Automatic segmentation method for dendritic structure image of laser cladding coating based on Dendritic Net

Abstract

The invention discloses an automatic segmentation method for a dendritic structure image of a laser cladding coating based on Dendritic Net. The automatic segmentation method comprises the following steps: S1, acquiring the dendritic structure image of the laser cladding coating by using a scanning electron microscope; s2, performing preprocessing, data annotation and data enhancement on the acquired image, constructing a dendritic structure data set, and dividing the data set into a training set and a verification set; s3, on the basis of a Pytorch framework, a dendritic structure image segmentation model Dendritic Net based on an HRNet model is established; s4, training the Dendritic Net model by using the training set in the dendritic structure data set, and checking the generalization ability of the Dendritic structure image segmentation model Dendritic Net by using the verification set; and S5, inputting a dendritic crystal structure image to be identified into the trained model, and identifying a dendritic crystal region and a dendritic crystal type in the image. According to the method, information of dendrites, blocky crystals and isometric crystals in the image is extracted in combination with a pixel segmentation method in a computer vision technology, and qualitative and quantitative characterization of the dendrite tissue structure of the laser cladding coating can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of laser cladding coating microstructure image analysis and processing, and particularly relates to an automatic segmentation method for dendritic structure images of laser cladding coatings based on DendriticNet. Background Art

[0002] Laser Cladding is an excellent surface modification technology. By focusing a high-energy laser beam to generate heat, it quickly melts metal powder to form a molten pool, and then quickly solidifies to form a coating with excellent mechanical and mechanical properties such as high hardness, good wear resistance, and strong corrosion resistance. Therefore, it has broad application prospects in aerospace, transportation, and manufacturing industries. The performance of laser cladding coatings is closely related to the dendritic size and morphology in the coating microstructure. Different dendritic sizes, sizes, and their distributions form different microstructures, which can directly affect key properties such as the mechanical strength, corrosion resistance, and oxidation resistance of alloys. Therefore, in order to more deeply explore the influence relationship between the microstructure and properties of the cladding layer, identifying and detecting different dendrites in the cladding layer microstructure is the key to characterizing the relationship between microstructure and properties.

[0003] Currently, most relevant researchers mostly use simple image processing software such as Image-J for auxiliary analysis. The algorithms used in these software are mostly traditional image segmentation algorithms such as threshold segmentation, edge detection segmentation, and watershed algorithm. Although it solves the problem of low efficiency of manual inspection, there are still problems such as missed recognition, misrecognition, incomplete detection of dendritic edges, and inability to further subdivide dendritic categories. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to propose an automatic segmentation method for dendritic structure images of laser cladding coatings based on DendriticNet, which combines deep learning technology and image processing technology to realize qualitative and quantitative characterization of dendrites, blocky crystals, and equiaxed crystals in dendritic structure images.

[0005] In order to achieve the above technical purpose, the technical solution adopted by the present invention is as follows:

[0006] An automatic segmentation method for dendritic structure images of laser cladding coatings based on DendriticNet, comprising:

[0007] S1: Collect dendritic structure images of laser cladding coatings using a scanning electron microscope;

[0008] S2: Preprocess, data label, and data augment the collected images to construct a dendritic structure dataset, and divide the dataset into a training set and a validation set;

[0009] S3: Based on the Pytorch framework, establish a dendritic structure image segmentation model DendriticNet based on the HRNet model. The DendriticNet introduces transfer learning and a residual dual attention module on the basis of the HRNet network model;

[0010] S4: Use the training set in the dendritic structure dataset to train the DendriticNet model, and use the validation set to test the generalization ability of the dendritic structure image segmentation model DendriticNet;

[0011] S5: Input the dendritic structure image to be recognized into the trained model to recognize the dendritic region and dendritic type in the image.

[0012] In some embodiments, in step S2, use the Image Labeler in Matlab to perform data annotation on each picture; the data augmentation expands the dataset by means of cropping, flipping, brightness, contrast, rotation, etc.; the image preprocessing steps include screening and noise removal.

[0013] In some embodiments, in step S2, the dataset is divided into a training set and a validation set according to a ratio of 9:1.

[0014] In some embodiments, the DendriticNet model includes:

[0015] Input layer: Receive the original image data;

[0016] Convolution layer: Extract local features of dendritic structures. By using a 7×7 convolution kernel to slide and scan the input data, spatial information is retained;

[0017] Res_CSP module: Consists of a residual atrous spatial convolution pooling pyramid and a convolutional block attention mechanism, fusing multi-scale dendritic structure context feature information, channel information, and spatial information to further extract and refine the edge features of different dendritic structures;

[0018] Pooling layer: By reducing the spatial resolution of the feature map, the computational amount and the number of parameters of the subsequent layers are reduced; through the aggregation operation of the local area, the model is made more robust to minor shape changes;

[0019] Fully connected layer: Map the extracted high-order features to the labeled sample space for final classification;

[0020] Softmax layer: Calculate the final output and convert the output of the fully connected layer into a probability distribution.

[0021] In some embodiments, the local features of dendritic structures include edges, textures, shapes, etc.

[0022] In some embodiments, the DendriticNet model uses the cross-entropy function as the loss function.

[0023] In some embodiments, during the model training process in step S4, the input image size is set to 512×512, the momentum is set to 0.9, the cosine annealing descent method is used for training, the batch size is set to 16, the Adam optimizer is used to dynamically optimize the learning rate during the learning process, the initial learning rate is set to 0.001, the weight decay coefficient is 1e-2, and the total number of training epochs is set to 200.

[0024] In some embodiments, the automatic segmentation method for the dendritic structure image of the laser cladding coating based on DendriticNet further includes S6: According to the pixel point information of the identified different dendritic regions, calculate the proportion, area, area ratio of different dendritic densities in the image, as well as the total proportion of dendritic density and the total dendritic area.

[0025] In some embodiments, S6 specifically includes the following steps:

[0026] S61: Obtain the pixel length t in the dendritic structure image;

[0027] S62: According to the pixel point information of the identified different dendritic regions, calculate the density proportion of different dendritic structures. The specific calculation method is as follows:

[0028]

[0029] Among them, D Dendritic crystal represents the density of dendritic crystals, ∑pixel(Dendritic crystal) represents the number of pixel points occupied by dendritic crystals, and Total pixel represents the total number of all pixel points in the image; D Bulk crystal represents the density of bulk crystals, and ∑pixel(Bulk block) represents the number of pixel points occupied by bulk crystals; D Equiax crystal represents the density of equiaxed crystals, and ∑pixel(Equiax crystal) represents the number of pixel points occupied by equiaxed crystals;

[0030] The total proportion of dendritic density is obtained by adding the density of dendritic crystals, the density of bulk crystals, and the density of equiaxed crystals;

[0031] S63: According to the pixel point information of the identified different dendritic regions and the pixel length t, calculate the area, area ratio, and total dendritic area of different dendrites;

[0032] Among them, the calculation method of the dendritic crystal area A Dendritic crystal is as follows:

[0033] A Dendritic crystal = t 2 × ∑ pixel(Dendritic crystal)

[0034] The area A of dendritic crystal Bulk crystal is calculated as follows:

[0035] A Bulk crystal = t 2 × ∑ pixel(Bulk crystal)

[0036] The area A of equiaxed crystal Equiax crystal is calculated as follows:

[0037] A Equiax crystal = t 2 × ∑ pixel(Equiax crystal)

[0038] The area ratio AR of dendritic crystal Dendritic crystal is calculated as follows:

[0039] AR Dendritic crystal = A Dendritic crystal / (A Dendritic crystal + A Bulk crystal + A Equiax crystal )

[0040] The area ratio AR of bulk crystal Bulk crystal is calculated as follows:

[0041] AR Bulk crystal = A Bulk crystal / (A Dendritic crystal + A Bulk crystal + A Equiax crystal )

[0042] The area ratio AR of equiaxed crystal Equiax crystal is calculated as follows:

[0043] AR Equiax crystal = A Equiaxcrystal / (A Dendritic crystal +A Bulk crystal +A Equiax crystal )

[0044] The calculation method of the total dendritic area is as follows:

[0045] A = A Dendritic crystal +A Bulk crystal +A Equiax crystal .

[0046] In some embodiments, the method for obtaining the pixel length t in the dendritic structure image in step S61 is as follows:

[0047] Extract the scale on the dendritic structure image and obtain the pixel length of the scale (i.e., the number of pixel points in a single row of the scale);

[0048] Calculate the pixel length t according to the marked length L of the scale and the corresponding pixel length S of the scale. The expression of t is: t = L / S.

[0049] Adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] 1) The present invention combines the pixel segmentation method in computer vision technology to extract the information of dendritic crystals, massive crystals and equiaxed crystals in the image, and uses the DendriticNet model to learn the extracted different dendritic information, so as to realize the qualitative and quantitative characterization of the dendritic structure of the laser cladding coating. The present invention uses a deep learning algorithm, combines image processing technology, reduces the error rate of manually distinguishing the categories of dendritic structure images, improves the recognition efficiency, and can realize the qualitative and quantitative analysis of dendritic structures at the micron scale, effectively saving labor costs and time costs.

[0051] 2) In the present invention, the DendriticNet model can accurately segment different dendritic structures in the laser cladding coating. To address the issue of the model's recognition robustness under different magnification factors, lighting conditions, and dendritic overlap situations, the DendriticNet model introduces a transfer learning strategy based on the HRNet network. Subsequently, to solve problems such as missed recognition, misrecognition, and rough dendritic edge detection that easily occur when the model identifies incomplete dendritic structures and fine dendrites, the DendriticNet network model adds a Res_CSP module after Stage four and before the network output layer of the HRNet network to enhance the network's ability to predict details. Finally, to avoid the occurrence of the gradient vanishing phenomenon, the cross-entropy function is used as the loss function of the DendriticNet network, and the cosine annealing descent method is applied to fine-tune the above-built model. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0053] Figure 1 is the flowchart of the embodiment of the present invention;

[0054] Figure 2 is the data annotation diagram of the dendritic structure image of the laser cladding coating;

[0055] Figure 3 is the schematic structural diagram of the dendritic structure image segmentation model DendriticNet;

[0056] Figure 4 is the flowchart of pixel length extraction;

[0057] Figure 5 is the total dendritic density and area ratio diagram measured by the method of the embodiment of the present invention;

[0058] Figure 6 is the different dendritic density and area ratio diagram measured by the method of the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0059] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be specifically pointed out that the following embodiments are only used to illustrate the present invention, but do not limit the scope of the present invention. Similarly, the following embodiments are only partial embodiments of the present invention rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present invention.

[0060] Referring to the attached Figure 1 As shown, this embodiment provides an automatic segmentation method for dendritic structure images of laser cladding coatings based on DendriticNet, including:

[0061] S1. Specimen selection: Select laser cladding coatings under different materials and different heat treatment states, and cut and select materials according to the specified specimen requirements.

[0062] S2. Specimen preparation: Operate on the specimens such as grinding, polishing, and etching according to the standard specimen preparation requirements.

[0063] S3. Use a scanning electron microscope (SEM) to collect dendritic structure images of the laser cladding coating; adjust the microscope focal length and brightness during the image shooting process to clearly display the details of the dendritic structure, select a suitable field of view, with a magnification of 2.0k or 3.0k, and collect three types of dendritic structures: dendritic crystals, massive crystals, and equiaxed crystals.

[0064] S4. After preprocessing the collected images such as screening and noise removal, use the image annotation tool Image Labeler in MATLAB-R2021a to annotate each dendritic structure image according to four regions: background, dendritic crystals, massive crystals, and equiaxed crystals (as shown in the attached Figure 2 figure); after the data annotation is completed, crop the images, and then use data augmentation methods such as flipping, brightness, contrast, and rotation on the cropped images to expand the dendritic structure dataset to construct a dendritic structure dataset, improve the robustness and generalization ability of the model, and then divide the dataset into a training set and a validation set according to a ratio of 9:1.

[0065] In this embodiment, after data screening and processing of the collected images, a total of 60 images are obtained. Subsequently, the images are cropped at intervals of 300 pixels into images of size 512×512. Data augmentation methods such as flipping, brightness contrast, and rotation are used on the cropped images to perform offline augmentation on the dataset, expanding the dataset to 1868 images, and improving the robustness and generalization ability of the model.

[0066] S5. Based on the Pytorch framework, the DendriticNet network model is built. To address the problems of misrecognition, missed recognition, and rough edge segmentation, transfer learning and the Residual Cross-stage Partial (Res_CSP) module are introduced on the basis of the HRNet network model, and the dendritic microstructure image segmentation model DendriticNet is proposed. The DendriticNet model uses the cross-entropy function as the loss function to improve the performance of the model.

[0067] Among them, the DendriticNet model includes:

[0068] Input layer: Receives the original image data;

[0069] Convolutional layer: Extracts local features of dendritic microstructures (such as edges, textures, shapes, etc.). By using a 7×7 convolutional kernel to slide and scan the input data, spatial information is retained;

[0070] Res_CSP module: Consists of a residual dilated spatial pyramid pooling and a convolutional block attention module, which fuses multi-scale context feature information, channel information, and spatial information of dendritic microstructures to further extract and refine the edge features of different dendritic microstructures;

[0071] Pooling layer: Reduces the spatial resolution of the feature map, reducing the computational amount and the number of parameters of the subsequent layers; Through the aggregation operation of the local area, the model is made more robust to small shape changes;

[0072] Fully connected layer: Maps the extracted high-order features to the labeled sample space for final classification;

[0073] Softmax layer: Calculates the final output and converts the output of the fully connected layer into a probability distribution.

[0074] S6: Use the training set in the dendritic microstructure dataset to train the DendriticNet model. Specifically, in this embodiment, an NVIDIA GeForce RTX3090 graphics card is used for acceleration to train the network model on the dendritic dataset. The input image size is set to 512×512, the momentum is set to 0.9, the cosine annealing descent method is used for training, the batch size is set to 16, the Adam optimizer is used to dynamically optimize the learning rate during the learning process, the initial learning rate is set to 0.001, the weight decay coefficient is 1e-2, and the total number of training epochs is set to 200.

[0075] Result test: Use the validation set to test the generalization ability of the dendritic microstructure image segmentation model, that is, to test the recognition accuracy of the model for the laser cladding coating dendritic microstructure images of non-training samples. After testing, the model finally obtained a recognition accuracy of 93.01%.

[0076] S7. Input the dendritic microstructure image to be recognized into the trained model to recognize the dendritic regions and dendritic types in the image; the DendriticNet model will accurately segment and identify different dendritic microstructures (dendritic crystals, bulk crystals, equiaxed crystals) in the image, and the predicted label map output by the model will display the pixel information of each region. Through the above steps, qualitative characterization of dendrites is achieved at the micron scale.

[0077] S8. According to the pixel point information of different dendritic regions recognized, calculate the proportion of different dendritic densities, area, area proportion, total dendritic density proportion and total dendritic area in the image; specifically, it includes the following steps:

[0078] S81: In order to convert the pixel point information into actual size information, first measure and extract the pixel size information of the coating dendritic microstructure image; the method for obtaining the pixel length t in the dendritic microstructure image is as follows:

[0079] Extract the scale on the dendritic microstructure image and obtain the pixel length of the scale (i.e., the number of pixel points in a single row of the scale);

[0080] Calculate the pixel length t according to the marked length L of the scale and the pixel length S corresponding to the scale. The expression of t is: t = L / S.

[0081] For easy understanding, the following provides an example for illustration. As shown in the appendix Figure 4 shown, the marked length L of the scale on the dendritic microstructure image is 30 μm, and the pixel length S corresponding to the scale is 380 pixels. Through further conversion, the pixel length t of the coating dendritic microstructure image can be calculated as: t = 0.079 μm / pixel.

[0082] S82: The segmentation map output by the dendritic microstructure image segmentation model shows the pixel information of four regions: background, dendritic crystal, bulk crystal and equiaxed crystal. According to the pixel point information of different dendritic regions recognized, calculate the density proportion of different dendritic microstructures. The calculation method is as follows:

[0083]

[0084] Among them, D Dendritic crystal represents the density of dendritic crystals, ∑pixel(Dendritic crystal) represents the number of pixel points occupied by dendritic crystals, and Total pixel represents the total number of pixel points in the image; D Bulk crystal represents the density of bulk crystals, and ∑pixel(Bulk block) represents the number of pixel points occupied by bulk crystals; D Equiax crystalThe density of equiaxed crystals is denoted as ρ(Equiax crystal), and ∑pixel(Equiax crystal) represents the number of pixels occupied by equiaxed crystals;

[0085] The total proportion of dendritic density is obtained by adding the density of dendritic crystals, the density of massive crystals, and the density of equiaxed crystals;

[0086] S83: Calculate the areas, area proportions, and total dendritic areas of different dendritic regions based on the pixel information and pixel length t of the identified different dendritic regions;

[0087] Among them, the dendritic crystal area A Dendritic crystal is equal to the product of the number of pixels marked as dendritic crystals in the prediction map and t 2 , and the expression is as follows:

[0088] A Dendritic crystal = t 2 ×∑pixel(Dendritic crystal)

[0089] The massive crystal area A Bulk crystal is equal to the product of the number of pixels marked as massive crystals in the prediction map and t 2 , and the expression is as follows:

[0090] A Bulk crystal = t 2 ×∑pixel(Bulk crystal)

[0091] The equiaxed crystal area A Equiax crystal is equal to the product of the number of pixels marked as equiaxed crystals in the prediction map and t 2 , and the expression is as follows:

[0092] A Equiax crystal = t 2 ×∑pixel(Equiax crystal)

[0093] The area proportion AR of dendritic crystals Dendritic crystal is calculated as follows:

[0094] AR Dendritic crystal = A Dendritic crystal / (A Dendritic crystal + A Bulk crystal + A Equiax crystal )

[0095] The area proportion AR of massive crystals Bulk crystal is calculated as follows:

[0096] AR Bulk crystal = A Bulk crystal / (A Dendritic crystal + A Bulk crystal + A Equiax crystal )

[0097] Area ratio AR of equiaxed crystals Equiax crystal The calculation method is as follows:

[0098] AR Equiax crystal = A Equiax crystal / (A Dendritic crystal + A Bulk crystal + A Equiax crystal )

[0099] The calculation method of the total dendritic area is as follows:

[0100] A = A Dendritic crystal + A Bulk crystal + A Equiax crystal .

[0101] By the above methods, the total dendritic density, area, and the density, area, and area ratio of different types of dendrites are measured, realizing the quantitative characterization of dendrites at the micron scale (as shown in the appendix). Figures 5 - 6 .

[0102] In practical applications, such as Figure 3 , the dendritic tissue image segmentation model inherits the advantages of the HRNet high-resolution network, enabling the model to continuously use high-resolution feature maps for repeated multi-scale fusion throughout the dendritic tissue feature processing stage, and being able to maintain a high-resolution image when extracting dendritic tissue feature information, ultimately obtaining rich high-resolution dendritic tissue features for prediction results. To improve the robustness and generalization ability of the model when identifying different magnifications, different illuminations, and dendritic overlaps, etc., and to enable the model to converge faster and improve performance, this paper uses transfer learning to initialize the model parameters obtained by training the HRNet-18 network on the large-scale dataset ImageNet and transfer them to the DendriticNet model proposed in this paper. To increase the receptive field of the network and extract more abundant multi-scale context dendritic tissue feature information, and further extract and refine the edge features of the dendritic tissue to improve the problems of unclear edge segmentation and mis-segmentation, a residual dual attention (Res_CSP) module is added after Stage four and before the network output layer to enhance the network's ability to predict details. To avoid the phenomenon of gradient disappearance, the cross-entropy function is used as the loss function of the DendriticNet network to improve the performance of the model.

[0103] After testing, the mIoU of the DendriticNet model in the dendritic microstructure image segmentation task is 80.31%, MP is 88.70%, MR is 88.43%, F1-score is 88.56%, and Accuracy is 93.01%. Compared with the HRNet basic network before improvement, the mIoU can be increased by 2.62%, MP by 1.55%, MR by 2.01%, F1-score by 1.78%, and Accuracy by 0.38%, realizing the qualitative characterization of dendrites at the micron scale.

[0104] The above are only some embodiments of the present invention, and thus do not limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall equally be included in the patent protection scope of the present invention.

Claims

1. An automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet, characterized in that, Including: S1: Collecting dendritic microstructure images of the laser cladding coating using a scanning electron microscope; S2: Preprocessing, data annotation, and data augmentation of the collected images to construct a dendritic microstructure dataset, and dividing the dataset into a training set and a validation set; S3: Based on the Pytorch framework, establishing a dendritic microstructure image segmentation model DendriticNet based on the HRNet model, where DendriticNet introduces transfer learning and a residual dual attention module on the basis of the HRNet network model; S4: Training the DendriticNet model using the training set in the dendritic microstructure dataset, and testing the generalization ability of the dendritic microstructure image segmentation model DendriticNet using the validation set; S5: Inputting the dendritic microstructure image to be recognized into the trained model to recognize the dendritic region and dendritic type in the image.

2. The automatic segmentation method of the dendritic structure image of the laser cladding coating based on DendriticNet according to claim 1, wherein, In step S2, the Image Labeler in Matlab is used to perform data annotation on each picture; the data augmentation expands the dataset by means of cropping, flipping, brightness, contrast, and rotation; the image preprocessing steps include screening and noise removal.

3. The automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet according to claim 1, characterized in that, In step S2, the dataset is divided into a training set and a validation set according to a ratio of 9:

1.

4. The automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet according to claim 1, characterized in that, The DendriticNet model includes: Input layer: Receiving the original image data; Convolutional layer: Extracting local features of dendritic microstructure, sliding and scanning the input data using a 7×7 convolutional kernel to retain spatial information; Res_CSP module: Composed of a residual dilated spatial convolution pooling pyramid and a convolutional block attention mechanism, fusing multi-scale dendritic microstructure context feature information, channel information, and spatial information to further extract and refine the edge features of different dendritic microstructures; Pooling layer: Reducing the spatial resolution of the feature map to reduce the computational amount and the number of parameters of the subsequent layers; making the model more robust to minor shape changes through local area aggregation operations; Fully connected layer: Mapping the extracted high-order features to the labeled sample space for final classification; Softmax layer: Calculating the final output and converting the output of the fully connected layer into a probability distribution.

5. The automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet according to claim 4, wherein, The local features of dendritic microstructure include edges, textures, and shapes.

6. The automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet according to claim 1, characterized in that The DendriticNet model uses the cross-entropy function as the loss function.

7. The automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet according to claim 1, characterized in that, In the model training process in step S4, the input image size is set to 512×512, the momentum is set to 0.9, the cosine annealing descent method is used for training, the batch size is set to 16, the Adam optimizer is used to dynamically optimize the learning rate during the learning process, the initial learning rate is set to 0.001, the weight decay coefficient is 1e-2, and the total number of training epochs is set to 200.

8. The automatic segmentation method of the dendritic structure image of the laser cladding coating based on DendriticNet according to claim 1, characterized in that It also includes S6: Calculating the proportion, area, area proportion, total dendritic density proportion, and total dendritic area of different dendritic densities in the image according to the pixel point information of the identified different dendritic regions.

9. The automatic segmentation method of the dendritic structure image of the laser cladding coating based on DendriticNet according to claim 1, characterized in that, S6 specifically includes the following steps: S61: Obtaining the pixel length t in the dendritic microstructure image; S62: Calculate the density proportion of different dendrite tissues based on the pixel point information of the identified different dendrite regions. The specific calculation method is as follows: Among them, D Dendriticcrystal represents the density of dendritic crystals, ∑pixel(Dendritic crystal) represents the number of pixels occupied by dendritic crystals, and Total pixel represents the total number of all pixels in the image; D Bulkcrystal represents the density of bulk crystals, ∑pixel(Bulk block) represents the number of pixels occupied by bulk crystals; D Equiaxcrystal represents the density of equiaxed crystals, ∑pixel(Equiax crystal) represents the number of pixels occupied by equiaxed crystals; The total dendrite density proportion is obtained by adding the densities of dendritic crystals, massive crystals, and equiaxed crystals; S63: Calculate the areas, area proportions, and total dendrite area of different dendrites based on the pixel point information and pixel length t of the identified different dendrite regions; Among them, the dendrite area A Dendriticcrystal is calculated as follows: A Dendriticcrystal = t 2 × ∑ pixel(Dendritic crystal) The area A of the blocky crystal Bulkcrystal is calculated as follows: A Bulkcrystal = t 2 × ∑ pixel(Bulk crystal) Equiaxed crystal area A Equiaxcrystal The calculation method is as follows: A Equiaxcrystal = t 2 × ∑ pixel (Equiax crystal) Area ratio AR of dendrite Dendriticcrystal The calculation method is as follows: AR Dendriticcrystal = A Dendriticcrystal / (A Dendriticcrystal + A Bulkcrystal + A Equiaxcrystal ) Area ratio AR of blocky crystals Bulkcrystal The calculation method is as follows: AR Bulkcrystal = A Bulkcrystal / (A Dendriticcrystal + A Bulkcrystal + A Equiaxcrystal ) Area ratio AR of equiaxed crystals Equiaxcrystal The calculation method is as follows: AR Equiaxcrystal = A Equiaxcrystal / (A Dendriticcrystal + A Bulkcrystal + A Equiaxcrystal ) The calculation method of the total dendrite area is as follows: A = A Dendriticcrystal +A Bulkcrystal +A Equiaxcrystal 。 10. The automatic segmentation method for dendritic microstructure images of laser cladding coatings based on DendriticNet according to claim 1, wherein, The method for obtaining the pixel length t in the dendrite tissue image in step S61 is as follows: Extract the scale on the dendrite tissue image and obtain the pixel length of the scale; Calculate the pixel length t based on the marked length L of the scale and the corresponding pixel length S of the scale. The expression of t is: t = L / S.

Images