining cross-sectional subimages. It should be noted that the multiple cross-sectional subimages generated based on the target cross-sectional image include all content on the target cross-sectional image.

[0096] By inputting multiple cross-sectional sub-graphs into the trained vascular plant stem cross-sectional structure prediction model, the fiber sheath prediction frame, fiber sheath positioning coordinates, vessel prediction frame and vessel positioning coordinates in each cross-sectional sub-graph can be obtained.

[0097] Then, the fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in multiple cross-sectional sub-images can be merged according to the size of the cross-sectional sub-image and its corresponding position on the target cross-sectional image, thereby obtaining the fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the target cross-sectional image.

[0098] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, the method further comprises:

[0099] Before generating a plurality of cross-sectional sub-images according to the target cross-sectional image, presetting the overlapping area of ​​adjacent cross-sectional sub-images in the plurality of cross-sectional sub-images; and

[0100] Before merging the fiber sheath prediction frames, fiber sheath positioning coordinates, catheter prediction frames, and catheter positioning coordinates in multiple cross-sectional sub-images, the repeated fiber sheath prediction frames, repeated fiber sheath positioning coordinates, repeated catheter prediction frames, and repeated catheter positioning coordinates in the multiple cross-sectional sub-images are removed based on the area of ​​the overlapping region.

[0101] Specifically, before generating a plurality of cross-sectional sub-images according to the target cross-sectional image, the overlapping areas of adjacent cross-sectional sub-images in the plurality of cross-sectional sub-images may be preset.

[0102] For example, multiple cross-sectional sub-images are obtained by shifting the upper left corner of the image to the right and downward in sequence. The overlapping area between adjacent cross-sectional sub-images can be defined as between 0% and 100%. A preferred method is to set the overlapping area between adjacent cross-sectional sub-images to 30%.

[0103] When the overlapping area is preset, repeated fiber sheath prediction frames, repeated fiber sheath positioning coordinates, repeated catheter prediction frames and repeated catheter positioning coordinates in each cross-sectional sub-image may be filtered according to the preset area.

[0104] In some implementations, non-maximum suppression may be performed on the overlapping portions to calculate the Intersection over Union (IOU) to remove duplicate prediction boxes.

[0105] After removing the repeated prediction frames and repeated positioning coordinates in each cross-sectional sub-image, the fibrous sheath prediction frames, fibrous sheath positioning coordinates, catheter prediction frames and catheter positioning coordinates in multiple cross-sectional sub-images are merged to obtain the fibrous sheath prediction frame, fibrous sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the target cross-sectional image.

[0106] According to a method for detecting cross-sectional structures of vascular plant stems provided by the present invention, before generating a plurality of cross-sectional sub-images based on a target cross-sectional image, the method further comprises:

[0107] Convert the target cross-section image to RGB and add gray bars before resizing it.

[0108] Add an iterative sample size dimension to the resized target cross-section image.

[0109] Specifically, before generating a plurality of cross-sectional sub-images according to the target cross-sectional image, the target cross-sectional image may be subjected to RGB conversion.

[0110] In order to resize the image and obtain a fixed input size, gray bars can be added to the image to achieve distortion-free resizing.

[0111] After changing the size, the sample size (batch_size) dimension of each iteration is added and input into the trained vascular plant stem cross-section structure prediction model.

[0112] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, a fiber sheath image and a duct image of a target cross-sectional image are obtained by a binarization method based on a fiber sheath prediction frame and a duct prediction frame, specifically comprising:

[0113] Filter the fibrous sheath prediction frame and the catheter prediction frame;

[0114] According to the filtered fiber sheath prediction frame and catheter prediction frame, the fiber sheath image and catheter image of the target cross-sectional image are obtained by a binarization method.

[0115] Specifically, after obtaining the fibrous sheath prediction frame and the catheter prediction frame, the fibrous sheath prediction frame and the catheter prediction frame may be filtered.

[0116] Since the shape of the catheter is approximately circular, the aspect ratio of the prediction box should be close to 1:1. In some implementations, when the aspect ratio of a catheter prediction box is less than 1:2 or greater than 2:1, the prediction box is filtered.

[0117] Among them, due to the large size differences of the fibrous sheaths (strands), even through non-maximum suppression, there may still be prediction boxes with too small an area. In some implementations, the fibrous sheaths (strands) can be clustered, and the clustering data indicators are confidence and area size. When the confidence of a fibrous sheath prediction box is less than 90%, the prediction box is filtered.

[0118] Then, based on the filtered fiber sheath prediction frame and the catheter prediction frame, the fiber sheath image and the catheter image of the target cross-sectional image are obtained by a binarization method, and the obtained fiber sheath image and the catheter image are more accurate.

[0119] According to a method for detecting cross-sectional structure of a vascular plant stem provided by the present invention, the method further comprises:

[0120] After obtaining the fiber sheath image and the catheter image of the target cross-sectional image, visualization processing is performed on the fiber sheath image and the catheter image.

[0121] Specifically, after obtaining the fiber sheath image and the catheter image in the target cross-sectional image, they can be visualized using methods such as heat maps.

[0122] In some implementations, the output prediction results can be processed using a sigmoid function and then a heat map can be drawn using the cv2 library to achieve visualization of the feature layer.

[0123] The obtained structural parameter data are visualized to facilitate more in-depth research and analysis of the morphology and structure of the cross-section of the vascular plant stem.

[0124] The vascular plant stem cross-sectional structure detection device provided by the present invention is described below. The vascular plant stem cross-sectional structure detection device described below and the vascular plant stem cross-sectional structure detection method described above can be referenced to each other.

[0125] Figure 3 The schematic diagram of the device for detecting the cross-sectional structure of a vascular plant stem provided by the present invention is as follows: Figure 3 As shown, the device includes:

[0126] Prediction module 300 is used to obtain a fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame, and duct positioning coordinates in the target cross-sectional image based on the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, and map the fiber sheath prediction frame and duct prediction frame to the target cross-sectional image based on the fiber sheath positioning coordinates and duct positioning coordinates;

[0127] An image module 310 is configured to obtain a fiber sheath image and a catheter image of a target cross-sectional image by a binarization method based on the fiber sheath prediction frame and the catheter prediction frame;

[0128] A calculation module 320 is used to calculate the fiber sheath area and the vessel area of ​​the vascular plant stem based on the fiber sheath image, the vessel image and the image resolution;

[0129] Among them, the vascular plant stem cross-sectional structure prediction model is trained based on sample cross-sectional images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels and duct positioning coordinate labels.

[0130] According to a vascular plant stem cross-sectional structure detection device provided by the present invention, a vascular plant stem cross-sectional structure prediction model is trained according to the following method:

[0131] The sample cross-sectional image is segmented into a plurality of sample sub-images, and the fiber sheath prediction frame label, the fiber sheath positioning coordinate label, the catheter prediction frame label, and the catheter positioning coordinate label in the sample cross-sectional image are synchronously segmented into the plurality of sample sub-images;

[0132] After data augmentation processing, multiple sample sub-graphs are sequentially passed through the target detection model with pre-trained weights, and trained in the freezing and thawing stages to obtain a vascular plant stem cross-sectional structure prediction model.

[0133] According to a vascular plant stem cross-sectional structure detection device provided by the present invention, a fiber sheath prediction frame, fiber sheath positioning coordinates, duct prediction frame, and duct positioning coordinates in the target cross-sectional image are obtained based on a target cross-sectional image and a trained vascular plant stem cross-sectional structure prediction model, specifically comprising:

[0134] Generate multiple cross-sectional sub-images according to the target cross-sectional image; the size of the cross-sectional sub-image is consistent with the size of the sample sub-image;

[0135] According to the multiple cross-sectional subgraphs and the trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinate, a vessel prediction frame, and a vessel positioning coordinate in the multiple cross-sectional subgraphs are obtained;

[0136] The fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in multiple cross-sectional sub-images are merged to obtain the fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the target cross-sectional image.

[0137] According to the present invention, a device for detecting cross-sectional structure of a vascular plant stem is provided, the device further comprising:

[0138] a preset module, used for presetting the overlapping area of ​​adjacent cross-sectional sub-images in the plurality of cross-sectional sub-images before generating the plurality of cross-sectional sub-images according to the target cross-sectional image; and

[0139] A deduplication module is used to remove duplicate fiber sheath prediction frames, duplicate fiber sheath positioning coordinates, duplicate catheter prediction frames and duplicate catheter positioning coordinates in multiple cross-sectional sub-images according to the area of ​​the overlapping area before merging the fiber sheath prediction frames, fiber sheath positioning coordinates, catheter prediction frames and catheter positioning coordinates in multiple cross-sectional sub-images.

[0140] According to a device for detecting cross-sectional structures of vascular plant stems provided by the present invention, before generating a plurality of cross-sectional sub-images based on a target cross-sectional image, the device further includes a pre-processing module for:

[0141] Convert the target cross-section image to RGB and add gray bars before resizing it.

[0142] Add an iterative sample size dimension to the resized target cross-section image.

[0143] According to a vascular plant stem cross-sectional structure detection device provided by the present invention, a fiber sheath image and a duct image of a target cross-sectional image are obtained by a binarization method based on a fiber sheath prediction frame and a duct prediction frame, specifically comprising:

[0144] Filter the fibrous sheath prediction frame and the catheter prediction frame;

[0145] According to the filtered fiber sheath prediction frame and catheter prediction frame, the fiber sheath image and catheter image of the target cross-sectional image are obtained by a binarization method.

[0146] According to the present invention, a device for detecting cross-sectional structure of a vascular plant stem is provided, the device further comprising:

[0147] The visualization module is used to perform visualization processing on the fiber sheath image and the catheter image after obtaining the fiber sheath image and the catheter image of the target cross-sectional image.

[0148] Figure 4 A schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the method for detecting the cross-sectional structure of a vascular plant stem, which includes:

[0149] According to the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, the fiber sheath prediction frame, the fiber sheath positioning coordinates, the duct prediction frame and the duct positioning coordinates in the target cross-sectional image are obtained, and the fiber sheath prediction frame and the duct prediction frame are mapped to the target cross-sectional image according to the fiber sheath positioning coordinates and the duct positioning coordinates;

[0150] According to the fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method;

[0151] Calculate the fiber sheath area and vessel area of ​​vascular plant stems based on the fiber sheath image, vessel image and image resolution;

[0152] Among them, the vascular plant stem cross-sectional structure prediction model is trained based on sample cross-sectional images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels and duct positioning coordinate labels.

[0153] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0154] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the vascular plant stem cross-sectional structure detection method provided by the above methods, which includes:

[0155] According to the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, the fiber sheath prediction frame, the fiber sheath positioning coordinates, the duct prediction frame and the duct positioning coordinates in the target cross-sectional image are obtained, and the fiber sheath prediction frame and the duct prediction frame are mapped to the target cross-sectional image according to the fiber sheath positioning coordinates and the duct positioning coordinates;

[0156] According to the fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method;

[0157] Calculate the fiber sheath area and vessel area of ​​vascular plant stems based on the fiber sheath image, vessel image and image resolution;

[0158] Among them, the vascular plant stem cross-sectional structure prediction model is trained based on sample cross-sectional images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels and duct positioning coordinate labels.

[0159] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for detecting the cross-sectional structure of a vascular plant stem provided by the above methods, the method comprising:

[0160] According to the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, the fiber sheath prediction frame, the fiber sheath positioning coordinates, the duct prediction frame and the duct positioning coordinates in the target cross-sectional image are obtained, and the fiber sheath prediction frame and the duct prediction frame are mapped to the target cross-sectional image according to the fiber sheath positioning coordinates and the duct positioning coordinates;

[0161] According to the fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method;

[0162] Calculate the fiber sheath area and vessel area of ​​vascular plant stems based on the fiber sheath image, vessel image and image resolution;

[0163] Among them, the vascular plant stem cross-sectional structure prediction model is trained based on sample cross-sectional images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels and duct positioning coordinate labels.

[0164] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0165] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting the cross-sectional structure of a vascular plant stem, characterized in that: include: According to the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinate, a duct prediction frame, and a duct positioning coordinate in the target cross-sectional image are obtained, and the fiber sheath prediction frame and the duct prediction frame are mapped to the target cross-sectional image according to the fiber sheath positioning coordinates and the duct positioning coordinates; According to the fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method; Calculating the fiber sheath area and the duct area of ​​the vascular plant stem according to the fiber sheath image, the duct image, and the image resolution; The vascular plant stem cross-section structure prediction model is trained based on sample cross-section images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels, and duct positioning coordinate labels; The vascular plant stem cross-sectional structure prediction model is trained according to the following method: Segmenting the sample cross-sectional image into a plurality of sample sub-images, and synchronously segmenting the fiber sheath prediction frame label, the fiber sheath positioning coordinate label, the catheter prediction frame label, and the catheter positioning coordinate label in the sample cross-sectional image into the plurality of sample sub-images; After the multiple sample subgraphs are subjected to data enhancement processing, they are sequentially passed through a target detection model with pre-trained weights, and trained in a freezing stage and a thawing stage to obtain the vascular plant stem cross-section structure prediction model.

2. The method for detecting cross-sectional structure of vascular plant stems according to claim 1, wherein: The method of obtaining the fiber sheath prediction frame, the fiber sheath positioning coordinates, the duct prediction frame, and the duct positioning coordinates in the target cross-sectional image based on the target cross-sectional image and the trained vascular plant stem cross-sectional structure prediction model specifically includes: generating a plurality of cross-sectional sub-images according to the target cross-sectional image; wherein the size of the cross-sectional sub-images is consistent with the size of the sample sub-image; According to the multiple cross-sectional subgraphs and the trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinate, a duct prediction frame, and a duct positioning coordinate in the multiple cross-sectional subgraphs are obtained; The fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the multiple cross-sectional sub-images are merged to obtain the fiber sheath prediction frame, fiber sheath positioning coordinates, catheter prediction frame and catheter positioning coordinates in the target cross-sectional image.

3. The method for detecting cross-sectional structure of vascular plant stems according to claim 2, wherein: The method further comprises: Before generating a plurality of cross-sectional sub-images according to the target cross-sectional image, presetting overlapping areas of adjacent cross-sectional sub-images in the plurality of cross-sectional sub-images; and Before merging the fiber sheath prediction frames, fiber sheath positioning coordinates, catheter prediction frames and catheter positioning coordinates in the multiple cross-sectional sub-images, the repeated fiber sheath prediction frames, repeated fiber sheath positioning coordinates, repeated catheter prediction frames and repeated catheter positioning coordinates in the multiple cross-sectional sub-images are removed based on the area of ​​the overlapping region.

4. The method for detecting cross-sectional structure of vascular plant stems according to claim 2, wherein: Before generating a plurality of cross-sectional sub-images according to the target cross-sectional image, the method further includes: Performing RGB conversion on the target cross-sectional image and adding gray bars before resizing; An iterative sample size dimension is added to the resized target cross-sectional image.

5. The method for detecting cross-sectional structure of vascular plant stems according to claim 1, wherein: The step of obtaining the fiber sheath image and the catheter image of the target cross-sectional image by a binarization method based on the fiber sheath prediction frame and the catheter prediction frame specifically includes: filtering the fiber sheath prediction frame and the catheter prediction frame; According to the filtered fiber sheath prediction frame and the catheter prediction frame, a fiber sheath image and a catheter image of the target cross-sectional image are obtained by a binarization method.

6. The method for detecting cross-sectional structure of vascular plant stems according to any one of claims 1 to 5, characterized in that: The method further comprises: After obtaining the fiber sheath image and the catheter image of the target cross-sectional image, visualization processing is performed on the fiber sheath image and the catheter image.

7. A device for detecting the cross-sectional structure of a vascular plant stem, characterized in that: include: a prediction module for obtaining, based on a target cross-sectional image and a trained vascular plant stem cross-sectional structure prediction model, a fiber sheath prediction frame, a fiber sheath positioning coordinates, a duct prediction frame, and a duct positioning coordinates in the target cross-sectional image, and mapping the fiber sheath prediction frame and the duct prediction frame to the target cross-sectional image based on the fiber sheath positioning coordinates and the duct positioning coordinates; An image module, configured to obtain a fiber sheath image and a catheter image of the target cross-sectional image by a binarization method according to the fiber sheath prediction frame and the catheter prediction frame; a calculation module, configured to calculate the fiber sheath area and the duct area of ​​the vascular plant stem based on the fiber sheath image, the duct image, and the image resolution; The vascular plant stem cross-section structure prediction model is trained based on sample cross-section images with fiber sheath prediction frame labels, fiber sheath positioning coordinate labels, duct prediction frame labels, and duct positioning coordinate labels; The vascular plant stem cross-sectional structure prediction model is trained according to the following method: Segmenting the sample cross-sectional image into a plurality of sample sub-images, and synchronously segmenting the fiber sheath prediction frame label, the fiber sheath positioning coordinate label, the catheter prediction frame label, and the catheter positioning coordinate label in the sample cross-sectional image into the plurality of sample sub-images; After the multiple sample subgraphs are subjected to data enhancement processing, they are sequentially passed through a target detection model with pre-trained weights, and trained in a freezing stage and a thawing stage to obtain the vascular plant stem cross-section structure prediction model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the method for detecting the cross-sectional structure of a vascular plant stem as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting the cross-sectional structure of a vascular plant stem as claimed in any one of claims 1 to 6 is implemented.

Images