CN116311237A - A steel microstructure identification algorithm based on deep learning network - Google Patents

Abstract

The invention relates to a steel microstructure identification algorithm based on a deep learning network, and belongs to the technical field of image identification methods. The technical scheme of the invention is as follows: carrying out grinding, polishing and corrosion operation on a sample by using a standard metallographic sample preparation technology, selecting proper magnification under a microscope, and acquiring images, thereby establishing image datasets of different microstructures; and (3) establishing a deep learning network training set by adopting an image size normalization processing method, and learning tissue characteristics by utilizing a deep learning ResNet50 network to realize steel microstructure identification. The beneficial effects of the invention are as follows: the error rate of manually distinguishing metallographic structure image categories is reduced by utilizing a deep learning algorithm and combining with image processing, the distinguishing efficiency is improved, and the personnel cost and the time cost are effectively saved.

Description

A steel microstructure identification algorithm based on deep learning network

technical field

The invention relates to a steel microstructure recognition algorithm based on a deep learning network, and belongs to the technical field of image recognition methods.

Background technique

Iron and steel microstructure analysis technology plays an important role in the study of metal material composition, heat treatment process, and organizational structure. At present, microstructure identification mainly adopts the methods of optical microscope observation record and manual analysis. Organizational analysis requires high technical ability of personnel, and ordinary inspectors have low awareness of difficult organizations, which requires years of experience accumulation and training. Different inspectors have different evaluation standards for similar samples, which are greatly influenced by subjective consciousness. The evaluation of most inspection items mainly depends on the personnel's years of inspection experience, and most links require the participation and checks of experts.

Contents of the invention

The purpose of the present invention is to provide a steel microstructure recognition algorithm based on a deep learning network, using the deep learning algorithm, combined with image processing, to reduce the error rate of manual resolution of metallographic structure image categories, improve resolution efficiency, and effectively save personnel costs and time Cost effectively solves the above-mentioned problems existing in the background technology.

The technical solution of the present invention is: a steel microstructure recognition algorithm based on a deep learning network, comprising the following steps:

S1. Sample material selection: select medium-carbon, high-carbon and low-carbon steel materials under different steel types and different heat treatment processes, and cut and select materials according to the requirements of metallographic samples;

S2. Sample preparation: grind, polish and corrode the sample according to the standard metallographic sample preparation requirements;

S3. Image collection: using a microscope to collect images of various tissues;

S4. Image preprocessing: normalize and segment the collected images, perform data enhancement by means of rotation, translation and cropping to form a training set;

S5. In-depth model training: build a Resnet50 model based on the Tensorflow framework, and use the training set to train the model;

S6. Achievement test: use the test set to test the generalization ability of the Resnet50 model, that is, to test the recognition accuracy of the model for microscopic tissue images other than the training samples.

In the step S3, use a Zeiss-level metallographic microscope to collect images of each tissue, adjust the brightness of the microscope focal length, and clearly display the details of the tissue, select a suitable field of view, and collect ferrite, pearlite, martensite, and bainite and austenite structure.

In the step S4, normalize and segment the collected image, divide the image into four equal parts after cropping, add corresponding gray scale bars on the top and bottom of the image, zoom in, and perform data enhancement to form a training set.

In the step S5, a Resnet50 network is built based on the TensorFlow framework, that is, residual learning is added to the convolutional neural network.

The beneficial effects of the present invention are: using a deep learning algorithm, combined with image processing, to reduce the error rate of manually distinguishing metallographic structure image categories, improve resolution efficiency, and effectively save personnel costs and time costs.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is the deep learning Resnet50 network model diagram of the present invention;;

Fig. 3 is the test figure of the embodiment of the present invention;

Fig. 4 is a diagram of test data results of the embodiment of the present invention.

Implementation

In order to make the purpose, technical solutions and advantages of the invention implementation cases clearer, the technical solutions in the implementation cases of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the implementation cases. Obviously, the implementation cases described are the present invention Invention of a small part of the implementation cases, not all of the implementation cases, based on the implementation cases in the present invention, all other implementation cases obtained by those of ordinary skill in the art without creative work, all belong to the protection scope of the present invention.

A steel microstructure identification algorithm based on a deep learning network, comprising the following steps:

S1. Sample material selection: select medium-carbon, high-carbon and low-carbon steel materials under different steel types and different heat treatment processes, and cut and select materials according to the requirements of metallographic samples;

S2. Sample preparation: grind, polish and corrode the sample according to the standard metallographic sample preparation requirements;

S3. Image collection: using a microscope to collect images of various tissues;

S4. Image preprocessing: normalize and segment the collected images, perform data enhancement by means of rotation, translation and cropping to form a training set;

S5. In-depth model training: build a Resnet50 model based on the Tensorflow framework, and use the training set to train the model;

S6. Achievement test: use the test set to test the generalization ability of the Resnet50 model, that is, to test the recognition accuracy of the model for microscopic tissue images other than the training samples.

In the step S3, use a Zeiss-level metallographic microscope to collect images of each tissue, adjust the brightness of the microscope focal length, and clearly display tissue details, select a suitable field of view, and a magnification of 500X or 1000X to collect ferrite, pearlite, There are 17 types of martensite, bainite and austenite, with a total of 4000 sheets.

In the step S4, the collected image is normalized and segmented. The original image is 1292×968 pixels in size, and the image is divided into four equal parts after cutting. The size of each image is 324×242. Corresponding grayscale bar, enlarged to 672×672 pixel size. After data enhancement, a training set is formed, and the total number of training sets is 16,000.

In the step S5, the Resnet50 network is built based on the TensorFlow framework, that is, residual learning is added to the convolutional neural network, the network structure has 50 layers, and the input image size is 672×672 pixels.

In practical applications, as shown in Figure 2, in the deep learning Resnet50 network model, CONV represents the convolution layer, Batch Norm represents the Batch normalization layer, ID BLOCK represents the Identity block, which is composed of multiple layers, and Conv BLOCK represents convolution A block, made up of multiple layers. In order to make the structure of the model clearer, two blocks, conv block and idblock, are extracted and encapsulated into functions respectively.

The method provided by the invention can realize the identification of steel microstructure. As for the method disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related part, please refer to the description of the method part.

In this paper, specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; meanwhile, for those of ordinary skill in the art, according to the present invention Thoughts, there will be changes in specific implementation methods and application ranges. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims (4)

1. A steel microstructure identification algorithm based on a deep learning network, characterized in that it comprises the following steps: S1. Sample material selection: select medium-carbon, high-carbon and low-carbon steel materials under different steel types and different heat treatment processes, and cut and select materials according to the requirements of metallographic samples; S2. Sample preparation: grind, polish and corrode the sample according to the standard metallographic sample preparation requirements; S3. Image collection: using a microscope to collect images of various tissues; S4. Image preprocessing: normalize and segment the collected images, perform data enhancement by means of rotation, translation and cropping to form a training set; S5. In-depth model training: build a Resnet50 model based on the Tensorflow framework, and use the training set to train the model; S6. Achievement test: use the test set to test the generalization ability of the Resnet50 model, that is, to test the recognition accuracy of the model for microscopic tissue images other than the training samples. 2. A steel microstructure recognition algorithm based on deep learning network according to claim 1, characterized in that: in the step S3, the images of each tissue are collected with a Zeiss-level metallographic microscope, and the brightness of the focal length of the microscope is adjusted to clearly Display microstructure details, select a suitable field of view, and collect ferrite, pearlite, martensite, bainite and austenite microstructures. 3. A kind of iron and steel microstructure recognition algorithm based on deep learning network according to claim 1, it is characterized in that: in described step S4, carry out normalization and segmentation processing with collected image, divide image into after clipping Four equal parts, add corresponding grayscale bars on the top and bottom of the image, zoom in, and perform data enhancement to form a training set. 4. A kind of iron and steel microstructure identification algorithm based on deep learning network according to claim 1, it is characterized in that: in described step S5, build Resnet50 network based on TensorFlow framework, promptly add residual learning in convolutional neural network .

Images