Forest root recognition method and system based on classification fusion idea

Abstract

This invention relates to the field of forest tree root system feature recognition technology, and particularly to a method and system for forest tree root system recognition based on the concept of classification fusion. The method includes: acquiring micro-root canal images of the root system of the trees to be tested, constructing a forest tree root system dataset, and constructing three sub-datasets—primary roots, mature roots, and dead roots—based on the color characteristics of roots at different growth stages; based on these three sub-datasets, introducing the attention mechanism PSA module into the U-net network structure to autonomously learn multiple features of the forest tree root system, such as color and morphology, and constructing recognition models for primary roots, mature roots, and dead roots; furthermore, it proposes the concept of classification fusion, superimposing the feature maps of primary roots, mature roots, and dead roots to obtain a feature fusion map that includes root system information at different growth stages, and using this feature fusion map to achieve accurate recognition of forest tree roots at different growth stages, thereby solving the problem of low accuracy in forest tree root system recognition.

Description

Technical Field

[0001] This invention relates to the field of forest tree root system feature recognition technology, and in particular to a forest tree root system recognition method and system based on the concept of classification fusion. Background Technology

[0002] With the development of computer technology, deep learning is increasingly being used in this field. Existing semantic segmentation methods are mainly effective at segmenting crop roots, uncolored young roots, and roots in black mud. However, the boundaries between forest tree roots and soil are blurred, and the soil contains interference such as cracks, stones, and sand that are similar in color to the roots. The root system itself is complex, and varying growth conditions result in different learnable characteristics for different root systems. Existing segmentation methods are first influenced by the background environment, segmenting interfering factors into root systems. Secondly, the different characteristics of roots at different growth stages also lead to a decrease in learning effectiveness, resulting in undersegmentation and oversegmentation of the target. Therefore, existing semantic segmentation methods cannot accurately describe the boundary features of forest tree roots at different growth stages in complex environments, leading to inaccurate identification.

[0003] In existing root system studies, directly extracting features from the acquired root system images results in low accuracy of the obtained feature information, low automation, and is time-consuming and labor-intensive. Summary of the Invention

[0004] This invention provides a method and system for identifying forest tree roots based on the concept of classification and fusion, in order to solve the technical problem of low accuracy in identifying the features of forest tree roots in the prior art.

[0005] On the one hand, this invention provides a method for identifying forest tree root systems based on the concept of classification fusion, comprising:

[0006] Acquire images of microroot canals in the root system of the trees to be tested;

[0007] Based on the different color characteristics of roots at different growth stages, feature extraction is performed on the microroot canal images to obtain primary root feature maps, mature root feature maps, and dead root feature maps, respectively.

[0008] The primary root feature map, mature root feature map, and dead root feature map are superimposed to obtain a feature fusion map;

[0009] Feature extraction is performed on the feature fusion map to obtain the root system feature information of the trees to be tested.

[0010] According to the classification fusion method for identifying forest tree roots provided by the present invention, before extracting features from the microroot canal image based on the different color features corresponding to roots at different growth stages, the method further includes:

[0011] The microroot canal image is cropped to remove the parts of the microroot canal image edge that do not contain root elements.

[0012] According to the present invention, a method for identifying forest tree roots based on the concept of classification fusion is provided. The step of extracting features from the micro-root canal image based on the different color features corresponding to roots at different growth stages to obtain primary root feature maps, mature root feature maps, and dead root feature maps includes:

[0013] Obtain the average RGB value of each pixel in the microroot canal image;

[0014] Based on the different RGB values ​​corresponding to primary roots, mature roots, and dead roots, the corresponding pixels in the microroot canal image are extracted, and the primary root feature map, mature root feature map, and dead root feature map are obtained based on the extracted pixels.

[0015] The primary root feature map, mature root feature map, and dead root feature map are all binary maps.

[0016] According to the present invention, a tree root identification method based on the idea of ​​classification fusion is provided, wherein the RGB values ​​corresponding to the primary roots are: R∈[180, 255], G∈[150, 200], B∈[80, 120];

[0017] The RGB values ​​corresponding to the mature roots are: R∈[180, 200], G∈[100, 130], B∈[30, 60];

[0018] The RGB values ​​corresponding to the dead root are: R∈[100, 180], G∈[80, 100], B∈[30, 60].

[0019] According to the present invention, a method for identifying forest tree roots based on the concept of classification fusion is provided, wherein the root system feature information of the forest trees includes root length density;

[0020] The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes:

[0021] The feature fusion map is subjected to skeletonization extraction to obtain a root system refinement map;

[0022] Scan the detailed root system diagram to obtain the starting point of the root system to be tested;

[0023] Starting from the starting point, each root system pixel is scanned sequentially according to the connectivity of the root system until the endpoint of the root system to be tested is found, and the pixel distance between each two adjacent root system pixels is recorded.

[0024] The root length of the root system to be tested is obtained based on the pixel distance between each two adjacent root system pixels and the calibration coefficient.

[0025] The root length density is obtained based on the root length, the area of ​​the feature fusion map, and the depth of the experimental field.

[0026] According to the present invention, a method for identifying forest tree roots based on the concept of classification fusion is provided, wherein the root system feature information of the forest trees includes root surface area density;

[0027] The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes:

[0028] Calculate the root surface area density using the following formula:

[0029]

[0030] Where S0=N0*C 2 C is the calibration coefficient, N0 is the number of root system pixels in the feature fusion map, A is the area of ​​the feature fusion map, and DOF is the experimental field depth.

[0031] According to the present invention, a method for identifying the root system of forest trees based on the concept of classification fusion is provided, wherein the root system feature information of the forest trees includes the number of roots;

[0032] The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes:

[0033] The feature fusion map is subjected to skeletonization extraction to obtain a root system refinement map;

[0034] Scan the detailed root system diagram to find all independent root systems, which are root systems that are not connected to another root system;

[0035] For each independent root system, scan the independent root system to obtain a starting point, and scan along the starting point until all root nodes connected to the starting point are found.

[0036] The number of roots is determined based on the number of root nodes included in all independent root systems.

[0037] According to the present invention, a method for identifying forest tree roots based on the concept of classification fusion is provided, wherein the root system feature information of the forest tree includes the number of lateral root branches;

[0038] The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes:

[0039] The feature fusion map is subjected to skeletonization extraction to obtain a root system refinement map;

[0040] Scan the detailed root system diagram to find all independent root systems, which are root systems that are not connected to another root system;

[0041] For each independent root system, scan the independent root system to obtain a starting point, and scan along the starting point until all root nodes connected to the starting point are found.

[0042] The number of lateral root branches included in the corresponding independent root system is determined based on the number of root nodes.

[0043] According to the present invention, a method for identifying forest tree roots based on the concept of classification fusion is provided. The step of extracting corresponding pixels from the micro-root canal image based on the different RGB values ​​corresponding to primary roots, mature roots, and dead roots, and obtaining the primary root feature map, mature root feature map, and dead root feature map based on the extracted pixels, includes:

[0044] The microroot canal image is input into one or more pre-trained neural network models to obtain the primary root feature map, mature root feature map, and dead root feature map;

[0045] The neural network model is a U-net neural network model that includes a PSA module.

[0046] On the other hand, the present invention also provides a forest tree root system identification system based on the concept of classification fusion, comprising:

[0047] The acquisition module is used to acquire images of the microroot canals of the tree root system under test.

[0048] The root image segmentation module is used to extract features from the microroot canal image based on the different color features corresponding to roots at different growth stages, and obtain the feature map of primary roots, mature roots and dead roots respectively.

[0049] The segmentation map fusion module is used to overlay the primary root feature map, mature root feature map, and dead root feature map to obtain a feature fusion map;

[0050] The feature extraction module is used to extract features from the feature fusion map to obtain the root system feature information of the trees to be tested.

[0051] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the forest root identification method based on the classification fusion idea described above.

[0052] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the forest root identification method based on the classification fusion idea described above.

[0053] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the forest root identification method based on the classification fusion idea described above.

[0054] The present invention provides a method for identifying forest tree roots based on the concept of classification and fusion. This method acquires micro-root canal images of the tree under test, and then extracts features from the micro-root canal images according to the different color characteristics corresponding to different growth stages of the root system, obtaining primary root feature maps, mature root feature maps, and dead root feature maps. These primary root feature maps, mature root feature maps, and dead root feature maps are then overlaid to obtain a feature fusion map. This feature fusion map includes all root feature information from different growth stages. Based on this feature fusion map, feature extraction is performed to obtain the root feature information of the tree under test, ensuring that the obtained root feature information is more accurate. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0056] Figure 1 This is a flowchart illustrating the forest tree root system identification method based on the classification fusion concept provided by the present invention.

[0057] Figure 2 These are microroot canal images of Populus tomentosa provided by this invention;

[0058] Figure 3 This is a schematic diagram of the feature extraction and feature fusion process provided by the present invention;

[0059] Figure 4 This is a feature fusion diagram of Populus tomentosa provided by the present invention;

[0060] Figure 5 This is a detailed diagram of the root system of Populus tomentosa provided by the present invention;

[0061] Figure 6 This is a diagram of the U-net neural network structure based on the attention mechanism PSA module provided by the present invention;

[0062] Figure 7This is a schematic diagram of the structure and processing flow of the PSA module provided by the present invention;

[0063] Figure 8 This is a schematic diagram of the structure of the forest tree root identification system provided by the present invention;

[0064] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0066] Based on the characteristics of root systems at different growth stages, this invention first extracts features from the acquired micro-root canal images to obtain root feature maps for three different growth stages. Then, these three root feature maps are superimposed to obtain a feature fusion map. This feature fusion map includes all root feature information at different growth stages. Based on this feature fusion map, feature extraction is performed to obtain the root feature information of the test tree, ensuring that the obtained root feature information is more accurate.

[0067] The following is combined Figures 1-9 The technical solution of the present invention will be further described below.

[0068] Example 1:

[0069] This embodiment provides a method for identifying forest tree root systems based on the concept of classification fusion, such as... Figure 1 The acquisition method includes:

[0070] Step 101: Obtain images of the microroot canals of the tree root system to be tested.

[0071] In this embodiment, taking the root microcanal image of Populus tomentosa as an example, after obtaining the root microcanal image of the tree to be tested, it is necessary to crop the microcanal image to remove the parts of the microcanal image that do not contain root elements. The cropped microcanal image is as follows. Figure 2 As shown. For example, rectangular cropping is performed on micro-root canal images to retain images containing root elements; black areas exist at the image edges, which do not contain root elements. In the segmentation of complex root images, this can affect segmentation accuracy and increase the actual computational load. Therefore, rectangular cropping is used to crop the original micro-root canal image, preserving the effective information areas in the root image. This can reduce the workload of subsequent image processing and improve work efficiency.

[0072] Step 102: Based on the different color characteristics of roots at different growth stages, feature extraction is performed on the microroot canal images to obtain primary root feature maps, mature root feature maps, and dead root feature maps.

[0073] This embodiment is based on knowledge of forest tree morphology. According to the characteristics of root systems in their growth stages, the obtained forest tree root images are divided into three categories: primary roots, mature roots, and dead roots. In their nascent stages, forest tree roots possess strong physiological functions, high vitality, and strong nutrient absorption and secretion capabilities. This leads to the oxidation of the surrounding soil, forming an oxidation zone. This oxidation zone absorbs soluble Fe from the surrounding soil. 2+ Oxidized to Fe 3+ , making Fe 2+ Unable to accumulate on the root surface, thus maintaining the "white" color of the root, these roots are primary roots. In the mature stage of forest tree roots, due to root aging, the cell walls of the outer cortex thicken, greatly reducing the root's absorption capacity and preventing toxic substances from penetrating the root interior. This is because Fe... 3+ Iron deposits on the roots, forming a yellowish-brown iron film, typically found on the surface of older roots and the base of the roots. This yellowish-brown iron film on older roots has a protective function, thus giving the roots their "yellow" appearance; these are mature roots. During the senescent stage of forest tree roots, due to factors such as waterlogging, the oxygen content around the roots is insufficient, leading to an increase in iron content. 2+ In addition to the above, the anaerobic decomposition of organic matter produces toxic substances such as H2S. When H2S combines with FeO, FeS is produced and precipitated on the root surface. This further deteriorates the physiological function of the roots, making them more susceptible to disease. Since FeS precipitate is black, the roots appear "black," indicating dead roots. The system overlays the training and validation set images from step 102, calculates the average RGB value of the root region, and reclassifies the images into primary roots, mature roots, and dead roots according to the following root classification criteria.

[0074] Specifically, in this embodiment, based on the different RGB values ​​corresponding to primary roots, mature roots, and dead roots, corresponding pixels in the micro-root canal image are extracted. Based on the extracted pixels, primary root feature maps, mature root feature maps, and dead root feature maps are obtained respectively. All three feature maps are binary images, where black pixels represent the soil background and white pixels represent the root system. The extracted primary root feature maps, mature root feature maps, and dead root feature maps are the same size as the original micro-root canal image.

[0075] Specifically, the RGB values ​​corresponding to the primary roots in this embodiment are shown in Table 1 below.

[0076] Table 1

[0077] R G B primary roots 180-255 150-200 80-120 Mature roots 180-200 100-130 30-60 Death Root 100-180 80-100 30-60

[0078] Specifically, in this embodiment, three trained neural network models are used to extract features from the input microroot canal image to obtain primary root feature maps, mature root feature maps, and dead root feature maps, respectively.

[0079] Step 103: Overlay the primary root feature map, mature root feature map, and dead root feature map to obtain a feature fusion map.

[0080] like Figure 3 In this embodiment, the three feature maps are superimposed to obtain a fused feature map that includes complete root system features. In other embodiments, the three types of feature maps can also be superimposed pairwise to obtain three fused feature maps, each of which includes root system features from two growth stages. For example, for example, ... Figure 2 The fused feature map obtained after processing the microroot canal image as described above is shown below. Figure 4 As shown.

[0081] Step 104: Extract features from the feature fusion map to obtain the root system feature information of the trees to be tested.

[0082] In one embodiment, the root system feature information of trees includes root length density. In this embodiment, the method for obtaining root length specifically includes: extracting a skeletonized image from the feature fusion map to obtain a refined root system image, for example, by... Figure 4 After skeleton extraction, the resulting root system refinement diagram is as follows: Figure 5 As shown. The root system refinement map is scanned to obtain the starting point of the root system to be tested. Starting from the starting point, each root system pixel is scanned sequentially according to the connectivity of the root system until the endpoint of the root system is found, while recording the pixel distance between each pair of adjacent root system pixels. The root system length is obtained based on the pixel distance between each pair of adjacent root system pixels and the calibration coefficient. The root length density is obtained based on the root system length, the area of ​​the feature fusion map, and the depth of the experimental field. Here, root system pixels represent the root system, i.e., the black pixels in the feature fusion map. The distance between two adjacent root system pixels is divided into two cases: if a pixel is located in the top, bottom, left, or right directions of its adjacent pixel, the distance between these two pixels is considered to be 1; if a pixel is located in the top left, top right, bottom left, or bottom right directions of its adjacent pixel, the distance between these two points (diagonal direction) is considered to be 1. The distances between all pixels are summed to obtain the pixel distance value, and then multiplied by a calibration coefficient to obtain the length of the root system to be measured. The calibration coefficient refers to the actual length corresponding to one pixel.

[0083] This embodiment extracts the skeleton of a fused image of forest tree root system features. Essentially, it reduces the width of multi-pixel regions within the root system to a single-pixel width using specific operations, thereby extracting the skeleton of the target root system. The system's refinement operation follows three principles: first, the refinement line obtains the center line of the original target, meaning the refinement process is symmetrical; second, it maintains the original connectivity and structure of the root system; and third, it preserves the basic morphological characteristics of the root system, meaning that after refinement, the root system should not exhibit shortened lines, broken corners, or the addition of unnecessary branches that were not originally present. Based on these principles, the system extracts and quantifies the root length density, root surface area density, number of roots, and number of root branches.

[0084] Root length density refers to the root system length per unit volume. In this embodiment, the root system length is obtained by calculating the pixels in the skeleton image using mathematical statistical methods and then multiplying the result by the distance between pixels. The thinned skeleton lines are 8-connected. First, the image is scanned to obtain a starting point for the root system thinning image. Starting from this starting point, scanning is performed using the connectivity of the skeleton lines. If the next scan point is located in the four directions above, below, left, or right of the current point, the distance between these two points is considered to be 1. If the next scan point is located in the four directions of the current point (upper left, upper right, lower left, and lower right), the distance between these two points (diagonal direction) is considered to be 1. Then, using the next scan point as the current point, repeat the above steps. Finally, during the entire scan process, record the number of times these two cases occur: the number of times distance 1 occurs is N1, and the number of times distance is... The number of times it appears is The mathematical expression for root length is: C is the calibration coefficient, meaning that the actual length of one pixel in the image is C millimeters.

[0085] The mathematical expression for the root length density of a region in a single image is:

[0086]

[0087] Where A is the area of ​​a single image and DOF is the depth of the experimental field.

[0088] In one embodiment, the root system feature information of the trees also includes root surface area density; feature extraction is performed on the feature fusion map to obtain the root system feature information of the trees to be tested, including:

[0089] Calculate the root surface area density using the following formula:

[0090]

[0091] Where S0=N0*C 2C is the calibration coefficient, N0 is the number of root system pixels in the feature fusion map, A is the area of ​​the feature fusion map, and DOF is the experimental field depth.

[0092] Specifically, root surface area density is the root surface area per unit volume. The feature fusion map of the root system can be viewed as a longitudinal section of the root system, and its area can be called the longitudinal sectional area, i.e., the projected area. According to the basic theory of forest nutrition, the radial growth of forest tree roots is uniform; therefore, the root system can be considered as a three-dimensional figure composed of many small cylindrical micro-elements stacked together. Assuming that the number of pixels in the root region in the segmented binary image is N0, then the mathematical expression for the projected area of ​​the root system is S0 = N0 * C. 2 C is the calibration coefficient, which means that the actual length of one pixel in the image is C millimeters.

[0093] Since the root system can be considered as being composed of n small cylindrical elements stacked together, its surface area is the sum of the lateral surfaces of these n small cylindrical elements. The mathematical expression for the surface area of ​​the root system is: The mathematical expression for the root length density of a region in a single image is: A represents the area of ​​a single image, and DOF represents the depth of the experimental field.

[0094] In one embodiment, the root system feature information of the trees includes the number of roots; the method in this embodiment further includes: extracting the skeleton from the feature fusion map to obtain a root system refinement map; scanning the root system refinement map to find all independent root systems, where an independent root system is a root system that is not connected to another root system; for each independent root system, scanning the independent root system to obtain a starting point, and scanning along the starting point until all root nodes connected to the starting point are found; determining the number of roots based on the number of root nodes included in all independent root systems, specifically, for an independent root system, the number of root nodes it has is its corresponding number of roots, and the sum of the number of roots in all independent root systems is the number of roots included in the current image.

[0095] In this embodiment, an independent root system can be understood as two root systems that do not intersect at all on the root system refinement map, meaning that the white pixels of the two root systems are not adjacent. The method for determining the starting point of an independent root system and all root nodes connected to the starting point is the same in this embodiment: the starting point and root nodes are both white pixels, and only one of the eight directions surrounding a white pixel contains a white pixel. If two of the eight directions surrounding a white pixel contain white pixels, then that point is a branching point.

[0096] In one embodiment, the root system feature information of the trees includes the number of lateral root branches. The method in this embodiment further includes: extracting the skeleton from the feature fusion map to obtain a root system refinement map; scanning the root system refinement map to find all independent root systems, where an independent root system is a root system not connected to another root system; for each independent root system, scanning the independent root system to obtain a starting point, and scanning along the starting point until all root nodes connected to the starting point are found; determining the number of lateral root branches included in the corresponding independent root system based on the number of root nodes. Generally, since each independent root system includes a main root, the lateral root branch index of an independent root system is its corresponding root node minus one. For example, if an independent root system has 5 root nodes, then its lateral root branch count is 4. The sum of the lateral root branch counts of all independent root systems is the number of lateral root branches included in the current image.

[0097] For example, let N T Let N be the number of roots. L N represents the number of lateral root branches. For an independent root system, N is... L =N T -1, such as Figure 5 The expression contains four independent root systems, with the number of nodes in each system being 5, 3, 2, and 1 respectively; therefore, N... T =5+1+2+3=11, N L = (5-1)+(1-1)+(2-1)+(3-1)=7. In step 102, to make the extracted root feature maps of different growth stages more accurate, this embodiment trains different neural network models for roots at different growth stages. In this embodiment, the micro-root canal image is input into one or more pre-trained neural network models to obtain primary root feature maps, mature root feature maps, and dead root feature maps; the neural network model is a U-net neural network model including a PSA module.

[0098] For example, in one embodiment, the network model selected is the U-net neural network model. Images of various root systems are used, and the images are randomly divided according to a training set:validation set:test set ratio of 7:2:1. The training set is used to train the model parameters, the validation set is used to evaluate the model's generalization ability after each batch of training and hyperparameter adjustment, and the test set is used to evaluate the model's segmentation performance for that type of root system. The U-net neural network based on the PSA module is trained on three types of root systems: "primitive roots," "mature roots," and "dead roots," respectively, resulting in models with good segmentation performance for each type. These three models are then used to segment root images, yielding segmentation maps of the three types of root systems for the same image, such as... Figure 3Three segmentation images are superimposed in pairs and then in groups of three to obtain four superimposed segmentation images, resulting in a total of seven images. This embodiment uses the feature fusion image obtained by superimposing three images for subsequent feature quantization. This embodiment improves upon the existing U-net neural network by adding a PSA (Attention Assist) module. The structure of the PSA module in this embodiment is as follows: Figure 7 As shown, the functions of the PSA module include:

[0099] 1) The SPC module is used to segment the channels, and multi-scale feature extraction is performed on the spatial information of the feature map of each channel. The SPC module divides the input feature map into S parts [x1, x2, ..., x...]. S-1 The number of channels in each part is [number missing]. Feature map x i GR C'×H×W , i = 0, 1, ..., S-1. The specific number of groups was obtained experimentally. k is the kernel size, and the optimal parameters for k are [3, 5, 7, 9], corresponding to the optimal parameters for G being [1, 4, 8, 16]. After passing through the SPC module, x is finally obtained. i Corresponding F i Concatenate them, F = Cat([F0,F1,…,F S-1 ]), F∈R C×H×W .

[0100] 2) The SE Weight module is used to extract channel attention from feature maps at different scales, resulting in channel attention vectors at each scale. The SE Weight module first performs Global Average Pooling on the output of the SPC module to obtain its global compressed feature value, and then obtains the weight of each channel through a two-layer fully connected bottleneck structure. The weighted feature map is then used as the input to the next layer of the network.

[0101] 3) Use Softmax to recalibrate the features of the multi-scale channel attention vector to obtain new attention weights after multi-scale channel interaction.

[0102] 4) Perform element-wise dot product on the recalibrated weights and the corresponding feature maps to output a feature map with multi-scale feature information and attention weights. This feature map has richer multi-scale information representation capabilities. The PSA module is a plug-and-play, lightweight, simple, and efficient attention mechanism that replaces the 3x3 convolution in the U-net network, improving the U-net network's segmentation performance on root images.

[0103] In this embodiment, the schematic diagram of the U-net network structure improved based on the attention mechanism PSA module is shown below. Figure 6 As shown. Figure 6 The system includes modules A and B. The image is input with three channels (R, G, and B), and a 3×3 convolutional layer generates a 64-channel feature map. After being input into the PSA module, the output is a feature map with the same number of channels and the same size, but with more learning weights assigned to the target. In downsampling, low-level features are mainly extracted through module A, while in upsampling, segmentation maps of different scales are mainly generated through module B.

[0104] In module A, the feature map is first downsampled by a factor of two through pooling. Then, a PSA module generates a feature map with twice the number of channels and assigns more learning weights to the target. Finally, a second PSA module outputs a feature map with the same number of channels and size but with more learning weights for the target. This achieves the operation of halving the size of the input feature map and doubling the number of channels. In module B, the feature map is first upsampled by a factor of two, doubling the size of the feature map and halving the number of channels. The upsampled feature map is then concatenated with the feature map of the same size and number of channels output from module A to fuse and learn more detailed low-level semantic features. The resulting feature map has the same size but doubled the number of channels. A PSA module generates a feature map with half the number of channels and assigns more learning weights to the target. A second PSA module outputs a feature map with the same number of channels and size but with more learning weights for the target. When a color image of a poplar root system is input, the U-net network based on the PSA module enhances its target learning ability, thus learning richer root features and improving the network's root segmentation ability.

[0105] Example 2:

[0106] The following describes the tree root system feature information acquisition device provided by the present invention. The tree root system feature information acquisition device described below can be referred to in correspondence with the tree root system identification method based on the classification fusion idea described above.

[0107] This embodiment presents a forest tree root system identification system based on the concept of classification and fusion, such as... Figure 8 The tree root system feature information acquisition device includes: acquisition module 201, root image segmentation module 202, segmentation image fusion module 203, and feature extraction module 204.

[0108] The system comprises three modules: an acquisition module 201, a root image acquisition module 202, and a root image segmentation module 203. The former extracts features from the micro-root canal images based on the different color characteristics corresponding to different growth stages of the root system, resulting in primary root feature images, mature root feature images, and dead root feature images. The latter overlays the primary root feature images, mature root feature images, and dead root feature images to obtain a feature fusion image. The former extracts features from the feature fusion image to obtain the root system feature information of the tree under test.

[0109] The implementation methods of the functions of each module in this embodiment are the same as those in Embodiment 1 above, and will not be repeated here.

[0110] Example 3:

[0111] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute the forest tree root system identification method based on the classification fusion idea provided in the above embodiment. The method includes: acquiring micro-root canal images of the root system of the forest tree to be tested; extracting features from the micro-root canal images according to the different color features corresponding to different growth stages of the root system to obtain primary root feature images, mature root feature images, and dead root feature images respectively; overlaying the primary root feature images, mature root feature images, and dead root feature images to obtain a feature fusion image; and extracting features from the feature fusion image to obtain the root system feature information of the forest tree to be tested.

[0112] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the forest tree root system identification method based on the classification fusion idea provided by the above methods. The method includes: acquiring micro-root canal images of the root system of the forest tree to be tested; extracting features from the micro-root canal images according to the different color features corresponding to different growth stages of the root system to obtain primary root feature images, mature root feature images, and dead root feature images respectively; superimposing the primary root feature images, mature root feature images, and dead root feature images to obtain a feature fusion image; and extracting features from the feature fusion image to obtain the root system feature information of the forest tree to be tested.

[0114] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the tree root system identification method based on the classification fusion concept provided by the above methods. The method includes: acquiring micro-root canal images of the tree root system to be tested; extracting features from the micro-root canal images according to the different color features corresponding to different growth stages of the root system to obtain primary root feature images, mature root feature images, and dead root feature images respectively; overlaying the primary root feature images, mature root feature images, and dead root feature images to obtain a feature fusion image; and extracting features from the feature fusion image to obtain the root system feature information of the tree to be tested.

[0115] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0116] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying forest tree root systems based on the concept of classification and fusion, characterized in that, include: Acquire images of microroot canals in the root system of the trees to be tested; Based on the different color characteristics of roots at different growth stages, feature extraction is performed on the microroot canal image to obtain primary root feature map, mature root feature map and dead root feature map, including: obtaining the average RGB value of each pixel in the microroot canal image; Based on the different RGB values ​​corresponding to primary roots, mature roots, and dead roots, the corresponding pixels in the microroot canal image are extracted, and the primary root feature map, mature root feature map, and dead root feature map are obtained based on the extracted pixels; the primary root feature map, mature root feature map, and dead root feature map are all binary images. The primary root feature map, mature root feature map, and dead root feature map are superimposed to obtain a feature fusion map; Feature extraction is performed on the feature fusion map to obtain the root system feature information of the trees to be tested.

2. The forest tree root system identification method based on the classification fusion idea according to claim 1, characterized in that, The RGB values ​​corresponding to the primary roots are: R∈[180, 255], G∈[150, 200], B∈[80, 120]; The RGB values ​​corresponding to the mature roots are: R∈[180, 200], G∈[100, 130], B∈[30, 60]; The RGB values ​​corresponding to the dead root are: R∈[100, 180], G∈[80, 100], B∈[30, 60].

3. The forest tree root system identification method based on the classification fusion idea according to claim 2, characterized in that, The step of extracting corresponding pixels from the microroot canal image based on the different RGB values ​​corresponding to primary roots, mature roots, and dead roots, and obtaining the primary root feature map, mature root feature map, and dead root feature map based on the extracted pixels includes: The microroot canal image is input into one or more pre-trained neural network models to obtain the primary root feature map, mature root feature map, and dead root feature map; The neural network model is a U-net neural network model that includes a PSA module.

4. The forest tree root system identification method based on the classification fusion idea according to claim 2, characterized in that, The root system characteristics of the trees include root length density; The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes: The feature fusion map is subjected to skeletonization extraction to obtain a root system refinement map; Scan the detailed root system diagram to obtain the starting point of the root system to be tested; Starting from the starting point, each root system pixel is scanned sequentially according to the connectivity of the root system until the endpoint of the root system to be tested is found, and the pixel distance between each two adjacent root system pixels is recorded. The root length of the root system to be tested is obtained based on the pixel distance between each two adjacent root system pixels and the calibration coefficient. The root length density is obtained based on the root length, the area of ​​the feature fusion map, and the depth of the experimental field.

5. The tree root system identification method based on classification fusion as described in claim 1, characterized in that, The root system characteristics of the trees include root surface area density; The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes: Calculate the root surface area density using the following formula: Where S0=N0*C 2 C is the calibration coefficient, N0 is the number of root system pixels in the feature fusion map, A is the area of ​​the feature fusion map, and DOF is the experimental field depth.

6. The forest tree root system identification method based on the classification fusion idea according to claim 1, characterized in that, The root system characteristics of the trees include the number of roots; The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes: The feature fusion map is subjected to skeletonization extraction to obtain a root system refinement map; Scan the detailed root system diagram to find all independent root systems, which are root systems that are not connected to another root system; For each independent root system, scan the independent root system to obtain a starting point, and scan along the starting point until all root nodes connected to the starting point are found. The number of roots is determined based on the number of root nodes included in all independent root systems.

7. The forest tree root system identification method based on the classification fusion idea according to claim 1, characterized in that, The root system characteristics of the trees include the number of lateral root branches; The step of extracting features from the feature fusion map to obtain the root system feature information of the trees to be tested includes: The feature fusion map is subjected to skeletonization extraction to obtain a root system refinement map; Scan the detailed root system diagram to find all independent root systems, which are root systems that are not connected to another root system; For each independent root system, scan the independent root system to obtain a starting point, and scan along the starting point until all root nodes connected to the starting point are found. The number of lateral root branches included in the corresponding independent root system is determined based on the number of root nodes.

8. A forest tree root system identification system based on the concept of classification and fusion, characterized in that, include: The acquisition module is used to acquire images of the microroot canals of the tree root system under test. The root image segmentation module is used to extract features from the microroot canal image based on the different color features corresponding to the root system at different growth stages, and to obtain the primary root feature map, mature root feature map and dead root feature map respectively, including: obtaining the average RGB value of each pixel in the microroot canal image; Based on the different RGB values ​​corresponding to primary roots, mature roots, and dead roots, the corresponding pixels in the microroot canal image are extracted, and the primary root feature map, mature root feature map, and dead root feature map are obtained based on the extracted pixels; the primary root feature map, mature root feature map, and dead root feature map are all binary images. The segmentation map fusion module is used to overlay the primary root feature map, mature root feature map, and dead root feature map to obtain a feature fusion map; The feature extraction module is used to extract features from the feature fusion map to obtain the root system feature information of the trees to be tested.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the forest root system identification method based on the classification fusion idea as described in any one of claims 1 to 7.

Images