A method for realizing intelligent temperature measurement of non-ferrous metal plates based on visible light images

Abstract

The application discloses a kind of non-ferrous metal plate based on visible light image realizes intelligent temperature measurement method, it is related to metal intelligent temperature measurement field.The specific steps include:1) under the condition of different temperature and illumination intensity, the image of non-ferrous metal plate surface is collected;2) the image is preprocessed by image format conversion, cropping and filtering denoising etc.;3) the high-level color feature of image is extracted, and the feature is screened by improved Fisher discriminant;4) the extracted image feature and corresponding temperature and illumination intensity label data are learned using K nearest neighbor regression algorithm, and non-ferrous metal plate temperature detection model is constructed.The application is suitable for the measurement of non-ferrous metal plate surface temperature under different illumination intensity, the measurement area is large, and it is easy to install, simple to operate.The problems of difficult installation and maintenance and small measurement area of traditional contact type temperature measurement are solved, and the problem that infrared temperature measurement is easily affected by environment and self emissivity change, resulting in precision decline.

Description

Technical Field

[0001] This invention belongs to the field of intelligent temperature measurement of non-ferrous metals, specifically relating to a method for intelligent temperature measurement of non-ferrous metal plates based on visible light images. Background Technology

[0002] In recent years, the demand for lightweight non-ferrous metal materials such as aluminum alloys, magnesium alloys, and copper alloys has been increasing, with widespread applications in aerospace, transportation, and 3C digital industries. However, non-ferrous metals are highly sensitive to temperature, especially during the rolling process, where rolling temperature significantly impacts the microstructure and properties of the rolled non-ferrous metal sheets. Excessively high temperatures lead to excessive plasticity in the non-ferrous metal sheets, making deformation difficult to control during rolling and prone to cracking and uneven deformation. Conversely, excessively low temperatures reduce the plasticity of the non-ferrous metal sheets, making deformation more difficult and hindering grain slippage during deformation, resulting in internal stress concentration and affecting the product's strength and toughness. Therefore, temperature control and monitoring during the hot rolling process of non-ferrous metal sheets are crucial, and this has become a significant factor limiting their development.

[0003] Currently, the main temperature measurement methods for non-ferrous metal sheets during hot rolling production are contact temperature measurement and infrared temperature measurement. Contact temperature measurement can only measure the temperature at a single point, which disrupts the continuity of the material's surface temperature curve. Furthermore, it is often performed manually in actual production, which can easily lead to safety hazards. Infrared temperature measurement and other thermal imaging methods do not require contact with the sheet material, but the emissivity of the metal surface varies with changes in surface condition, resulting in low reliability. In addition, the high surface reflectivity of non-ferrous metals can cause interference from other heat sources in the temperature measurement results. This invention combines visible light imaging and intelligent algorithms for temperature detection of non-ferrous metal sheets. It acquires surface images of non-ferrous metal sheets under different light intensities and temperatures, extracts image features in the RGB color space, and uses an improved Fisher criterion to filter features. It then constructs an intelligent temperature measurement model based on the K-nearest neighbor regression algorithm, effectively overcoming the problems of discontinuous temperature curves, difficult installation and maintenance, and small measurement area in traditional contact temperature measurement. Furthermore, it addresses the issue that infrared temperature measurement is susceptible to the influence of environmental dust, water vapor, heat sources, and changes in its own emissivity, leading to decreased accuracy. This invention enables accurate detection of high-temperature ranges in non-ferrous metal sheets, which is of great significance for improving the processing and forming quality of non-ferrous metal sheets. Summary of the Invention

[0004] To address the difficulty of temperature detection during the rolling process of non-ferrous metal sheets, this invention provides a method for intelligent temperature measurement of non-ferrous metal sheets based on visible light images, comprising the following steps: Step 1: Acquire visible light images under known temperature range and light intensity conditions. First, construct a closed, opaque dark environment around the tubular furnace using professional physical light-blocking cloth, and use LED fluorescent lights to supplement the light and control the light intensity. Then, place the non-ferrous metal sheet in a fixed position in the center of the tubular furnace, and adjust the angle of the sheet to ensure uniform temperature across the entire sheet. Finally, set different light intensities corresponding to different heating temperature ranges, and use a digital camera to acquire RAW format images of the non-ferrous metal sheet surface at fixed temperature intervals. Step 2: Preprocess the images obtained in Step 1. First, label the captured visible light images, including temperature and light intensity labels. Then, convert the image format, transforming the uneditable RAW format visible light images into lossless compressed and editable TIFF format images. Cropped TIFF images, keeping the same positions, are then cropped to the same size. Finally, perform median filtering noise reduction on the cropped TIFF images. Step 3: First, calculate the grayscale frequency of each color channel (R, G, B) in the RGB color space of all preprocessed images: In the formula: for Component images have gray levels The number of pixels, Represents grayscale levels, and , Let be the size of the image, where the image length is . pixels, width is Pixel; Then, the mean, median, mode, range, variance, standard deviation, kurtosis, and skewness of the grayscale frequencies of each color channel (R, G, B) in the RGB color space, as well as the image's energy and entropy, are extracted, totaling 30 feature values. The median represents the middle value of the grayscale frequency; the mode represents the value with the most frequent grayscale frequency; the remaining feature values ​​are calculated according to the following formula: Mean: ; Range: ; variance: ; Standard deviation: ; Kuroshi: ; Skewness: ; energy: ; entropy: , In the formula: For image grayscale, The maximum gray level, The minimum gray level, For grayscale frequency, The length of the image in pixels. The width of the image in pixels; Step 4: Use the improved Fisher criterion to evaluate the 30 extracted color features and select the criterion value. Greater than the discrimination threshold The features are analyzed, and the remaining feature data is removed, resulting in the selected image. These features form the final key dataset. The improved Fisher criterion adds a uniform interval coefficient. This can reflect the uniformity of the distribution of samples at different temperatures; Improved Fisher criterion : In the formula: For the first Within-class variance of each feature For the first The inter-class variance of each feature; ; , In the formula: Assign a feature number; Temperature rating , For temperature ratings; Temperature rating The first of a single sample One feature; Temperature rating is The sample set; Temperature rating The first of a single sample The average of the features; For the sample number The average value of each feature; Temperature rating The number of samples; The total number of samples; To reflect the uniformity of sample distribution at different temperatures, the Fisher criterion incorporates a uniformity interval coefficient. : , Defined as the ratio of the median of the intervals between the mean values ​​of characteristic quantities of samples at different temperatures to the ideal uniform interval.

[0005] Step 5: Constructing the K-nearest neighbor algorithm model includes the following steps: (1) Data preprocessing: The data of the key dataset is normalized by deviation standardization, and the feature parameters of the key dataset are used as the input of the K-nearest neighbor regression algorithm model, and the temperature of the surface of the non-ferrous metal plate is used as the output. The deviation standardization formula is as follows: In the formula: The maximum value of the sample data. The minimum value of the sample data; (2) Selecting the K value: The optimal K value of the temperature measurement model of non-ferrous metal plates based on the K-nearest neighbor algorithm is obtained by cross-validation. The training dataset is divided into several parts, and one part is used as the validation set in turn, while the rest is used as the training set for model training. (3) Calculate the distance: Calculate the distance between each test image sample and each sample in the training set using Euclidean distance. The calculation formula is as follows: In the formula: Represents image samples eigenvectors With image samples eigenvectors The distance between them; (4) Determine the neighbors: Determine the K nearest training samples based on the distance between the test image samples and each sample in the training set; (5) Output calculated value: The average temperature of the K nearest training samples is used as the output temperature; (6) Model evaluation: The performance of the metal sheet temperature measurement model is evaluated using the mean absolute error of the test set.

[0006] Step 6: Input the feature values ​​selected from the visible light image of the surface of the metal sheet to be detected, and calculate the average temperature based on the K-nearest neighbor regression algorithm.

[0007] Compared with existing methods for measuring the temperature of non-ferrous metal sheets, the above technical solution can achieve the following beneficial effects: The present invention provides a method for intelligent temperature measurement of non-ferrous metal plates based on visible light images. This method overcomes the problem of reduced accuracy caused by the influence of environmental factors such as dust, water vapor, heat sources, and changes in reflectivity on infrared thermometry. It also avoids the difficulties in installation and maintenance, small measurement area, and errors caused by manual operation inherent in traditional contact temperature measurement methods. Furthermore, by applying machine learning algorithms, an intelligent temperature measurement model for non-ferrous metal plates based on the K-nearest neighbor regression algorithm is constructed. This method exhibits strong environmental adaptability, simple installation and maintenance, a large measurement area, and high temperature measurement accuracy. Attached Figure Description

[0008] Figure 1 This is a flowchart of a method for intelligent temperature measurement of non-ferrous metal sheets based on visible light images. Detailed Implementation

[0009] This invention provides a method for intelligent temperature measurement of non-ferrous metal sheets based on visible light images.

[0010] The process of capturing the image in this example is as follows: Taking a cast-rolled AZ31 magnesium alloy sheet and a copper alloy sheet with dimensions of 70mm x 100mm as an example, three light intensities of 90 LUX, 120 LUX, and 140 LUX were photographed in a dark environment with professional physical light-blocking cloth forming a closed and opaque environment. The temperature range was set from 250℃ to 450℃, with 5℃ as the temperature interval. Twenty raw format images of the sheet surface were taken at each temperature, resulting in 2460 images for both the magnesium alloy sheet and the copper alloy sheet.

[0011] The preprocessing process in this example is as follows: First, convert the RAW image to an editable TIFF format. Then, crop a 400×700 pixel block from the same location within the TIFF image. Finally, apply median filtering to the cropped image for noise reduction.

[0012] The grayscale frequencies of the three primary colors for each image of the magnesium alloy and copper alloy after preprocessing were calculated. The grayscale range is [0, 255], with a total of 256 grayscale levels.

[0013]

[0014]

[0015] Extract the mean, range, variance, standard deviation, kurtosis, and skewness of the grayscale frequencies of the R, G, and B channels of magnesium alloy and copper alloy images, as well as the energy and entropy of the images.

[0016]

[0017]

[0018] The improved Fisher discriminant formula was used to select 30 parameters extracted from magnesium and copper alloys, and the criterion value for each feature was output. .

[0019]

[0020]

[0021] Set threshold =0.1, and finally select features with a criterion value greater than the threshold of 0.1. The features selected for magnesium alloy are: R channel kurtosis, B channel kurtosis, R channel range and B channel skewness; the features selected for copper alloy are: R channel entropy, R channel range, G channel kurtosis and B channel entropy.

[0022] Construct a K-nearest neighbor regression algorithm.

[0023] First, the data in the key dataset is normalized using deviation standardization. Then, cross-validation is used to obtain the K value, and the training dataset is divided into 10 parts. One part is used as the validation set in turn, and the rest are used as the training set for model training.

[0024]

[0025]

[0026] For magnesium alloys, the minimum selection error is K=4. For copper alloys, the minimum selection error is K=3.

[0027] The distance between the test sample and the remaining samples is calculated using Euclidean distance. For magnesium alloy, the four closest samples are selected, and for copper alloy, the three closest samples are selected. The average value is then used as the regression prediction value for the test sample.

Claims

1. A method for intelligent temperature measurement of non-ferrous metal sheets based on visible light images, characterized in that, Includes the following steps: Step 1: In a professional physical shaded environment, take RAW format images of the surface of non-ferrous metal sheets at a temperature range of 250~450℃ and under different sunlight conditions to establish a visible light image library; Step 2: Label the captured visible light images, including temperature and light intensity labels, and then convert the RAW format images to an editable TIFF format; keep the same position in the image and crop the image to the same size; finally, use median filtering to reduce noise in the cropped image and build a temperature measurement image library for non-ferrous metal plates. Step 3: In the RGB color space, calculate the mean, median, mode, range, variance, standard deviation, kurtosis, and skewness of the grayscale frequencies of the three color channels (R, G, and B) of all images in the non-ferrous metal sheet temperature measurement image library, as well as the data of 30 features such as the energy and entropy of the images. Step 4: Use the improved Fisher criterion to discriminate the 30 extracted color features and select the criterion value. Based on the characteristics, remove the remaining feature data, and summarize the selected images. Data with specific characteristics forms a key dataset; Step 5: Construct a K-nearest neighbor regression algorithm model for intelligent temperature measurement of non-ferrous metal plates based on visible light images; Step 6: Based on the model established in Step 5, input the feature data selected from the visible light image of the surface of the non-ferrous metal sheet to be detected, and calculate the average temperature.

2. The method for intelligent temperature measurement of non-ferrous metal plates based on visible light images according to claim 1, characterized in that, The process of establishing the visible light image library in step 1 is as follows: First, a closed, opaque dark environment is constructed around the constant-temperature, visible heating furnace using professional physical light-blocking cloth, and LED fluorescent lamps are used for supplemental lighting and the light intensity is controlled; then, the non-ferrous metal sheet is placed in a fixed position in the middle of the heating furnace to ensure that the temperature of the entire sheet is uniform; finally, different light intensities and heating temperature ranges are set, and a digital camera is used to collect RAW format images of the surface of the non-ferrous metal sheet at fixed temperature intervals.

3. The method for intelligent temperature measurement of non-ferrous metal plates based on visible light images according to claim 1, characterized in that, The mean, range, variance, standard deviation, kurtosis, skewness, and image energy and entropy of the 30 features in step 3 are calculated according to the following formula: Mean: ; Range: ; variance: ; Standard deviation: ; Kuroshi: ; Skewness: ; energy: ; entropy: , In the formula: For image grayscale, The maximum gray level, The minimum gray level, For grayscale frequency, The pixel length of the image. The pixel width of the image.

4. The method for intelligent temperature measurement of non-ferrous metal plates based on visible light images according to claim 1, characterized in that, The improved Fisher criterion calculation formula in step 4 is as follows: ； ； ； ， In the formula: This is the Fisher criterion value. For the first The within-class variance of each feature, For the first The inter-class variance of each feature The uniform interval coefficient, For feature numbering, Temperature rating , For temperature levels, Temperature rating The first of a single sample One characteristic, Temperature rating is The sample set, Temperature rating is The first of a single sample The average of the features, For the sample number The average of the features; Temperature rating The number of samples, The total number of samples.

5. The method for intelligent temperature measurement of non-ferrous metal plates based on visible light images according to claim 1, characterized in that, The process of establishing the K-nearest neighbor regression algorithm model in step 5 is as follows: (1) Data preprocessing: The data of the key dataset is normalized by using deviation standardization, and the feature parameters of the key dataset are used as the input of the K-nearest neighbor regression algorithm model, and the temperature of the surface of the non-ferrous metal plate is used as the output. (2) Selecting the K value: The optimal K value of the temperature measurement model for non-ferrous metal plates based on the K-nearest neighbor regression algorithm was obtained by cross-validation; (3) Calculate the distance: Calculate the distance between each test image sample and each sample in the training set using Euclidean distance. The calculation formula is as follows: ， In the formula: Represents image samples eigenvectors With image samples eigenvectors The distance between them; (4) Determine the neighbors: Determine the K training samples with the shortest distances based on the distances between the test image samples and each sample in the training set; (5) Fitting the output: The average temperature of the K nearest training samples is used as the output temperature; (6) Model evaluation: The reliability of the temperature measurement model for non-ferrous metal plates was evaluated using the mean absolute error of the test set.

Images