Water turbine maintenance tool identification method based on ternary network

Abstract

A water turbine maintenance tool identification method based on a ternary network comprises the steps that a large number of maintenance tool images are collected and subjected to data enhancement processing, and a complete maintenance tool data set is formed; the data set is divided into a training set, a verification set and a test set according to the proportion, and a foundation is laid for constructing a water turbine maintenance tool recognition model; and constructing a ternary network structure based on the ResNeXt50 backbone network, and inputting the training set into a ternary network model of the ResNeXt50 backbone network for training so as to establish an efficient and accurate identification model. In the training process, the verification set is used for evaluating model performance, and an optimal model is determined to generate a water turbine maintenance tool recognition model based on the ternary network; according to the method, an independent test set is utilized to comprehensively test the generated ternary network-based water turbine maintenance tool identification model so as to verify the performance and accuracy of the model in practical application, an efficient and accurate solution is provided, and the method has a wide application prospect.

Description

Technical Field

[0001] The invention belongs to the technical field of computer vision and relates to a method for identifying a turbine maintenance tool based on a ternary network. Background Art

[0002] In the daily maintenance work of turbine units, accurate classification and identification of maintenance tools are crucial. Traditional classification methods usually rely on manual operation, which has the problems of large subjective factors, high misjudgment rate, time-consuming and low efficiency. How to achieve efficient, accurate and automated methods to identify turbine maintenance tools to improve work efficiency and accuracy is the key issue currently faced. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method for identifying turbine maintenance tools based on a ternary network. The method based on deep learning realizes accurate classification and identification of different categories of tools through training and learning of a large number of maintenance tool images, avoids interference from human subjective factors, and improves recognition accuracy and efficiency.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for identifying a hydraulic turbine maintenance tool based on a ternary network, characterized in that it comprises the following steps: S1, collect maintenance tool images and perform data enhancement processing to form a maintenance tool data set; S2, divide the maintenance tool dataset into training set, validation set and test set according to the proportional relationship; S3, constructing a turbine maintenance tool identification model based on a ternary network structure; S4, training a turbine maintenance tool recognition model based on a ternary network structure; S5, generating a turbine maintenance tool identification model based on a ternary network structure.

[0005] In S1, data enhancement processing includes horizontal flipping, vertical flipping, multi-angle rotation, scaling, random cropping and other processing.

[0006] In S2, the maintenance tool dataset in S1 is divided into training set, validation set and test set according to the ratio of 6:2:2.

[0007] In S3, three network models with ResNeXt50 as the backbone network are constructed. The three network branches are called anchor sample branch, positive sample branch and negative sample branch respectively.

[0008] In S4, it specifically includes: S4-1, train the ternary network model by inputting the training set into the ResNeXt50 backbone network to establish an efficient and accurate recognition model; S4-2, in the training set of the turbine maintenance tool recognition model based on the ternary network structure, select an image of a maintenance tool as an anchor sample, then select an image of a maintenance tool of the same type as the anchor sample but taken at a different angle as a positive sample, and select an image of a maintenance tool of a different type from the anchor sample as a negative sample; S4-3, use the ResNeXt50 model to extract the image features of the three samples, calculate the Euclidean distance between the anchor sample and the positive sample and the Euclidean distance between the anchor sample and the negative sample, and identify the maintenance tool category through the softmax algorithm.

[0009] In S5, the trained network weights are saved, and a turbine maintenance tool recognition model based on a ternary network structure is generated for recognizing turbine maintenance tools.

[0010] In S4, a maintenance tool image is selected as an anchor sample in the training set of the turbine maintenance tool recognition model based on the ternary network structure. , and then select an image of a maintenance tool of the same type as the anchor sample but taken at a different angle as the positive sample and an image of a maintenance tool of a different class from the anchor sample as a negative sample , the anchor sample, positive sample and negative sample are combined into a triplet ( ), and normalize each image to generate a color image with a width and height of 224, that is, a triplet of image size 3×224×224 ( ). The ResNeXt50 model is used to extract the image features of the three samples and obtain three feature maps ( ), in, , , , C represents the number of channels, and the preferred value in this example is 2048, H represents the height of the feature map, and W represents the width of the feature map.

[0011] In S4, the Euclidean distance between the anchor sample and the positive sample is calculated and the Euclidean distance between the anchor sample and the negative sample , use formula (1) (2) to calculate: ; (1) ; (2) in, , and The feature maps representing the anchor sample, positive sample, and negative sample respectively have the pixel value at the kth channel, row i, and column j.

[0012] In S4, the triplet loss function is used to iterate and train the optimal model. (3) Among them, the parameters express and the degree of spacing between The bigger, the easier it is to distinguish and , in this example, ; Finally, the maintenance tool category is identified through the softmax function.

[0013] The main beneficial effects of the present invention are: The deep learning-based method can accurately classify and identify different types of tools by training and learning a large number of maintenance tool images, avoiding interference from human subjective factors and improving recognition accuracy and efficiency.

[0014] The introduction of the ResNeXt50 backbone network structure based on the ternary network helps to better extract and learn the features of the maintenance tool images, thereby realizing the accurate recognition of the turbine maintenance tools, thus bringing unprecedented convenience and efficiency improvement to the maintenance work of the turbine units. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 This is a maintenance tool identification model diagram of the present invention.

[0018] Figure 3 This is the ResNeXt50 network model diagram of the present invention. DETAILED DESCRIPTION

[0019] like Figure 1~Figure 3 In the present invention, a method for identifying a hydraulic turbine maintenance tool based on a ternary network comprises the following steps: S1, collect maintenance tool images and perform data enhancement processing to form a maintenance tool data set; Embodiment 1, The hydroturbine maintenance tools, including hammers, screwdrivers, hexagon sockets, sockets, wrenches and other tools of different types and specifications, were selected as research objects, and images of each tool were taken from multiple perspectives using industrial cameras. The collected images were expanded using data enhancement methods, including horizontal flipping, vertical flipping, multi-angle rotation, scaling, random cropping and other processing, to form a rich and diverse data set.

[0020] S2, divide the maintenance tool dataset into training set, validation set and test set according to the proportional relationship; Preferably, the maintenance tool data set in S1 is divided into a training set, a validation set and a test set in a ratio of 6:2:2.

[0021] S3, constructing a turbine maintenance tool identification model based on a ternary network structure; Embodiment 2, like Figure 2 As shown in the figure, three network models with ResNeXt50 as the backbone networks are constructed, and the three network branches are respectively called anchor sample branch, positive sample branch and negative sample branch.

[0022] The structure diagram of ResNeXt50 is as follows Figure 3 As shown in the figure, there are five stages from the input to the output. The first stage consists of a convolution layer and a maximum pooling layer. The second stage consists of three BLOCK modules. In the BLOCK module, the input features with C channels are first divided into 32 groups, each group is compressed to 4 channels and then processed. After the 32 groups are added, they are connected with the original feature residuals and then output with C1 channels. The third and fourth stages are also composed of BLOCK modules. The number of BLOCK modules, input image size and output image size are different in each stage.

[0023] S4, training a turbine maintenance tool recognition model based on a ternary network structure; Embodiment 3, The training set is input into the ternary network model of the ResNeXt50 backbone network for training to establish an efficient and accurate recognition model.

[0024] In the training set of the turbine maintenance tool recognition model based on the ternary network structure, a maintenance tool image is selected as the anchor sample. , and then select an image of a maintenance tool of the same type as the anchor sample but taken at a different angle as the positive sample and an image of a maintenance tool of a different class from the anchor sample as a negative sample , the anchor sample, positive sample and negative sample are combined into a triplet ( ), and normalize each image to generate a color image with a width and height of 224, that is, a triplet of image size 3×224×224 ( ).

[0025] The ResNeXt50 model is used to extract the image features of the three samples and obtain three feature maps ( ), in, , , , C represents the number of channels, and the preferred value in this example is 2048, H represents the height of the feature map, and the value in this example is 7, and W represents the width of the feature map, and the value in this example is 7.

[0026] Calculate the Euclidean distance between the anchor sample and the positive sample and the Euclidean distance between the anchor sample and the negative sample , use formula (1) (2) to calculate: (1) (2) in, , and The feature maps representing the anchor sample, positive sample, and negative sample respectively have the pixel value at the kth channel, row i, and column j.

[0027] Use the triplet loss function to iterate and train the optimal model (3) Among them, the parameters express and the degree of spacing between The bigger, the easier it is to distinguish and , in this example, .

[0028] Finally, the maintenance tool category is identified through the softmax function.

[0029] S5, generating a turbine maintenance tool identification model based on a ternary network structure.

[0030] The trained network weights are saved to generate a turbine maintenance tool recognition model based on a ternary network structure for identifying turbine maintenance tools.

[0031] Among the above methods, the deep learning-based method can achieve accurate classification and identification of different categories of tools by training and learning a large number of maintenance tool images, avoiding the interference of human subjective factors and improving recognition accuracy and efficiency.

[0032] The introduction of the ResNeXt50 backbone network structure based on the ternary network helps to better extract and learn the features of the maintenance tool images, thereby realizing the accurate recognition of the turbine maintenance tools, bringing unprecedented convenience and efficiency improvement to the maintenance work of the turbine units; it also provides an efficient and accurate solution for the identification of turbine maintenance tools, which has broad application prospects.

[0033] The above-mentioned embodiments are only preferred technical solutions of the present invention and should not be regarded as limitations of the present invention. The embodiments and features in the embodiments of the present application can be arbitrarily combined with each other without conflict. The protection scope of the present invention shall be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for identifying a hydraulic turbine maintenance tool based on a ternary network, characterized in that: The steps include: S1, collect maintenance tool images and perform data enhancement processing to form a maintenance tool data set; S2, divide the maintenance tool dataset into training set, validation set and test set according to the proportional relationship; S3, constructing a turbine maintenance tool identification model based on a ternary network structure; S4, training a turbine maintenance tool recognition model based on a ternary network structure; S5, generating a turbine maintenance tool identification model based on a ternary network structure.

2. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 1 is characterized in that: In S1, data enhancement processing includes horizontal flipping, vertical flipping, multi-angle rotation, scaling, random cropping and other processing.

3. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 1 is characterized in that: In S2, the maintenance tool dataset in S1 is divided into training set, validation set and test set according to the ratio of 6:2:

2.

4. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 1 is characterized in that: In S3, three network models with ResNeXt50 as the backbone network are constructed. The three network branches are called anchor sample branch, positive sample branch and negative sample branch respectively.

5. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 1 is characterized in that: In S4, it specifically includes: S4-1, train the ternary network model by inputting the training set into the ResNeXt50 backbone network to establish an efficient and accurate recognition model; S4-2, in the training set of the turbine maintenance tool recognition model based on the ternary network structure, select an image of a maintenance tool as an anchor sample, then select an image of a maintenance tool of the same type as the anchor sample but taken at a different angle as a positive sample, and select an image of a maintenance tool of a different type from the anchor sample as a negative sample; S4-3, use the ResNeXt50 model to extract the image features of the three samples, calculate the Euclidean distance between the anchor sample and the positive sample and the Euclidean distance between the anchor sample and the negative sample, and identify the maintenance tool category through the softmax algorithm.

6. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 1 is characterized by: In S5, the trained network weights are saved, and a turbine maintenance tool recognition model based on a ternary network structure is generated for recognizing turbine maintenance tools.

7. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 5 is characterized by: In S4, a maintenance tool image is selected as an anchor sample in the training set of the turbine maintenance tool recognition model based on the ternary network structure. , and then select an image of a maintenance tool of the same type as the anchor sample but taken at a different angle as the positive sample and an image of a maintenance tool of a different class from the anchor sample as a negative sample , the anchor sample, positive sample and negative sample are combined into a triplet ( ), and normalize each image to generate a color image with a width and height of 224, that is, a triplet of image size 3×224×224 ( ). The ResNeXt50 model is used to extract the image features of the three samples and obtain three feature maps ( ), in, , , , C represents the number of channels, and the preferred value in this example is 2048, H represents the height of the feature map, and W represents the width of the feature map.

8. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 7 is characterized by: In S4, the Euclidean distance between the anchor sample and the positive sample is calculated and the Euclidean distance between the anchor sample and the negative sample , use formula (1) (2) to calculate: ；（1） ；（2） in, , and The feature maps representing the anchor sample, positive sample, and negative sample respectively have the pixel value at the kth channel, row i, and column j.

9. The method for identifying a hydraulic turbine maintenance tool based on a ternary network according to claim 7 is characterized by: In S4, the triplet loss function is used to iterate and train the optimal model. （3） Among them, the parameters express and the degree of spacing between; The bigger, the easier it is to distinguish and , in this example, ; Finally, the maintenance tool category is identified through the softmax function.

Images